ProgrammableMarketing/kepler.ai (opens in a new tab)
PDF

Views

  • Plan
  • Roadmap
  • Research

On this page

  • The brief
  • Start here
  • Idea portfolio
  • Effort vs impact
  • This week · 6
  • This quarter · 8
  • Big bets · 3
  • Wildcards · 4
  • How we can help
Generated Oct 18, 2026
  1. Programmable
  2. Playbooks
  3. Kepler
Kepler logo

growth playbook · Oct 18, 2026

kepler.ai (opens in a new tab)

Kepler growth playbook

Diagnosis

Kepler has a sharp technical claim (AI where the model never touches the numbers), a strong team from Palantir and Citadel, an Anthropic case study and a verified Claude connector.

Its hiring says it is moving from founder-led selling to a real go-to-market: a first Founding Marketer who owns brand, proof, the CEO's voice, partners and AI search; Forward Deployed Engineers who win lighthouse customers through design partners; and a talent lead for a team expected to grow several times over.

The gaps are proof and paths in: no customer is named anywhere, there is no pricing, the connector ends in a demo form, and the name is shared with a busy European broker.

Growth Score

6/10

Biggest Opportunity

Make the claim testable everywhere: a test pack analysts run against their current tool, a provenance mark on every forwarded output, a trial from the Claude connector, and public company pages with traced numbers.

Biggest Risk

Larger finance AI tools (Rogo, AlphaSense, Hebbia) already claim citations and have thousands of customers; without named proof and a faster way to try it, "verifiable" sounds like everyone else's message.

01

Start here

The three moves to make first. Each one is written up in full below.

  1. 1

    One case study with a measured result from a live engagement

    Ask the Forward Deployed Engineer which engagement has a before-and-after number the customer would sign off.

    High impact First result in 3-4 weeks
  2. 2

    A 14-day public-filings trial for people who find Kepler in Claude's connector directory

    Decide the trial limits (public data only, 14 days, a daily query cap) and which email domains qualify.

    High impact First result in 3 weeks
  3. 3

    Turn the compare page's worked examples into a 25-question "check your current tool" pack

    Pick 25 questions from recent 10-Ks and 10-Qs where a typical tool gets the number wrong, and record Kepler's traced answer for each.

    High impact First result in 2 weeks

02

Your idea portfolio

21 ideas, grouped by when to run them. Open any idea for how to run it: the steps, who to approach, copy you can use, and how to measure it.

  • 8 ideas are marked Unexpected: plays you are unlikely to have tried.

Every idea placed by the effort it takes against the impact it could have. Quick wins sit top left; select a number to open that idea. The filters below narrow the list by status, timing, category and what Programmable can build.

Effort vs impactEach number is an idea. Select one to open it.
Impact →
Quick wins · 4Big bets · 9Fill-ins · 6Reconsider · 2
Effort →

Showing 21 of 21 ideas

This week 6

Quick wins with a clear first action.

  • 01

    The compare page proves its point with three worked examples (Alphabet margin, total debt, CoreWeave concentration). Analysts believe what they test themselves, and the page gives them nothing to run.

    High impactLow effort First result in 2 weeks Founding Marketer

    First action

    Pick 25 questions from recent 10-Ks and 10-Qs where a typical tool gets the number wrong, and record Kepler's traced answer for each.

    How to measure it

    Demo requests from the test pack

    Now
    –
    Target
    30
    By
    6 weeks

    Steps

    1. 1

      Pick the questions

      Margins, debt, segment revenue and customer concentration, from filings in the last two quarters.

      Founding Marketer 1 day

    2. 2

      Trace every answer

      Filing, page, line and arithmetic for each answer, from Kepler output.

      Forward Deployed Engineer 1 day

    3. 3

      Build the sheet

      One row per question: question, answer, source link, how it is computed.

      Founding Marketer 3 hours

    4. 4

      Add it to /compare

      A "Run these in your tool" section above the comparison table.

      Founding Marketer 2 hours

    5. 5

      Seed it

      The CEO posts five of the questions on LinkedIn with the answer key link.

      CEO 30 minutes

    6. 6

      Refresh quarterly

      Swap in new filings each earnings season so the answers stay current.

      Founding Marketer 1 day a quarter

    First result
    2 weeks
    Cost
    $0
    Owner
    Founding Marketer

    Smallest test

    Post five questions and the answer key on LinkedIn before building the page.

    Why it works

    Publish 25 questions about public filings, each with the traced answer, the filing and the arithmetic, as a downloadable sheet. Invite analysts to run them in whatever AI tool they use and compare. Gate nothing; the "Request Demo" sits next to the answers.

    The Founding Marketer is asked to know "what competitors claim and where we actually win". A test pack makes the win reproducible without naming competitors, and it travels inside funds as a forwarded spreadsheet.

    Ready to use

    PostCEO launch post

    Ask your AI tool for Alphabet's FY2025 operating margin.
    
    If it says 31.4% or 31.5%, it made the number up. The 10-K says 32.0% (129,039 / 402,836).
    
    We put 25 questions like this in one sheet, each with the filing and the arithmetic. Run them in whatever you use today: {link}

    Risks and how to handle them

    • A competitor answers some questions correctly

      Publish the results honestly; the point is that every Kepler answer is traceable.

    Evidence

    • Your websiteCompare page: Kepler against unnamed "frontier models on their own" and "finance AI tools"; worked examples (Alphabet operating margin 31.4% from a typical tool against 32.0% computed from the 10-K; total debt $10.9B against $35.3B; CoreWeave's Microsoft concentration of about 67% missed); research claims of 99.4% scale accuracy and 94% taxonomy mapping with a model trained for $9.Source
    • HiringFounding Marketer (first marketing hire): owns the brand system ("spread across a brandbook, a website, and a dozen decks that don't quite match"), the website ("you'll ship to it yourself with Claude"), positioning, customer proof from live engagements, the template an analyst sees in a Kepler output, conference materials, the CEO's LinkedIn and X, the blog, recruiting posts, launches and partner co-marketing, and "Search and AI answers".Source
    • Your websiteHomepage: "AI that proves it's right." and "enterprise intelligence for high-stakes industries"; 26M+ SEC filings, 14,000+ companies, 120+ countries, 30+ languages; SOC 2 Type II. No named customers and no pricing; three unnamed testimonials (a Managing Director at a multi-billion dollar fund, a Partner at a global law firm, a VP at a global investment bank). The main CTA is "Request Demo".Source
  • 02

    Analysts forward Kepler outputs, workbooks and slides to the people who make the decision. Those readers are the buyers, and today the output carries no sign of where it came from.

    High impactLow effort First result in 4 weeks Founding Marketer

    First action

    Mock up the footer on the redesigned output template and the Excel add-in export.

    How to measure it

    Demo requests from output footers

    Now
    –
    Target
    10 a month
    By
    8 weeks

    Steps

    1. 1

      Design the mark

      One line, small, matching the new template; no logo wall.

      Founding Marketer 2 hours

    2. 2

      Build the landing page

      What "traced" means, one worked example, Request Demo.

      Founding Marketer 3 hours

    3. 3

      Add to exports

      PDF output, Excel workbooks and PowerPoint slides.

      Full Stack Engineer 2 days

    4. 4

      Give admins a switch

      Off by firm, on by default for trials.

      Full Stack Engineer 2 hours

    5. 5

      Tag the link

      Per customer, so you see which accounts forward outputs.

      Full Stack Engineer 1 hour

    6. 6

      Tell customers

      One note to each account explaining the footer and the switch.

      Forward Deployed Engineer 1 hour

    First result
    4 weeks
    Cost
    $0, about three days of engineering
    Owner
    Founding Marketer

    Smallest test

    Add the footer to trial accounts only for one month.

    Why it works

    As part of the template redesign the job post already asks for, add a quiet footer to exports, Excel workbooks and PowerPoint slides: "Every number traced to source by Kepler", with a link to a short page that explains provenance and offers a demo. Admins can turn it off.

    In finance, work products move up and across firms. A provenance mark makes the product's main claim visible exactly where trust is decided, and it turns each forwarded memo into a referral.

    Risks and how to handle them

    • Customers do not want vendor marks on client work

      Off switch per firm, and default off for client-facing decks if any customer asks.

