AI Prompts for Angel and Early-Stage Investment Diligence
The one variable most closely linked to angel returns is the one newly liquid founders skip — because operating instinct feels like diligence, and it isn’t.
The five things that matter
- Hours, not shortcuts. Above the 20-hour median, 45% of exits returned less than 1x. Below it, 65% did. Same deals, different depth.
- Ask the magnitude question, not the quality question. Operators ask “is this a good business?” Angels have to ask “can this return the whole portfolio?” A durable, profitable company can still be a bad angel investment.
- Your diligence hours are mostly going to the wrong places. The deck and the model are the lowest-information sources available — the founder controls both. Customers and off-list references carry far more per hour.
- Portfolio size beats deal selection. Twenty to twenty-five positions is the commonly cited floor for a high probability of positive return. Under ten has been observed returning less than capital in.
- Total loss is the base case for any single position, not the downside scenario. Roughly half fail completely; another 30–40% return less than 1x.
Recently liquid founders and operators writing their first 5–25 angel cheques, and anyone whose deal flow arrives mostly through people they know.
You’re a professional investor with an established process — this will be familiar. Also skip if you’re not yet clear on whether this capital is genuinely loseable.
On this page
How much diligence should an angel actually do?
Twenty hours per deal is a floor, not a target. Research on angel investors in groups found diligence time was among the clearest correlates of return in the whole dataset.
Use these prompts to spend the same hours on better questions — not fewer hours on the same ones.
That’s an awkward thing for a site that publishes AI prompt systems to say. It’s also the only reading of the data that survives contact with it.
Why do successful founders make poor angel investors?
Because operating pattern recognition feels like diligence. You’ve seen a hundred pitches, you know what a real customer conversation sounds like, and you can spot a fake metric in ten seconds. All of that is real — and none of it is diligence. It’s a very fast first filter that feels like a conclusion.
Five specific failure modes, and they compound:
You were a founder. You rate founders well because you have a rich model of what good looks like — which makes you confident, and confidence shortens diligence. But founder quality is the hardest thing to assess from outside and the easiest to be charmed by.
Having built a B2B SaaS company doesn’t make you expert in this B2B SaaS market. The research suggests expertise correlates with returns — but it’s expertise in the specific area, not general operating experience.
The deals that look most like your own company get the least scrutiny, because you feel you already understand them. That’s exactly backwards: recognition is a reason to check whether you’re pattern-matching to your own experience rather than to this business.
Post-exit capital tempts concentration. But returns follow a power law, and the commonly cited floor for a high probability of positive return is 20–25 positions. Five large cheques is not a portfolio, it’s five lottery tickets with your name on them.
After a visible exit, deals arrive from people who know you. That’s a filter on your social graph, not on quality — and it comes wrapped in relationship pressure that makes a clean “no” expensive. This is the least discussed and possibly most damaging of the five.
What’s the difference between validating an idea and evaluating an investment?
The founder is asking a yes/no question. The investor is asking a magnitude question. If you’ve used our idea validation prompts, this guide is the same machinery pointed the other way — and the inversion is not cosmetic.
“Will this work?”
A yes/no about viability. A business that reaches profitability and sustains itself is a success.
“Can this return my entire portfolio?”
A magnitude question about the tail. That same profitable, sustainable business may be a failed angel investment.
A company can be good and still be a bad angel investment. Founder-angels miss this more than any other single thing, because “is this a good business?” is the operator’s question and they never stop asking it.
The mechanism is portfolio maths. If roughly half your positions go to zero and a further 30–40% return less than 1x, the portfolio only works if something in it returns enormously. A company that gets to $8m of revenue, throws off cash and never sells has produced a fine outcome for its founder and a write-off-shaped outcome for you.
Prompt 1: What would have to be true?
Start here, before any other analysis. This converts a pitch into a testable list — and the list is your diligence plan.
