You're a junior investor at a fund backing young companies in the Middle East and North Africa. The best ones often never email you. Describe the kind of startup your fund wants to back. Alpha Scout returns a scored shortlist from the web, each score backed by a quote and a link.
A fund that waits for founders to come to it meets them after faster funds do. The vision: you search from the profile of one of your best portfolio companies, on the sources you trust, and every company on the shortlist comes with checked evidence.
You find startups through founders who contact the fund, and through people you know.
Funding stage, size and region in news articles are often wrong or out of date.
AI-written company summaries can invent facts.
Companies found in different places have no common scores.
The same ineligible companies come back in every search.
| Your job | Hypothesis | Metric and target, then result | Expected impact |
|---|---|---|---|
| When I look for deals, I want startups like our best portfolio company found on sources I trust, so I see them before they ask for money. | If you search from the profile of one of your best portfolio companies, you find eligible companies you didn't know, because the search covers sources you don't read daily. | Share of shortlisted companies not yet in the CRM: up. Baseline unobserved. Guardrail: share outside the thesis stays flat | More first meetings with founders (unobserved) |
| When a company shows up, I want its stage, size and region checked, so I only review companies we can back. | If each company is checked against three sources, fewer ineligible companies reach you, because one article is often out of date. | Share of shortlisted companies you mark ineligible: down. Baseline unobserved | Less time on companies you can't back (unobserved) |
| When I read a score, I want its quote and link, so I can check it in seconds. | If every score shows a checked quote, you verify faster and trust the shortlist more, because you check one quote instead of reading the article. | Share of scores whose quote is found in the source: up. Measured per company; not yet added up | A shortlist that holds up at committee |
| When I compare companies, I want the same 4 criteria for all, so my colleague and I agree on who to call first. | If companies are scored on the same 4 criteria, two investors agree more on who to contact first, because they compare like with like. | Top-3 picks two investors share on the same search: up. Baseline unobserved | Faster agreement on who to contact (unobserved) |
| When I search again, I want known misses skipped, so I don't pay twice. | If ineligible companies go on a skip list, later searches make fewer paid API calls per eligible company, because known misses are skipped before the extra checks. | API calls per eligible company over repeated searches: down. Baseline unobserved | Lower cost per eligible company |
No investor has measured how many results turn out useful. Every baseline is unobserved.
A company that fits your fund gets written up on a regional startup news site. Nobody at your fund reads that article, so a rival fund calls the founders first.
The weak link is belief, for both. The cheapest fix: track 3 months of shortlists and count how many lead to a first meeting with founders.
The model companies are fictional. Results name real companies found in public sources. Live searches share a free allowance that resets daily. Video walkthrough (YouTube).
This build tackles 1 to 5. Outreach is a non-goal.
| Metric | Type | Tests hypothesis | How measured |
|---|---|---|---|
| Share of shortlist not yet in the CRM | Leading | 1 | CRM match per result |
| Share an investor marks ineligible | Leading | 2 | Label per result |
| Share of quotes found in their source | Leading | 3 | Quote-check flag per score |
| Top-3 picks shared by two investors | Leading | 4 | Both rank the same search |
| Shortlisted companies that get a first meeting | Lagging | 1 to 4 | CRM status after 3 months |
| In scope | Non-goal (incl. untackled sub-problems) |
|---|---|
| Search for similar companies from a model company profile, on chosen sources | 6 Outreach: no automatic contact. The investor decides and writes |
| Website, LinkedIn and funding stage checks; size, region and stage filters | Ranking pitch decks founders send in: that's Deck Rank |
| Scoring on 4 criteria with quote, link and quote-check flag | Running the AI model in-house to keep data private |
| Comparison table, 2x2 matrix, report, Excel and PDF export | Syncing with the CRM |
| Target list, saved searches, skip list, chat with the AI about targets | Investment decisions |
| Chose | Over | Gain | Cost |
|---|---|---|---|
| search only chosen sources | search the whole web | the investor controls quality | misses companies those sources haven't covered |
| drop names not found in source text | accept every name the AI pulls out | no invented companies | fewer companies |
| check eligibility on 3 sources | trust the article | fewer ineligible results | about 3 extra searches per company |
| skip list for ineligible companies | re-check every time | cheaper repeat searches | a company that grows into eligibility stays hidden until removed |
Shared goal: see the best companies the fund wants to back first, and be the first to commit. Alpha Scout is the half where the fund goes looking (see first). Deck Rank is the half that ranks pitch decks founders send in (act first).
| Stage | Artifact | Repeatable with |
|---|---|---|
| Discovery | User stories table per persona (README) | /prd-draft |
| Strategy | 3 ways to find companies: like a portfolio company, like regional success stories, from founders who apply | /write-prod-strategy |
| Build | AI steps for search, extra checks and scoring, each logged in Langfuse (an AI monitoring tool) | /code-first-draft |
| Evaluation | An AI grading the scores, quality checks on the extra checks, review step | /feature-metrics |
| Launch | Video walkthrough | /prototype |
| Showcase | This page, real-name check, demo with usage limits, saved examples | /portfolio-case-study, /name-audit |
Skills are mapped from the artifact each stage produced. They show how each stage repeats with PM OS; the original working sessions were not logged.
/prd-draft /write-prod-strategy /code-first-draft /feature-metrics /prototype /portfolio-case-study /name-audit