    Evidence

    • HiringFounding Marketer (first marketing hire): owns the brand system ("spread across a brandbook, a website, and a dozen decks that don't quite match"), the website ("you'll ship to it yourself with Claude"), positioning, customer proof from live engagements, the template an analyst sees in a Kepler output, conference materials, the CEO's LinkedIn and X, the blog, recruiting posts, launches and partner co-marketing, and "Search and AI answers".Source
    • Your websiteProducts include Excel and PowerPoint add-ins, "Build on Kepler" (Kepler MCP and API, docs on Mintlify), workbooks and company pages in the app (an NVIDIA entity page shows 29,329 mentions across 139 filers).Source
    • Your websiteHomepage: "AI that proves it's right." and "enterprise intelligence for high-stakes industries"; 26M+ SEC filings, 14,000+ companies, 120+ countries, 30+ languages; SOC 2 Type II. No named customers and no pricing; three unnamed testimonials (a Managing Director at a multi-billion dollar fund, a Partner at a global law firm, a VP at a global investment bank). The main CTA is "Request Demo".Source
  • 03

    Every testimonial on the site is unnamed and no customer is named anywhere, including the Anthropic case study. For buyers who "sign large enterprise deals", proof from a peer firm is what moves a demo to a pilot.

    High impactLow effort First result in 3-4 weeks Founding Marketer

    First action

    Ask the Forward Deployed Engineer which engagement has a before-and-after number the customer would sign off.

    How to measure it

    Demo to pilot rate

    Now
    –
    Target
    +10 points
    By
    end of quarter

    Steps

    1. 1

      Pick the engagement

      The one with a clear workflow and a number the customer agrees with.

      Forward Deployed Engineer 1 day

    2. 2

      Measure before and after

      Time per task or errors found, over two weeks.

      Forward Deployed Engineer 2 weeks

    3. 3

      Interview the user

      An analyst and their manager, 30 minutes each.

      Founding Marketer 1 hour

    4. 4

      Write it

      One page: problem, workflow, number, quote.

      Founding Marketer 1 day

    5. 5

      Get approval

      Customer legal and compliance sign-off, anonymized if needed.

      CEO 1 week

    6. 6

      Use it everywhere

      Homepage, Finance page, decks and the demo follow-up email.

      Founding Marketer 1 day

    First result
    3 to 4 weeks
    Cost
    $0
    Owner
    Founding Marketer

    Smallest test

    Get one approved quote with a number this week and put it on the Finance page.

    Why it works

    Work with the Forward Deployed Engineer on the strongest live engagement to publish one case study: the firm type and size, the workflow, a before-and-after number (hours per model, comps built per week, errors caught) and a quote with a title. Anonymize the firm if needed, but not the number.

    Customer proof is named in the Founding Marketer's brief. A specific measured outcome is more credible to a skeptical buyer than three unnamed quotes, and it gives every deck and page the same proof point.

    Risks and how to handle them

    • Compliance blocks naming the firm

      Use "a $10B+ AUM hedge fund" and keep the number and title.

    Evidence

    • Your websiteHomepage: "AI that proves it's right." and "enterprise intelligence for high-stakes industries"; 26M+ SEC filings, 14,000+ companies, 120+ countries, 30+ languages; SOC 2 Type II. No named customers and no pricing; three unnamed testimonials (a Managing Director at a multi-billion dollar fund, a Partner at a global law firm, a VP at a global investment bank). The main CTA is "Request Demo".Source
    • HiringFounding Marketer (first marketing hire): owns the brand system ("spread across a brandbook, a website, and a dozen decks that don't quite match"), the website ("you'll ship to it yourself with Claude"), positioning, customer proof from live engagements, the template an analyst sees in a Kepler output, conference materials, the CEO's LinkedIn and X, the blog, recruiting posts, launches and partner co-marketing, and "Search and AI answers".Source
    • HiringForward Deployed Engineer: embed with "lighthouse customers"; "Sit with a Portfolio Manager at an institution with $10B+ AUM"; ship a vertical app for a design partner; integrate with a client's research stack (Bloomberg, CapIQ) with provenance preserved; 30 to 50% travel.Source
    • Search resultsAnthropic case study (April 30, 2026): 26M+ SEC filings, 50M+ public and 1M+ private documents, 14,000+ companies across 27 global markets; the founders interviewed 147 financial firms before building; 94% accuracy mapping statement labels to taxonomy codes against 38 to 46% for other models. No customers are named.Source
  • 04

    A news scan for "Kepler" returned 87 mentions in 30 days, led by Kepler Cheuvreux price-target stories. Buyers in finance are exactly the people who know the broker, and an AI assistant asked about "Kepler" may answer about the wrong company.

    Medium impactLow effort First result in 4-8 weeks Founding Marketer

    First action

    Ask three AI assistants "What is Kepler AI for finance?" and record what they say.

    How to measure it

    Correct answers across 20 test prompts

    Now
    –
    Target
    18 of 20
    By
    8 weeks

    Steps

    1. 1

      Record the baseline

      20 prompts across ChatGPT, Claude, Perplexity and Google.

      Founding Marketer 2 hours

    2. 2

      Add Organization schema

      Name, alternate name, logo, founders, sameAs links.

      Founding Marketer 1 hour

    3. 3

      Align the profiles

      LinkedIn, X, Crunchbase, Wellfound and the connector listing use "Kepler AI".

      Founding Marketer 2 hours

    4. 4

      Create a Wikidata item

      With sources from the Anthropic case study and press.

      Founding Marketer 1 hour

    5. 5

      Rewrite the About page opener

      One sentence that says what Kepler is and for whom.

      Founding Marketer 1 hour

    6. 6

      Recheck monthly

      Same 20 prompts, tracked in a sheet.

      Founding Marketer 1 hour a month

    First result
    4 to 8 weeks
    Cost
    $0
    Owner
    Founding Marketer

    Smallest test

    Add the schema and fix the LinkedIn and X names, then rerun the prompts in a month.

    Why it works

    Make the entity unambiguous: Organization schema with "Kepler AI" as an alternate name, consistent naming on LinkedIn, X, Crunchbase, the Anthropic connector listing and press, a Wikidata entry, and an About page that states what Kepler is in the first sentence. Then check what ChatGPT, Claude and Perplexity say about "Kepler AI finance".

    "Search and AI answers" is in the Founding Marketer's brief. Disambiguation is cheap, compounds, and stops prospects who search after a conference from landing on the broker.

    Risks and how to handle them

    • Changing the name in public feels like a rebrand

      Keep "Kepler" as the brand and use "Kepler AI" only where disambiguation is needed.

    Evidence

    • programmable.infoA news scan for "Kepler" found 87 mentions in 30 days; the latest was about Kepler Cheuvreux, a European broker, cutting a price target. Kepler's X handle is @kepler_ai_hq.Source
    • HiringFounding Marketer (first marketing hire): owns the brand system ("spread across a brandbook, a website, and a dozen decks that don't quite match"), the website ("you'll ship to it yourself with Claude"), positioning, customer proof from live engagements, the template an analyst sees in a Kepler output, conference materials, the CEO's LinkedIn and X, the blog, recruiting posts, launches and partner co-marketing, and "Search and AI answers".Source
    • HiringCompany facts: $7.25M seed (April 2025); investors include founders of OpenAI, Meta AI Research, MotherDuck, dbt Labs and Square, plus Pebblebed, Company Ventures and Mantis VC; founders with 40+ combined years at Palantir; the CEO scaled a data company to $15M ARR and was Citadel's first Head of Business Engineering.Source
  • 05

    Kepler is a verified connector in Claude with nothing to install, which is the shortest possible path to value, but the connector post ends with "Request Demo". Analysts who try a connector on a Tuesday night will not wait for a sales call.

    High impactMedium effort First result in 3 weeks CEO

    First action

    Decide the trial limits (public data only, 14 days, a daily query cap) and which email domains qualify.

    How to measure it

    Trial to sales call from investment firm domains

    Now
    –
    Target
    25%
    By
    8 weeks

    Steps

    1. 1

      Set the limits

      Public data only, 14 days, a daily cap, work email required.

      CEO 2 hours

    2. 2

      Build the trial account

      Sign up from the connector with no demo step.

      Full Stack Engineer 1 week

    3. 3

      Show the value fast

      First-run prompts that return a traced comp set and a model.