## ROLE You are helping me assess an early-stage investment. Your value is not enthusiasm and not caution — it is converting a narrative into a list of specific conditions I can go and test. I will do the testing. ## THE OPPORTUNITY What they do, in my own words: [WRITE IT YOURSELF. If you can't explain it in four sentences without their deck, that's your first finding.] Stage / raise / valuation: [ROUND, AMOUNT, PRE OR POST] My cheque: [AMOUNT] · My target portfolio size: [N POSITIONS] Traction as stated: [NUMBERS THEY GAVE — mark each as VERIFIED or CLAIMED] What I know independently: [ANYTHING NOT FROM THEM] ## PRODUCE 1. THE MAGNITUDE CASE. For this to return 20x+ on my cheque, what would have to be true? List 8–12 specific, checkable conditions — not "the market grows" but "they reach X customers at Y price by Z". 2. RANK BY (IMPORTANCE × UNCERTAINTY). Which conditions most determine the outcome AND are least established? That ranking is my diligence plan — say so explicitly and order it. 3. THE LOAD-BEARING ASSUMPTION. Of everything above, which single condition, if false, kills the investment? State it in one sentence. Then: how would I find out cheaply? 4. THE CEILING TEST. Assume they execute perfectly and everything goes right. How big can this realistically get? If the honest ceiling doesn't support a 20x+ outcome at this valuation, say so plainly — a good business at the wrong entry price is still a poor angel investment. 5. WHAT I'M NOT ASKING. Based on this sector and stage, which standard diligence questions are missing from my framing above? Name them. 6. MY PATTERN-MATCHING RISK. Given my background [ONE LINE ON YOUR OPERATING HISTORY], where am I most likely to be substituting my own experience for analysis of this company? ## RULES - Use only what I've given you. Do not supply market figures, competitor names or benchmarks from memory — flag where I need to find real data and say what kind. - Do not tell me whether to invest. - Do not soften point 4. I want the ceiling honestly.
Prompt 2: The question set the founder hasn’t rehearsed
Founders have rehearsed the standard questions. “What’s your CAC?” produces a prepared answer that carries almost no information. Questions they haven’t anticipated carry a great deal.
THE COMPANY: [PASTE PROMPT 1'S OUTPUT] SECTOR / STAGE: [SPECIFIC] MY LOAD-BEARING ASSUMPTION: [FROM PROMPT 1, POINT 3] Generate three tiers of questions: TIER 1 — THE UNREHEARSED (8 questions). Questions a founder is unlikely to have prepared, that reveal thinking rather than positioning. Good examples of the type: - What do you believe about this market that most people working in it disagree with? - What would have to happen for you to shut this down? - Which customer did you lose that you still think about? - What's the last thing you changed your mind about? Generate 8 more in that spirit, specific to THIS company. For each, note what a good answer sounds like and what an evasive one sounds like. TIER 2 — THE LOAD-BEARING (5 questions). Questions that directly test my load-bearing assumption. These should be answerable with evidence, not opinion. For each: what evidence should exist if the answer is true, and what would its absence tell me? TIER 3 — THE UNCOMFORTABLE (4 questions). Questions I will feel socially awkward asking — about co-founder equity and vesting, previous failed initiatives, cap table cleanliness, why the last senior person left. For each, note the specific risk that not asking leaves on the table. THEN: A. THE ORDER. Which order should I ask these in, and why? B. THE FOLLOW-UP. For the three most important, what's the follow-up question if the first answer is vague? C. WHAT I SHOULD NOT ASK. Questions that will make me look naive or that the founder can't be expected to answer at this stage — and what to ask instead. Do not generate generic diligence checklists. If a question would apply unchanged to any company, replace it.
Tier 3 is the one people quietly delete. The uncomfortable questions have the highest information density in the entire process, precisely because the social cost of asking them means most investors don’t — so the answers haven’t been polished.
Prompt 3: The reference interrogation
The references a founder gives you will say good things. That’s what makes them references. The value is in what you ask, and in who you find that isn’t on the list.
THE COMPANY: [BRIEF] REFERENCE TYPE: [Customer / former colleague / investor / former employee / partner] ON-LIST OR OFF-LIST: [Did the founder provide them?] WHAT I NEED TO LEARN: [YOUR LOAD-BEARING ASSUMPTION] Produce: 1. THE CALL STRUCTURE. 20 minutes. What order, and where the real question sits. (It is rarely first and never last.) 2. THE CALIBRATION QUESTION. One question early that tells me how candid this person will be, so I know how to weight everything after it. 3. TEN QUESTIONS, phrased to make honesty easy. Prefer specific and behavioural over evaluative: NOT "were they good to work with?" BUT "walk me through the last time a deadline slipped — what happened?" 4. THE THIRD-PERSON ROUTE. Two questions that let the reference report a concern as someone else's view. People will say "some of the team felt..." long before they'll say "I thought...". 5. THE CLOSING QUESTION. The one that most often produces the useful answer, plus what to do with silence after it. Do not fill the silence. 6. WHO ELSE. Based on this company and stage, which off-list references would carry the most information, and how would I plausibly reach them? 7. READING THE SIGNAL. For a customer reference specifically: what distinguishes genuine enthusiasm from politeness? Give me three concrete tells. RULES: - Do not suggest anything deceptive. No pretexting, no misrepresenting who I am or why I'm calling. - Assume everything I say gets back to the founder. - Do not draft anything I'd be embarrassed to have quoted.