      Full Stack Engineer 2 days

    4. 4

      Flag firm domains

      Accounts from investment firms routed to a person on day 1.

      Full Stack Engineer 1 day

    5. 5

      Day-7 outreach

      A note that references what they asked, from the engineer who can help.

      Forward Deployed Engineer 30 minutes per account

    6. 6

      Update the connector post

      Replace "Request Demo" with "Start a 14-day trial".

      Founding Marketer 30 minutes

    First result
    3 weeks
    Cost
    Inference costs for trial usage, capped per account
    Owner
    CEO

    Smallest test

    Hand out 20 trial accounts to analysts who request demos and measure usage.

    Why it works

    Let anyone with a work email connect the Claude connector for 14 days on public data only (SEC filings, transcripts, market data), with a cap on queries. At day 7, the Forward Deployed Engineer or CEO reaches out to accounts from investment firms with what they asked.

    The connector is distribution Kepler already has. A public-data trial shows the traced answers without any private documents or security review, and the usage tells sales which firms to call.

    Risks and how to handle them

    • Trials attract students and non-buyers

      Require a work email and route only investment firm domains to sales.

    • Self-serve undercuts enterprise pricing

      Trial is public data only; private documents and firm ontology stay enterprise.

    Evidence

    • Your websiteKepler in Claude (September 2, 2026): a verified connector in Claude's connector directory, "nothing to install locally and no API key to manage", but the post ends with "Request Demo"; no trial or pricing is stated.Source
    • Your websiteProducts include Excel and PowerPoint add-ins, "Build on Kepler" (Kepler MCP and API, docs on Mintlify), workbooks and company pages in the app (an NVIDIA entity page shows 29,329 mentions across 139 filers).Source
    • HiringForward Deployed Engineer: embed with "lighthouse customers"; "Sit with a Portfolio Manager at an institution with $10B+ AUM"; ship a vertical app for a design partner; integrate with a client's research stack (Bloomberg, CapIQ) with provenance preserved; 30 to 50% travel.Source
  • 06

    The CEO came from Citadel and the team writes thoughtful research, but the posts are long essays. Analysts scroll LinkedIn between models; a single traced number with its arithmetic is a format they read and share.

    Medium impactLow effort First result in 4 weeks Founding Marketer

    First action

    Draft the first post from the Oura S-1 breakdown.

    How to measure it

    Demo requests citing LinkedIn

    Now
    –
    Target
    5 a month
    By
    8 weeks

    Steps

    1. 1

      Pick the number

      From the week's filings: a margin, a concentration, a restatement.

      Founding Marketer 1 hour a week

    2. 2

      Trace it in Kepler

      Screenshot of the trace and the arithmetic.

      Founding Marketer 30 minutes

    3. 3

      Draft in the CEO's voice

      Five lines and the image.

      Founding Marketer 30 minutes

    4. 4

      CEO edits and posts

      Same day each week.

      CEO 15 minutes

    5. 5

      Cross-post

      To X and the blog as a short note.

      Founding Marketer 15 minutes

    6. 6

      Track

      Profile views from finance titles and inbound demos.

      Founding Marketer 15 minutes a week

    First result
    4 weeks
    Cost
    $0
    Owner
    Founding Marketer

    Smallest test

    Post three in three weeks and compare engagement with the CEO's last three posts.

    Why it works

    Every week, the CEO posts one number from a fresh filing that headlines or AI tools got wrong or missed, with the screenshot of Kepler's trace and the arithmetic. The Founding Marketer drafts; the CEO edits and posts.

    The job post asks for the CEO's LinkedIn and X and for talks turned into posts. A repeatable format keeps the cadence without a content calendar, and each post is a demo.

    Risks and how to handle them

    • Calling out errors in public offends someone

      Point at the filing, not at a person or a named tool.

    Evidence

    • HiringFounding Marketer (first marketing hire): owns the brand system ("spread across a brandbook, a website, and a dozen decks that don't quite match"), the website ("you'll ship to it yourself with Claude"), positioning, customer proof from live engagements, the template an analyst sees in a Kepler output, conference materials, the CEO's LinkedIn and X, the blog, recruiting posts, launches and partner co-marketing, and "Search and AI answers".Source
    • HiringCompany facts: $7.25M seed (April 2025); investors include founders of OpenAI, Meta AI Research, MotherDuck, dbt Labs and Square, plus Pebblebed, Company Ventures and Mantis VC; founders with 40+ combined years at Palantir; the CEO scaled a data company to $15M ARR and was Citadel's first Head of Business Engineering.Source
    • programmable.infoThe site has 39 URLs: 21 blog posts and 5 research pieces (including "Reclassified to Conform", which found 194 of 200 companies restate statements in ways XBRL tags miss, and an Oura S-1 breakdown filed September 3, 2026), written mostly by engineers. There are no case study, customer or public company pages.Source

This quarter 8

The core plays for the next three months.

  • 07

    Kepler can build a traced three-statement model from an 8-K in minutes (the NVIDIA example ran with 26 citations). Earnings week is when every analyst is rebuilding the same models under time pressure.

    High impactMedium effort First result in First earnings season Founding Marketer

    First action

    List the 20 companies for the next earnings season and their report dates.

    How to measure it

    Analyst email sign-ups

    Now
    –
    Target
    1,000
    By
    end of the first earnings season

    Steps

    1. 1

      Pick the 20

      Mega caps and the most-debated names in your buyers' coverage.

      Founding Marketer 1 hour

    2. 2

      Template the breakdown

      Five numbers, one chart, one workbook, the trace links.

      Founding Marketer 2 days

    3. 3

      Run each report

      Kepler output reviewed by an engineer before posting.

      Forward Deployed Engineer 2 hours per report

    4. 4

      Email capture

      A sign-up for "the breakdown in your inbox the morning after".

      Full Stack Engineer 1 day

    5. 5

      Distribute

      CEO post, X, and the email list.

      Founding Marketer 30 minutes per report

    6. 6

      Review after the season

      Which companies drew sign-ups and demos.

      Founding Marketer 2 hours

    First result
    First earnings season
    Cost
    $0, about two days a week during earnings season
    Owner
    Founding Marketer

    Smallest test

    Publish five breakdowns in one week and count sign-ups.

    Why it works

    During earnings season, publish a traced breakdown for the 20 most-watched companies within 24 hours: headline numbers, what changed, a downloadable workbook, every figure linked to the filing. Send them to an email list of analysts who sign up.

    The Oura S-1 breakdown shows the format already works. Doing it at the moment of peak demand builds an audience of exactly the users Kepler sells to, and the workbook is the product.

    Risks and how to handle them

    • A published number is wrong

      Engineer review before posting and a public correction policy.

    Evidence

    • programmable.infoThe site has 39 URLs: 21 blog posts and 5 research pieces (including "Reclassified to Conform", which found 194 of 200 companies restate statements in ways XBRL tags miss, and an Oura S-1 breakdown filed September 3, 2026), written mostly by engineers. There are no case study, customer or public company pages.Source
    • Your websiteKepler in Claude (September 2, 2026): a verified connector in Claude's connector directory, "nothing to install locally and no API key to manage", but the post ends with "Request Demo"; no trial or pricing is stated.Source
    • HiringFounding Marketer (first marketing hire): owns the brand system ("spread across a brandbook, a website, and a dozen decks that don't quite match"), the website ("you'll ship to it yourself with Claude"), positioning, customer proof from live engagements, the template an analyst sees in a Kepler output, conference materials, the CEO's LinkedIn and X, the blog, recruiting posts, launches and partner co-marketing, and "Search and AI answers".Source
  • 08

    Kepler for Industry lists manufacturing, legal, insurance and healthcare with sample questions but no customers. The Forward Deployed Engineer role is built around shipping vertical apps for design partners.

    High impactMedium effort First result in 1 quarter Forward Deployed Engineer

    First action

    Pick the first industry (legal or manufacturing) based on which anonymous testimonial customer would join.

    How to measure it

    Design partners signed

    Now
    –
    Target
    3 in the first industry
    By
    end of quarter

    Steps

    1. 1

      Pick the first industry

      Where an existing relationship gives a first partner.

      CEO 1 day

    2. 2

      Write the offer

      90 days, what partners give, what they get, the price after.

      CEO 1 day

    3. 3

      Publish the page

      On the Industry page, with an application form.

      Founding Marketer 1 day

    4. 4

      Ask investors for intros

      Each investor names two possible partners.

      CEO 1 week

    5. 5

      Run the builds

      One FDE per partner, weekly demos.

      Forward Deployed Engineer 90 days

    6. 6

      Write the case studies

      Co-authored with each partner at day 90.

      Founding Marketer 1 week

    First result
    1 quarter
    Cost
    FDE time, discounted first-year pricing
    Owner
    Forward Deployed Engineer

    Smallest test

    Offer the program by email to five contacts in one industry.

    Why it works

    Publish a design-partner offer per industry: three seats, a 90-day build with a dedicated Forward Deployed Engineer, preferential pricing, and the right to co-author the first case study. Promote it to the partners and CEOs in the CEO's and investors' networks.

    A structured program turns "we could serve legal" into a pipeline with a deadline, gives the FDE work that becomes product, and produces the first proof in each industry.

    Risks and how to handle them

    • Too many verticals at once

      One industry until three partners are live.

    Evidence

    • Your websiteKepler for Industry: manufacturing, legal, insurance and healthcare (energy and government in the menu), with sample questions per industry; no named customers, and testimonials from an unnamed CEO at a global manufacturer and a Partner at a global law firm.Source
    • HiringForward Deployed Engineer: embed with "lighthouse customers"; "Sit with a Portfolio Manager at an institution with $10B+ AUM"; ship a vertical app for a design partner; integrate with a client's research stack (Bloomberg, CapIQ) with provenance preserved; 30 to 50% travel.Source
    • HiringCompany facts: $7.25M seed (April 2025); investors include founders of OpenAI, Meta AI Research, MotherDuck, dbt Labs and Square, plus Pebblebed, Company Ventures and Mantis VC; founders with 40+ combined years at Palantir; the CEO scaled a data company to $15M ARR and was Citadel's first Head of Business Engineering.Source
  • 09

    Anthropic already published a case study on Kepler and lists Kepler as a verified Claude connector. That is the strongest third-party proof Kepler has, and it has only been used as a link.

    High impactMedium effort First result in 6-8 weeks Founding Marketer

    First action

    Email the Anthropic contact from the case study with a one-paragraph session proposal and two dates.

    How to measure it

    Demos from the joint session

    Now
    –
    Target
    20
    By
    4 weeks after the session

    Steps

    1. 1

      Pitch the session

      One paragraph, two dates, the outline.

      Founding Marketer 1 hour

    2. 2

      Build the demo

      A live traced model in Claude with the connector.

      Forward Deployed Engineer 3 days

    3. 3

      Promote

      Both companies' channels and the CEO's network.

      Founding Marketer 2 weeks

    4. 4

      Run it

      CTO and the Anthropic speaker, 45 minutes.

      CTO 1 day

    5. 5

      Cut clips

      Five short clips for LinkedIn and sales follow-up.

      Founding Marketer 2 days

    6. 6

      Follow up

      Every registrant from an investment firm gets a personal note.

      CEO 1 day

    First result
    6 to 8 weeks
    Cost
    $0 to $2,000 for production
    Owner
    Founding Marketer

    Smallest test

    Ask Anthropic for a quote about Kepler for the Finance page first.

    Why it works

    Propose a joint 45-minute session with Anthropic's financial services team: how the CTO separates reasoning from retrieval, a live traced model in Claude, and a Q&A. Record it, cut it into clips, and send it to every open deal.

    "Joint pieces we ship with partners" is in the Founding Marketer's brief. Anthropic's audience of financial services teams adopting Claude is the right buyer, already sold on the model and looking for trustworthy data.

    Risks and how to handle them

    • Partner timelines slip

      Run a Kepler-only session with the case study if a date isn't set in four weeks.

    Evidence

    • Search resultsAnthropic case study (April 30, 2026): 26M+ SEC filings, 50M+ public and 1M+ private documents, 14,000+ companies across 27 global markets; the founders interviewed 147 financial firms before building; 94% accuracy mapping statement labels to taxonomy codes against 38 to 46% for other models. No customers are named.Source
    • Your websiteKepler in Claude (September 2, 2026): a verified connector in Claude's connector directory, "nothing to install locally and no API key to manage", but the post ends with "Request Demo"; no trial or pricing is stated.Source
    • HiringFounding Marketer (first marketing hire): owns the brand system ("spread across a brandbook, a website, and a dozen decks that don't quite match"), the website ("you'll ship to it yourself with Claude"), positioning, customer proof from live engagements, the template an analyst sees in a Kepler output, conference materials, the CEO's LinkedIn and X, the blog, recruiting posts, launches and partner co-marketing, and "Search and AI answers".Source
  • 10

    The first question from an analyst is whether a new tool fits their stack. Kepler has Excel and PowerPoint add-ins and the FDE builds integrations with CapIQ and Bloomberg for clients, but the site has no integration pages.

    Medium impactMedium effort First result in 6 weeks Forward Deployed Engineer

    First action

    Record a 60-second screen capture of the Excel add-in building a traced model.

    How to measure it

    Demos from integration pages

    Now
    –
    Target
    15
    By
    end of quarter

    Steps

    1. 1

      List what exists

      Integrations that are live today, and those built for clients.

      Forward Deployed Engineer 2 hours

    2. 2

      Record each one

      60 seconds, real filing, provenance visible.

      Forward Deployed Engineer 1 day

    3. 3

      Write the pages

      What it does, what stays traceable, setup steps.

      Founding Marketer 2 days

    4. 4

      Check client permissions

      Only publish integrations that aren't a client's private work.

      CEO 1 hour

    5. 5

      Link from the nav

      An Integrations item under Product.

      Founding Marketer 1 hour

    6. 6

      Use in follow-ups

      Send the matching page after each demo.

      Forward Deployed Engineer ongoing

    First result
    6 weeks
    Cost
    $0
    Owner
    Forward Deployed Engineer

    Smallest test

    Publish the Excel add-in page alone and send it after five demos.

    Why it works

    Publish one page per integration that exists today (Excel add-in, PowerPoint add-in, Claude connector, MCP and API) and, once delivered for a client, the research-stack integrations, each with a 60-second video and what provenance looks like inside that tool.

    Integration pages answer a buying objection and rank for searches like "AI Excel add-in for SEC filings". They also turn FDE work for one client into marketing for the next.

    Risks and how to handle them

    • Implying official partnerships with data vendors

      Describe integrations, not partnerships, and follow each vendor's trademark rules.

    Evidence

    • Your websiteProducts include Excel and PowerPoint add-ins, "Build on Kepler" (Kepler MCP and API, docs on Mintlify), workbooks and company pages in the app (an NVIDIA entity page shows 29,329 mentions across 139 filers).Source
    • HiringForward Deployed Engineer: embed with "lighthouse customers"; "Sit with a Portfolio Manager at an institution with $10B+ AUM"; ship a vertical app for a design partner; integrate with a client's research stack (Bloomberg, CapIQ) with provenance preserved; 30 to 50% travel.Source
    • programmable.infoThe site has 39 URLs: 21 blog posts and 5 research pieces (including "Reclassified to Conform", which found 194 of 200 companies restate statements in ways XBRL tags miss, and an Oura S-1 breakdown filed September 3, 2026), written mostly by engineers. There are no case study, customer or public company pages.Source
  • 11

    Conference materials are in the Founding Marketer's brief. Booths full of decks look like every other AI vendor; a live test with the visitor's own question is the demo Kepler is built for.

    High impactMedium effort First result in First event Founding Marketer

    First action

    Pick two events in the next four months where your buyers attend and book a small booth or demo table.

    How to measure it

    Meetings from each event

    Now
    –
    Target
    25
    By
    two weeks after each event

    Steps

    1. 1

      Pick the events

      Where portfolio managers and analysts attend, not vendor shows.

      CEO 1 day

    2. 2

      Build the station

      A screen, a printer, and a fast question form.

      Founding Marketer 1 week

    3. 3

      Staff with engineers

      An FDE answers follow-ups on the spot.

      Forward Deployed Engineer event days

    4. 4

      Print the answer

      One page with the trace, branded with the new template.

      Founding Marketer 1 day

    5. 5

      Follow up in 48 hours

      Email the answer and a trial or demo link.

      Founding Marketer 1 day

    6. 6

      Publish the best questions

      A post after each event with the hardest five.