Prompt 4: Does this fit the portfolio I’m building?
The most common error isn’t picking a bad company. It’s picking good companies in a shape that can’t produce returns.
MY PORTFOLIO SO FAR Positions to date: [N] · Target total: [N] Total committed: [£/$] · Total allocated to this asset class: [£/$] Existing positions by sector/stage/geography: [LIST] Typical cheque: [AMOUNT] · Reserves for follow-on: [AMOUNT OR NONE] THIS OPPORTUNITY Sector / stage / geography: [___] Cheque: [AMOUNT] · Entry valuation: [___] Assess: 1. CORRELATION. How correlated is this with what I already hold? Not just sector — shared customer type, shared funding-environment dependency, shared macro exposure. Two "different" companies selling to the same buyer in the same downturn are one position. 2. PACE. At my current rate and cheque size, do I reach my target position count before my allocation runs out? Show the arithmetic. If the answer is no, say what has to change — cheque size, pace, or target. 3. RESERVES. What proportion am I holding back for follow-on, and is that consistent with how this asset class actually works? If I hold none, state plainly what I'm giving up. 4. THE CONCENTRATION QUESTION. If this position went to zero tomorrow — the base case for any single position — what would that mean financially and psychologically? If the honest answer to the second part is "a lot", the cheque is too big regardless of the company. 5. THE OPPORTUNITY COST. This cheque is not being written against cash. It's being written against the deals I see in the next 12 months. Is this in my top decile of what I expect to see? 6. THE RELATIONSHIP FLAG. If this deal came through a personal relationship, how would I assess it if a stranger had sent it? Name the difference. That difference is the relationship premium I'm paying. RULES: - Do not evaluate the company. This is portfolio structure. - Show arithmetic explicitly so I can check it. Do not round in a way that hides a problem. - If my structure can't produce power-law returns, say so directly.
A real deal, run through the sequence
A B2B workflow tool for mid-market logistics firms. $2m seed at $10m post. Founder is a second-time operator, previously exited a smaller company in an adjacent sector. Investor here is a recently exited SaaS founder considering $50k.
Conditions for a 20x+ outcome — top four by importance × uncertainty:
- That mid-market logistics firms will change workflow software at all. Stated ARR is $340k across 11 customers — but 11 customers cannot distinguish “solved a real problem” from “sold well to 11 people”.
- That the $31k average contract value holds or rises as they move beyond design partners. Early customers are frequently priced as relationships.
- That this becomes a platform rather than a feature. At a $10m post, a $50–80m acquisition returns 5–8x, not 20x. The 20x case requires this to be a standalone business, not an acqui-feature.
- That the second-time-founder advantage transfers across sector. Adjacent is not the same.
Load-bearing assumption: that logistics mid-market will switch. Cheapest test: four conversations with non-customer logistics operations managers about what they use now and what would make them change. Cost: about six hours. This is the highest-value six hours in the entire process.
Ceiling test: honest realistic ceiling looks like $40–90m enterprise value in a good outcome. At $10m post, that’s 4–9x. A 20x outcome requires the platform case in point 3 to be true. It isn’t yet demonstrated.
Pattern-matching risk: the investor built B2B SaaS. Their instinct will read the ARR curve as familiar and good. The relevant difference is buyer behaviour — logistics mid-market procurement is not SaaS mid-market procurement, and the investor has no data on it.
The deck was about the product. The diligence was about the buyer. Everything material sat in whether a conservative, thin-margin, relationship-driven industry changes software — a question the deck never addressed, because the founder had already answered it for themselves.
The ceiling test reframed the entire decision. This is plausibly a good company and, at this entry price, unlikely to be a portfolio-returning position. Those are different findings, and only the second one is the investor’s question.