      Founding Marketer 1 day

    First result
    First event
    Cost
    $8,000 to $20,000 per event
    Owner
    Founding Marketer

    Smallest test

    Run the "bring a filing" format at a small dinner before a booth.

    Why it works

    At the next two buy-side events, run a booth where visitors ask any question about any public filing and get a traced answer printed or emailed, with a leaderboard of the hardest questions. Collect the questions as content and follow up within 48 hours.

    The product claim is "AI that proves it's right". Letting skeptical analysts test it live is more convincing than any slide, and the questions show what buyers care about.

    Risks and how to handle them

    • A question exposes a gap live

      Treat it as product feedback and say so; it shows the system refuses rather than guesses.

    Evidence

    • HiringFounding Marketer (first marketing hire): owns the brand system ("spread across a brandbook, a website, and a dozen decks that don't quite match"), the website ("you'll ship to it yourself with Claude"), positioning, customer proof from live engagements, the template an analyst sees in a Kepler output, conference materials, the CEO's LinkedIn and X, the blog, recruiting posts, launches and partner co-marketing, and "Search and AI answers".Source
    • Your websiteHomepage: "AI that proves it's right." and "enterprise intelligence for high-stakes industries"; 26M+ SEC filings, 14,000+ companies, 120+ countries, 30+ languages; SOC 2 Type II. No named customers and no pricing; three unnamed testimonials (a Managing Director at a multi-billion dollar fund, a Partner at a global law firm, a VP at a global investment bank). The main CTA is "Request Demo".Source
    • Your websiteCompare page: Kepler against unnamed "frontier models on their own" and "finance AI tools"; worked examples (Alphabet operating margin 31.4% from a typical tool against 32.0% computed from the 10-K; total debt $10.9B against $35.3B; CoreWeave's Microsoft concentration of about 67% missed); research claims of 99.4% scale accuracy and 94% taxonomy mapping with a model trained for $9.Source
  • 12

    Kepler expects to grow several times over in two years, and its engineers already write research posts. Those posts are read by the same skeptical engineers it wants to hire and by the technical buyers at funds.

    Medium impactLow effort First result in 1 month Founding Talent

    First action

    Add role and demo links to the last five research posts.

    How to measure it

    Engineering applications citing a post

    Now
    –
    Target
    20%
    By
    end of quarter

    Steps

    1. 1

      Map posts to roles

      Each post to the role closest to its subject.

      Founding Talent 1 hour

    2. 2

      Add both endings

      Hiring line and demo line on every post.

      Founding Marketer 2 hours

    3. 3

      Share in alumni networks

      Palantir and Meta alumni groups, by the author.

      Founding Talent 30 minutes a post

    4. 4

      Ask applicants

      A "what brought you here" question in the application.

      Founding Talent 15 minutes

    5. 5

      Plan one post a fortnight

      Engineers pick topics; marketing edits.

      Founding Talent ongoing

    6. 6

      Report monthly

      Applications and demos attributed to posts.

      Founding Talent 30 minutes a month

    First result
    1 month
    Cost
    $0
    Owner
    Founding Talent

    Smallest test

    Add both links to the next post and track clicks.

    Why it works

    Give every research and blog post two endings: a short "we're hiring for this" line linking the matching role, and "see it on your filings" linking a demo. The Founding Talent lead shares each post in the Palantir and Meta alumni networks the team came from.

    Founding Talent is told the team's network is the best hiring source but hasn't been worked. Technical writing is the strongest recruiting content for engineers, and it proves depth to buyers at the same time.

    Risks and how to handle them

    • Engineers stop writing under pressure

      One post a fortnight across the whole team, edited by marketing.

    Evidence

    • HiringFounding Talent: "we're about a dozen people today and expect to grow several times over in the next two years"; the team's network is the best source of hires, with outbound and a few agency partners added recently.Source
    • programmable.infoThe site has 39 URLs: 21 blog posts and 5 research pieces (including "Reclassified to Conform", which found 194 of 200 companies restate statements in ways XBRL tags miss, and an Oura S-1 breakdown filed September 3, 2026), written mostly by engineers. There are no case study, customer or public company pages.Source
    • HiringCareers page: 10 open roles, all in New York: Engineering 5 (backend, full stack, frontend, agentic systems, machine learning), Business 2 (Founding Marketer, Founding Talent), Forward Deployed Engineering 1, Research 1 (Head of AI Research), plus general interest.Source
  • 13

    Kepler's backers include venture firms whose own teams read filings, model comps and diligence companies. They know the product, trust the founders and talk to other funds and LPs.

    High impactLow effort First result in 6 weeks CEO

    First action

    Email each fund partner an offer of a 60-day pilot for their team.

    How to measure it

    Qualified introductions from investors

    Now
    –
    Target
    15
    By
    end of quarter

    Steps

    1. 1

      Make the offer

      A 60-day pilot for each fund's investment team.

      CEO 1 day

    2. 2

      Run the pilots

      One FDE per fund for the first two weeks.

      Forward Deployed Engineer 60 days

    3. 3

      Collect a result

      What they used it for, and one number.

      Founding Marketer 1 hour per fund

    4. 4

      Ask for two introductions

      Peer funds or LPs, with the result attached.

      CEO 1 hour per fund

    5. 5

      Ask the angels

      A short list of finance contacts each one knows.

      CEO 1 week

    6. 6

      Track the referrals

      Introductions to meetings to pilots.

      CEO weekly

    First result
    6 weeks
    Cost
    $0
    Owner
    CEO

    Smallest test

    Offer one fund a pilot this month.

    Why it works

    Offer each fund on the cap table a 60-day pilot for its investment team, then ask each one to introduce Kepler to two peer funds or LPs with the pilot result. Ask the angel investors (founders of OpenAI, dbt Labs, MotherDuck, Square) for introductions to the finance teams and family offices they know.

    Warm investor introductions shorten trust-heavy enterprise sales, and an investor using the product is a reference other investors believe.

    Risks and how to handle them

    • Investors feel pressured

      Make the pilot free and genuinely optional.

    Evidence

    • HiringCompany facts: $7.25M seed (April 2025); investors include founders of OpenAI, Meta AI Research, MotherDuck, dbt Labs and Square, plus Pebblebed, Company Ventures and Mantis VC; founders with 40+ combined years at Palantir; the CEO scaled a data company to $15M ARR and was Citadel's first Head of Business Engineering.Source
    • HiringForward Deployed Engineer: embed with "lighthouse customers"; "Sit with a Portfolio Manager at an institution with $10B+ AUM"; ship a vertical app for a design partner; integrate with a client's research stack (Bloomberg, CapIQ) with provenance preserved; 30 to 50% travel.Source
    • Your websiteHomepage: "AI that proves it's right." and "enterprise intelligence for high-stakes industries"; 26M+ SEC filings, 14,000+ companies, 120+ countries, 30+ languages; SOC 2 Type II. No named customers and no pricing; three unnamed testimonials (a Managing Director at a multi-billion dollar fund, a Partner at a global law firm, a VP at a global investment bank). The main CTA is "Request Demo".Source
  • 14

    There is no pricing anywhere on the site, the connector post or the docs. Enterprise buyers still need to know whether a tool is $20k or $500k before they put an internal champion's reputation on it.

    Medium impactLow effort First result in 1 month CEO

    First action

    Write down the price of the last five deals and the pilot terms, and pick a starting price you are comfortable publishing.

    How to measure it

    Days from first call to pilot

    Now
    –
    Target
    -25%
    By
    end of quarter

    Steps

    1. 1

      List recent deals

      Price, unit and terms for the last five.

      CEO 1 hour

    2. 2

      Pick the anchor

      "Teams from $X a year" and what changes it.

      CEO 2 hours

    3. 3

      Write the page

      Unit, starting price, drivers, pilot terms, FAQ.

      Founding Marketer 1 day

    4. 4

      Add to demo form

      Ask team size so the first call starts with a quote.

      Founding Marketer 1 hour

    5. 5

      Link from the connector post

      So trial users see the path to paying.

      Founding Marketer 15 minutes

    6. 6

      Review after a quarter

      Did qualified demos rise or fall?

      CEO 1 hour

    First result
    1 month
    Cost
    $0
    Owner
    CEO

    Smallest test

    Put the starting price in the demo confirmation email first.

    Why it works

    Publish a short page: the pricing unit (seats, firm platform, or usage), a starting price for a team, what moves the price (private documents, firm ontology, deployment), and what a pilot costs.

    A price anchor qualifies buyers before the first call and lets an analyst champion get budget approval without a meeting. Competitors mostly hide prices, so a clear page is a difference.

    Risks and how to handle them

    • Published prices limit negotiation

      Publish a starting price only; keep enterprise quotes custom.

    Evidence

    • Your websiteHomepage: "AI that proves it's right." and "enterprise intelligence for high-stakes industries"; 26M+ SEC filings, 14,000+ companies, 120+ countries, 30+ languages; SOC 2 Type II. No named customers and no pricing; three unnamed testimonials (a Managing Director at a multi-billion dollar fund, a Partner at a global law firm, a VP at a global investment bank). The main CTA is "Request Demo".Source
    • Your websiteKepler in Claude (September 2, 2026): a verified connector in Claude's connector directory, "nothing to install locally and no API key to manage", but the post ends with "Request Demo"; no trial or pricing is stated.Source
    • Competitor researchCategory: Rogo reports 35,000+ professionals at 250+ institutions; AlphaSense passed $500M ARR with 6,500+ customers; Hebbia links every cell to a sentence-level citation. Most 2026 comparisons are written by vendors.Source

Big bets 3

Larger plays that can change the trajectory.

  • 15

    Kepler covers 14,000+ companies and already builds company pages in the app (NVIDIA: 29,329 mentions across 139 filers). The public site has 39 URLs. Every analyst search for a company's margin, debt or segment data is a chance to show a traced number.

    High impactHigh effort First result in 1 quarter Founding Marketer

    First action

    Build pages for the 50 most-searched companies and check them against the filings.

    How to measure it

    Trials from public company pages

    Now
    –
    Target
    300 a month
    By
    2 quarters after launch

    Steps

    1. 1

      Pick the 10 metrics

      Ones analysts search for and AI tools often misstate.

      Founding Marketer 1 day

    2. 2

      Build the template

      Metric, value, filing link, arithmetic, last updated.

      Full Stack Engineer 2 weeks

    3. 3

      Launch 50 companies

      Reviewed by hand before publishing.

      Forward Deployed Engineer 1 week

    4. 4

      Scale to 14,000

      Automated updates on each filing, with checks.

      Full Stack Engineer 6 weeks

    5. 5

      Add structured data

      So search and AI answers can cite the numbers.

      Founding Marketer 2 days

    6. 6

      Link to trials

      "Build the model in Kepler" on every page.

      Full Stack Engineer 2 days

    First result
    1 quarter
    Cost
    About two months of engineering plus compute
    Owner
    Founding Marketer

    Smallest test

    Publish 50 pages and watch search impressions for six weeks.

    Why it works

    Publish a public page per company with 10 traced metrics (revenue, margins, debt, cash flow, customer concentration, restatements), each linking to the filing line, updated with each filing, and an "Open in Kepler" button for the full model.

    Search and AI answers are in the Founding Marketer's brief, and AI assistants cite pages that state sourced numbers. Public pages show the product's difference on every visit and capture search demand no deck can.

    Risks and how to handle them

    • A wrong public number damages the brand

      Publish only metrics that pass deterministic checks and show the trace.

    • Data licensing limits

      Use only SEC and other public filings on public pages.

    Evidence

    • Your websiteProducts include Excel and PowerPoint add-ins, "Build on Kepler" (Kepler MCP and API, docs on Mintlify), workbooks and company pages in the app (an NVIDIA entity page shows 29,329 mentions across 139 filers).Source
    • programmable.infoThe site has 39 URLs: 21 blog posts and 5 research pieces (including "Reclassified to Conform", which found 194 of 200 companies restate statements in ways XBRL tags miss, and an Oura S-1 breakdown filed September 3, 2026), written mostly by engineers. There are no case study, customer or public company pages.Source
    • HiringFounding Marketer (first marketing hire): owns the brand system ("spread across a brandbook, a website, and a dozen decks that don't quite match"), the website ("you'll ship to it yourself with Claude"), positioning, customer proof from live engagements, the template an analyst sees in a Kepler output, conference materials, the CEO's LinkedIn and X, the blog, recruiting posts, launches and partner co-marketing, and "Search and AI answers".Source
    • Your websiteHomepage: "AI that proves it's right." and "enterprise intelligence for high-stakes industries"; 26M+ SEC filings, 14,000+ companies, 120+ countries, 30+ languages; SOC 2 Type II. No named customers and no pricing; three unnamed testimonials (a Managing Director at a multi-billion dollar fund, a Partner at a global law firm, a VP at a global investment bank). The main CTA is "Request Demo".Source
  • 16

    The FDE post names chemicals, legal and healthcare; the Industry page names manufacturing, legal, insurance and healthcare. A platform that serves every industry is hard to buy; one industry with one case study is easy to buy.

    High impactHigh effort First result in 2 quarters CEO

    First action

    Set the bar for choosing the industry: three live partners, one measured result, and a clear budget owner.

    How to measure it

    Signed customers in the new industry

    Now
    –
    Target
    5
    By
    2 quarters after launch

    Steps

    1. 1

      Choose the industry

      Using the design-partner results.

      CEO 1 day

    2. 2

      Build the product page

      Industry questions, sources, and the trace.

      Founding Marketer 1 week

    3. 3

      Publish the case study

      Co-authored with the partner.

      Founding Marketer 2 weeks

    4. 4

      Make a test pack

      25 domain questions with traced answers.

      Forward Deployed Engineer 1 week

    5. 5

      Pick one conference

      Where that industry's buyers gather.

      Founding Marketer 1 day

    6. 6

      Hire for it

      An FDE with domain experience.

      Founding Talent 1 quarter

    First result
    2 quarters
    Cost
    FDE time, one conference, and a domain hire
    Owner
    CEO

    Smallest test

    Run the domain test pack with three prospects before building the page.

    Why it works

    After the design-partner program, pick the industry with the strongest results and launch it properly: a dedicated product page, the co-authored case study, a domain-specific test pack like the finance one, and a presence at that industry's main conference.

    The trust layer is the same, but buyers buy from someone who knows their domain. A focused launch makes Kepler the default answer in one new market before others claim it.

    Risks and how to handle them

    • Splitting focus away from finance

      Only launch once finance has a repeatable sales motion.

    Builds on

    Evidence

    • Your websiteKepler for Industry: manufacturing, legal, insurance and healthcare (energy and government in the menu), with sample questions per industry; no named customers, and testimonials from an unnamed CEO at a global manufacturer and a Partner at a global law firm.Source
    • HiringForward Deployed Engineer: embed with "lighthouse customers"; "Sit with a Portfolio Manager at an institution with $10B+ AUM"; ship a vertical app for a design partner; integrate with a client's research stack (Bloomberg, CapIQ) with provenance preserved; 30 to 50% travel.Source
    • Competitor researchCategory: Rogo reports 35,000+ professionals at 250+ institutions; AlphaSense passed $500M ARR with 6,500+ customers; Hebbia links every cell to a sentence-level citation. Most 2026 comparisons are written by vendors.Source
  • 17

    "Build on Kepler" offers an MCP and API with docs, but access runs through sales. Internal AI teams at banks and fintech startups are building their own agents and need exactly what Kepler sells: traced numbers from filings.

    High impactHigh effort First result in 1 quarter CEO

    First action

    Count how many API and MCP accounts exist today and what they use.

    How to measure it

    Monthly API revenue

    Now
    –
    Target
    $20k
    By
    3 quarters

    Steps

    1. 1

      Define the free tier

      Public filings only, capped calls.

      CEO 1 day

    2. 2

      Self-serve keys and billing

      Sign up, key, usage meter, card.

      Full Stack Engineer 4 weeks

    3. 3

      Publish samples

      Agent frameworks and MCP clients.

      Agentic Systems Engineer 2 weeks

    4. 4

      Launch to developers

      A technical post and a demo video.

      Founding Marketer 1 week

    5. 5

      Route big users

      Usage above a threshold goes to an FDE.

      Forward Deployed Engineer ongoing

    6. 6

      Review pricing

      After a quarter of usage data.

      CEO 1 day

    First result
    1 quarter after launch
    Cost
    About two months of engineering, plus inference costs
    Owner
    CEO

    Smallest test

    Give 20 developers keys by hand and watch what they build.