The outcome wasn’t “no”. It was a specific, cheap, six-hour test with a pre-committed decision rule: if fewer than two of four operations managers describe active dissatisfaction with their current tool, pass. Deciding the rule before the calls is what stops you interpreting ambiguous answers favourably at the point you’ve already grown fond of the founder.
Level-up: the 40-hour allocator
This is the part competitors won’t have, and it follows directly from the data at the top of the page.
If diligence hours correlate with returns, the obvious next question is one almost nobody asks: where should the hours go? Most angels spend the bulk of theirs on the deck, the model and founder meetings — which are the three lowest-information sources available, because the founder controls all three.
I have [N] hours for diligence on this deal. Allocate them. THE DEAL: [PROMPT 1 OUTPUT] LOAD-BEARING ASSUMPTION: [FROM PROMPT 1] WHAT I ALREADY KNOW: [AND HOW I KNOW IT] MY ACCESS: [Who I can realistically reach — customers, operators in this sector, other investors, former staff] Produce: 1. THE ALLOCATION. Hours per activity, ranked by expected information per hour, not by convention. Include: founder meetings · customer conversations · off-list references · sector expert calls · market/desk research · financial review · legal and cap table · competitor product use. Justify each in one line. 2. THE INFORMATION HIERARCHY. Rank my available sources from highest to lowest information per hour, and be explicit that founder-controlled material sits near the bottom. Say why for each. 3. THE FIRST SIX HOURS. If I could only spend six, what would I do? This is my kill-fast test — the cheapest route to a defensible no. 4. THE DECISION RULE, WRITTEN IN ADVANCE. For each of the top three tests: what result means proceed, what result means pass, and what result means "keep looking". Write these before I run the tests. 5. THE DIMINISHING RETURN POINT. At what stage does more diligence stop adding information for a deal of this type? I want to know when I'm procrastinating rather than working. 6. WHAT I'LL BE TEMPTED TO SKIP. Given this deal's attractive features, which of the above will I most want to skip — and what does skipping it cost? RULES: - Do not allocate the majority of hours to reading materials the founder produced. - Assume I will feel time pressure from the round closing. Build the allocation to survive it. - Do not recommend a decision.
Point 4 is the discipline that makes the rest work. Diligence that doesn’t have a pre-committed decision rule becomes a search for permission. You will find something reassuring — there is always something reassuring — and you’ll weight it heavily because by hour thirty you like these people. Writing the rule at hour zero is the only defence, and it costs nothing.
Bad prompt vs good prompt
| Weak prompt | Why it fails |
|---|---|
| “Here’s a pitch deck. Should I invest?” | Asks for a decision the model cannot make, on material the founder controls. You’ll get a confident, fluent, evidence-free answer — and its confidence will feel like information. |
| “Analyse this company’s market opportunity.” | Invites the model to generate market sizing from memory. It will produce plausible figures that may be wrong, stale, or invented, and you have no way to tell which. |
| “What are the risks with this investment?” | Produces a generic risk list applicable to any startup. Reads as thorough, contains nothing you didn’t know. |
| “Summarise the deck and pull out the key metrics.” | This is the shortcut the data warns against. You’ve replaced reading with skimming and called it diligence. |
| Strong prompt | Why it works |
|---|---|
| “For this to return 20x+, what would have to be true? Rank by importance × uncertainty.” | Converts a narrative into a testable list, and the ranking is the work plan. Asks a magnitude question rather than a quality one. |
| “Do not supply market figures from memory. Flag where I need real data and say what kind.” | Closes off the single largest fabrication risk and converts it into a research task with a specification. |
| “Given my background in X, where am I most likely to substitute my experience for analysis?” | Uses the model for the one thing you cannot do unaided — seeing your own pattern-matching from outside. |
| “Write the decision rule before I run the test: what result means proceed, pass, or keep looking?” | Pre-commits the interpretation, which is the only real defence against motivated reasoning at hour thirty. |
What AI must never do here
| Never | Why |
|---|---|
| Supply market data from memory | Model-generated market sizes, growth rates and competitor figures are frequently wrong or stale and always confident. Every number needs a source you can open. |
| Do the portfolio arithmetic unchecked | Multi-step numerical reasoning remains unreliable. Reproduce any calculation that affects a decision in a spreadsheet. |
| Assess legal documents or the cap table | Term sheets, side letters, liquidation preferences and option pools carry consequences a model will miss. This is lawyer work. |
| Verify a claim | It cannot call a customer, check a bank statement or confirm a contract exists. It can only reason about what you’ve told it. |
| Substitute for the hours | The central finding of this page. Compressing diligence with AI may remove exactly the input that correlated with returns. |
| Hold confidential deal material | Don’t paste NDA’d decks, data room contents or founder financials into a consumer tool. Use an enterprise tier with appropriate data terms, and check your NDA first. |
Which model, and one setup note
Use a reasoning-tier model for Prompt 1 and the allocator — both reward careful multi-step analysis. Use a large-context model when working across a full data room. For anything numerical, verify independently regardless of tier.