    Why it works

    Open API keys with usage pricing for public-filing data and calculations, a free tier for prototyping, and sales-led plans for private documents. Publish code samples for the agent frameworks and the MCP clients analysts use.

    Kepler calls the harness, not the model, the product. Developers building agents are a second channel that needs no demo, and each production integration becomes an enterprise lead.

    Risks and how to handle them

    • Developers build competing front ends

      Keep the analyst workspace, private documents and ontology enterprise-only.

    Evidence

    • Your websiteProducts include Excel and PowerPoint add-ins, "Build on Kepler" (Kepler MCP and API, docs on Mintlify), workbooks and company pages in the app (an NVIDIA entity page shows 29,329 mentions across 139 filers).Source
    • Your websiteKepler in Claude (September 2, 2026): a verified connector in Claude's connector directory, "nothing to install locally and no API key to manage", but the post ends with "Request Demo"; no trial or pricing is stated.Source
    • HiringCareers page: 10 open roles, all in New York: Engineering 5 (backend, full stack, frontend, agentic systems, machine learning), Business 2 (Founding Marketer, Founding Talent), Forward Deployed Engineering 1, Research 1 (Head of AI Research), plus general interest.Source

Wildcards 4

Unexpected experiments, each with a cheap way to test it.

  • 18

    Kepler showed a frontier model giving two different Alphabet margins (31.5% and 29.8%) when the 10-K says 32.0%. Nobody measures this systematically, and the financial press covers AI reliability constantly.

    High impactMedium effort First result in 8 weeks Head of AI Research

    First action

    Write the 500-question set from the most recent earnings season.

    How to measure it

    Press citations

    Now
    –
    Target
    5 per report
    By
    second report

    Steps

    1. 1

      Design the set

      500 questions, balanced by type, with ground truth.

      Head of AI Research 2 weeks

    2. 2

      Run the tools

      The same prompts across assistants, recorded.

      Head of AI Research 1 week

    3. 3

      Score openly

      Publish the method and the raw answers.

      Head of AI Research 1 week

    4. 4

      Write the report

      Five findings, one chart each.

      Founding Marketer 1 week

    5. 5

      Brief reporters

      Financial and AI desks, a week before.

      Founding Marketer 1 week

    6. 6

      Repeat quarterly

      New filings, same method.

      Head of AI Research quarterly

    First result
    8 weeks
    Cost
    Under $2,000 in tool usage per report
    Owner
    Head of AI Research

    Smallest test

    Publish a 50-question version as a blog post.

    Why it works

    Each quarter, ask the leading AI assistants 500 questions about fresh filings, score each answer against the traced figure, and publish the error rates by question type with the full dataset. Pitch it to financial and AI reporters.

    The Head of AI Research leads evaluation; publishing a rigorous, repeated measurement builds authority that a vendor claim cannot, and every reporter who covers it explains Kepler's thesis for them.

    Risks and how to handle them

    • Seen as self-serving

      Publish the method and data, and include Kepler's own refusals and misses.

    Evidence

    • Your websiteCompare page: Kepler against unnamed "frontier models on their own" and "finance AI tools"; worked examples (Alphabet operating margin 31.4% from a typical tool against 32.0% computed from the 10-K; total debt $10.9B against $35.3B; CoreWeave's Microsoft concentration of about 67% missed); research claims of 99.4% scale accuracy and 94% taxonomy mapping with a model trained for $9.Source
    • programmable.infoThe site has 39 URLs: 21 blog posts and 5 research pieces (including "Reclassified to Conform", which found 194 of 200 companies restate statements in ways XBRL tags miss, and an Oura S-1 breakdown filed September 3, 2026), written mostly by engineers. There are no case study, customer or public company pages.Source
    • Search resultsAnthropic case study (April 30, 2026): 26M+ SEC filings, 50M+ public and 1M+ private documents, 14,000+ companies across 27 global markets; the founders interviewed 147 financial firms before building; 94% accuracy mapping statement labels to taxonomy codes against 38 to 46% for other models. No customers are named.Source
  • 19

    Investor relations and finance teams spend months on filings, and AI assistants now summarise them for investors, sometimes wrongly. Kepler can show a company exactly how its last 10-K is misread and what is restated in ways tags miss.

    Medium impactMedium effort First result in 6 weeks Founding Marketer

    First action

    Run the report on three companies with recent restatements and send it to their IR contacts.

    How to measure it

    IR report requests

    Now
    –
    Target
    25
    By
    8 weeks

    Steps

    1. 1

      Build the report template

      Misreadings, missed restatements, traced figures.

      Head of AI Research 1 week

    2. 2

      Run three examples

      Companies with known restatements.

      Head of AI Research 3 days

    3. 3

      Find IR contacts

      From each company's investor site.

      Founding Marketer 1 day

    4. 4

      Send personally

      A short note with their report attached.

      CEO 1 day

    5. 5

      Publish a request page

      Any public company can ask for theirs.

      Founding Marketer 2 days

    6. 6

      Follow up

      A call to walk through the findings.

      CEO ongoing

    First result
    6 weeks
    Cost
    $0
    Owner
    Founding Marketer

    Smallest test

    Send three reports and see how many reply.

    Why it works

    Offer IR teams a free report on their latest filing: what popular AI tools get wrong, the restatements XBRL tags miss (from the "Reclassified to Conform" method), and the traced figures. Use it to start conversations with corporates and with the funds that cover them.

    It opens a new buyer (corporate finance and IR) with a problem they did not know they had, and puts Kepler in front of the companies analysts cover.

    Risks and how to handle them

    • Companies read it as criticism

      Frame it as protecting how investors see their numbers.

    Evidence

    • programmable.infoThe site has 39 URLs: 21 blog posts and 5 research pieces (including "Reclassified to Conform", which found 194 of 200 companies restate statements in ways XBRL tags miss, and an Oura S-1 breakdown filed September 3, 2026), written mostly by engineers. There are no case study, customer or public company pages.Source
    • Your websiteCompare page: Kepler against unnamed "frontier models on their own" and "finance AI tools"; worked examples (Alphabet operating margin 31.4% from a typical tool against 32.0% computed from the 10-K; total debt $10.9B against $35.3B; CoreWeave's Microsoft concentration of about 67% missed); research claims of 99.4% scale accuracy and 94% taxonomy mapping with a model trained for $9.Source
    • Your websiteKepler for Industry: manufacturing, legal, insurance and healthcare (energy and government in the menu), with sample questions per industry; no named customers, and testimonials from an unnamed CEO at a global manufacturer and a Partner at a global law firm.Source
  • 20

    Analysts adopt the tools they learned on. Finance clubs and pitch competitions model the same filings every term, and they are a hiring pool for a company expecting to grow several times over.

    Medium impactLow effort First result in One term Founding Talent

    First action

    Email the presidents of five finance clubs with the offer.

    How to measure it

    Active student users per term

    Now
    –
    Target
    500
    By
    end of the next term

    Steps

    1. 1

      Pick ten schools

      Where your buyers recruit analysts.

      Founding Talent 1 day

    2. 2

      Set student access

      Public data, .edu email, term-length accounts.

      Full Stack Engineer 2 days

    3. 3

      Sponsor two competitions

      A traced-model prize.

      Founding Talent 1 week

    4. 4

      Run a workshop

      An engineer shows how tracing works.

      Forward Deployed Engineer 1 hour per school

    5. 5

      Recruit winners

      Internship invitations.

      Founding Talent ongoing

    6. 6

      Track alumni

      Students who later sign up from a firm domain.

      Founding Talent yearly

    First result
    One term
    Cost
    $5,000 to $10,000 in prizes and travel
    Owner
    Founding Talent

    Smallest test

    Offer it to one club for one competition.

    Why it works

    Give finance clubs at ten target schools free public-data access for pitch competitions, sponsor two competitions with a "best traced model" prize, and invite winners to apply for internships.

    It builds habit in future buyers at low cost and gives Founding Talent a pipeline of finance-savvy engineers and analysts.

    Risks and how to handle them

    • Student usage costs compute

      Cap queries per account.

    Evidence

    • HiringFounding Talent: "we're about a dozen people today and expect to grow several times over in the next two years"; the team's network is the best source of hires, with outbound and a few agency partners added recently.Source
    • Your websiteKepler in Claude (September 2, 2026): a verified connector in Claude's connector directory, "nothing to install locally and no API key to manage", but the post ends with "Request Demo"; no trial or pricing is stated.Source
    • Your websiteProducts include Excel and PowerPoint add-ins, "Build on Kepler" (Kepler MCP and API, docs on Mintlify), workbooks and company pages in the app (an NVIDIA entity page shows 29,329 mentions across 139 filers).Source
  • 21

    Every finance AI tool says it cites sources. Kepler says the model never touches the numbers. A public bounty turns that architecture claim into a promise buyers can test.

    Medium impactLow effort First result in 1 month CEO

    First action

    Define what counts as untraceable and how a claim is checked.

    How to measure it

    Deals where the bounty was raised

    Now
    –
    Target
    30% of demos
    By
    end of quarter

    Steps

    1. 1

      Write the rules

      What counts, how to submit, how it is checked.

      CEO 2 hours

    2. 2

      Set up submissions

      A form that captures the output and the number.

      Full Stack Engineer 1 day

    3. 3

      Check claims fast

      Within two business days, by an engineer.

      CTO ongoing

    4. 4

      Publish fix notes

      Each paid claim with what was fixed.

      Founding Marketer as needed

    5. 5

      Add to the compare page

      One line and the running count.

      Founding Marketer 1 hour

    6. 6

      Report quarterly

      Claims, payouts, fixes.

      CEO 1 hour a quarter

    First result
    1 month
    Cost
    $250 per valid claim
    Owner
    CEO

    Smallest test

    Offer the bounty to current customers only for one quarter.

    Why it works

    Publish a standing offer: any customer or trial user who finds a number in a Kepler output that does not trace to its stated source gets $250 and a public fix note. Report the count every quarter.

    It gives the compare page a concrete guarantee, gives sales a line no competitor can match, and creates a feedback loop the Head of AI Research can use.

    Risks and how to handle them

    • Many claims at once

      Limit to outputs from the current release and cap payouts per user.

    Evidence

    • Your websiteCompare page: Kepler against unnamed "frontier models on their own" and "finance AI tools"; worked examples (Alphabet operating margin 31.4% from a typical tool against 32.0% computed from the 10-K; total debt $10.9B against $35.3B; CoreWeave's Microsoft concentration of about 67% missed); research claims of 99.4% scale accuracy and 94% taxonomy mapping with a model trained for $9.Source
    • Your websiteHomepage: "AI that proves it's right." and "enterprise intelligence for high-stakes industries"; 26M+ SEC filings, 14,000+ companies, 120+ countries, 30+ languages; SOC 2 Type II. No named customers and no pricing; three unnamed testimonials (a Managing Director at a multi-billion dollar fund, a Partner at a global law firm, a VP at a global investment bank). The main CTA is "Request Demo".Source
    • Search resultsAnthropic case study (April 30, 2026): 26M+ SEC filings, 50M+ public and 1M+ private documents, 14,000+ companies across 27 global markets; the founders interviewed 147 financial firms before building; 94% accuracy mapping statement labels to taxonomy codes against 38 to 46% for other models. No customers are named.Source

03

How Programmable can help

Where an idea in this playbook is something Programmable builds or runs, it's listed here with what we'd do. Everything else is yours to run.

ProgrammableTalk

Fill the events in this plan

These ideas are events. Talk gets registrants for them and tells you who's coming and what their company does before the day.

Plan an event (opens programmable.talk in a new tab)
ProgrammableDigital

Show it, don't just describe it

These ideas need something to show: a demo, a video, an animation, graphics or a deck. Digital turns screenshots or the site into all of them, in the company's own brand.

Make a demo (opens programmable.digital in a new tab)
More example playbooks and reviews

More growth playbooks

Other companies we have researched, each read from what they are hiring for.

  • LegalOS growth strategyAI-assisted U.S. employment immigration: attorneys plus AI workflows that prepare and file work visas (H-1B, L-1A, L-1B, O-1, TN) and…22 growth ideas
  • Attentive growth strategyAn AI marketing platform for 1:1 personalization across SMS, RCS, email and push, with identity (Signal), AI sign-up, AI journeys, MCP…22 growth ideas
  • Pulse growth strategyA free app for making friends in real life: it matches people into small groups, breaks the ice with an AI guide (Pip) and helps them plan…20 growth ideas
  • VaynerMedia growth strategyA global integrated creative and media agency, "the Modern Agency of Record", with social at the center: creative, media, strategy, PR…21 growth ideas
  • White Elm growth strategyA women-owned vegan bag brand from Georgetown, Texas: convertible backpacks, work and travel totes, crossbody bags, wallets and…22 growth ideas
  • Libas growth strategyA fast-fashion Indian wear brand for women: kurtas, kurta sets, suits, sarees and co-ords, plus sub-brands Libas Extra Love (plus size)…23 growth ideas
  • Aiwyn growth strategyA platform for accounting firms covering payments and collections, practice management, a client experience portal with engagement…21 growth ideas
  • Payhip growth strategyAn e-commerce platform for creators to sell digital downloads, courses, coaching, memberships and physical products from their own store…21 growth ideas

See every playbook and website review

Independent research. Programmable wrote this playbook from public sources, as they stood on October 18, 2026, and cites them throughout. Kepler did not commission, review or endorse it, and Programmable is not affiliated with Kepler. Company and product names are trademarks of their owners, used only to identify them. The ideas are suggestions and opinion, not professional, legal or financial advice, and not statements about how Kepler runs its business.

Something wrong, or do you represent Kepler? Email [email protected] and we will correct it or take it down.

30/60/90 roadmap

What to focus on in each month, and the ideas that carry it.

  1. Days 1 to 308 ideas

    Make the claim testable and visible: the test pack, the output footer, the first case study, the name fix, the connector trial and the weekly CEO post; then plan earnings breakdowns and the design-partner program.

  2. Days 31 to 607 ideas

    Partners and proof: the Anthropic session, integration pages, the first conference booth, the dual-use blog, investor pilots, the pricing page and the bounty.

  3. Days 61 to 906 ideas

    Reach beyond today's buyers (the AI audit, IR reports, finance clubs), then commit to the big bets on the first 60 days of evidence.

The research behind the plan

3 research areas, each summarised in a few lines. Open any of them for the full detail.

Customers

What customers say

Customers value every number traced to the filing line. The most cited problem: "I can't put a number in front of a client if I can't show where it came from".

  • Calculations that give the same answer every time
  • Double-checking every number AI produces
  • In their words: “show where it came from”

Market

Competitive landscape

4 competitors compared: AlphaSense, Rogo, Hebbia and Frontier models alone.

  • Gap to exploit: Numbers computed by code, not written by a model
  • AlphaSense: Search first, not computed and traced numbers
  • Rogo: Model-written numbers

Hiring signals

Kepler has 10 open roles, all in New York: five engineers, a Head of AI Research, a Forward Deployed Engineer, and its first two business hires, a Founding Marketer and Founding Talent.

  • Founding Marketer: First marketing hire: brand, website, customer proof, the analyst output template, the CEO's LinkedIn and X, partner…
  • Forward Deployed Engineer: Lighthouse customers and design partners, vertical apps, Bloomberg and CapIQ integrations: enterprise land and expand…
  • Founding Talent: About a dozen people now, growing several times over in two years; the network is the best source. Ideas 12, 20.

From the team that wrote this playbook

All products
  • ProgrammableAgencyWorking AI systems in days, not quarters (opens programmable.agency in a new tab)
  • ProgrammableInfoAn agent that finds, researches and asks first (opens programmable.info in a new tab)
  • ProgrammableDigitalScreenshots in, an animated product demo out (opens programmable.digital in a new tab)
  • ProgrammableWatchKnow the day a supplier falls out of compliance (opens programmable.watch in a new tab)
  • ProgrammableTalkGet registrants for your events, and know who’s coming (opens programmable.talk in a new tab)
  • ProgrammablePageText what you need, your customer gets a page that works (opens programmable.page in a new tab)
  • ProgrammableCommunityForums with labelled AI members, never empty (opens programmable.community in a new tab)
ProgrammableMarketing
ExamplesLeaderboardsProductsTermsPrivacyCookiesDPALegalContact

© 2026 AI-Powered Growth Intelligence