The setup step worth doing before your next deal: write your investment thesis and portfolio parameters — target position count, cheque size, sectors you’ll consider, sectors you won’t — into a single document, and paste it at the top of every prompt in this guide. Diligence without a thesis becomes a series of independent judgements about whether each company is impressive, which is how portfolios end up correlated and undersized.
Model capability, disclosure rules and market conditions all change. We re-verify on each review cycle. Confirm anything decision-relevant independently.
The Angel Diligence Question Set
The four prompts plus the question bank — built to be worked through per deal rather than read once.
◦ 60+ sector-specific questions
◦ Tier 3: the uncomfortable questions
◦ Reference call scripts, 5 types
◦ The 40-hour allocation template
◦ Decision-rule worksheet
Free. Unbranded and free to use inside a syndicate or angel group.
Upgrade — the interactive version
The Diligence Hour Allocator. Enter deal stage, sector, your available hours and your access, and it returns an hour-by-hour plan ranked by information per hour — plus a decision-rule worksheet that locks your pass/proceed thresholds before you run the tests, and won’t let you edit them afterwards.
[[LINK WHEN BUILT — the locking behaviour is the whole point. A checklist you can revise mid-diligence is a rationalisation tool.]]
Common questions
How much due diligence should an angel investor do?
Research on angel investors in groups found that diligence hours were among the clearest correlates of returns, with investors spending forty or more hours per deal averaging a 7.1x multiple against 1.1x for those spending under twenty hours. Twenty hours per deal is a reasonable floor. It is worth noting this is a correlation rather than proof of causation, since hours may also proxy for expertise, access and post-investment engagement.
How do I evaluate a startup investment?
Ask what would have to be true for the investment to return more than twenty times, rather than whether the business is good. List those conditions, rank them by importance and uncertainty, and spend your diligence time on the most uncertain and most important. Then seek evidence that would disprove your thesis rather than evidence that supports it, and finally assess whether the position fits the portfolio you are building.
Why do successful founders often make poor angel investors?
Because operating pattern recognition feels like diligence and is not. Founders tend to over-weight the founder because they were one, to mistake operating expertise for sector expertise, to shorten diligence when a deal feels familiar, and to write too few and too large cheques. Deal flow also arrives through social proximity after an exit, which selects for who knows you rather than for quality.
How many angel investments do I need to make?
Commonly cited guidance is a minimum of twenty to twenty-five positions to reach a high probability of a positive overall return, because returns follow a power law in which a small number of investments produce most of the gain. Portfolios of fewer than ten companies have been observed returning less than the capital invested. Deciding your total number of positions before your first cheque is the most consequential decision you will make.
Can AI do due diligence for me?
No, and using it to reduce your hours may remove the thing that correlated with returns in the first place. AI is useful for generating better questions, structuring analysis, building the case against, and identifying what you have not asked. It cannot verify a claim, call a reference, read a cap table reliably, or assess legal documents. Every output is a starting point for human work rather than a substitute for it.
What questions should I ask a startup founder before investing?
The most useful questions are ones the founder cannot have rehearsed. Ask what they believe about the market that most people in it disagree with, what would have to happen for them to shut the company down, which customer they lost that they still think about, and what the last thing they changed their mind about was. Rehearsed answers to standard questions carry very little information.
Do I need to be an accredited investor to angel invest?
In the United States most early-stage private offerings are made under exemptions that require investors to be accredited, which is defined by income, net worth or certain professional certifications. Rules differ by jurisdiction and change over time. Confirm your status and the applicable exemption with a securities lawyer before investing, since the requirements sit on the company issuing the securities as well as on you.
What percentage of angel investments fail?
Commonly cited figures suggest roughly half of early-stage portfolio companies fail completely and a further thirty to forty per cent return less than the capital invested. Total loss is the base case for any individual position rather than the worst case. This is why portfolio size matters more than deal selection for most angels, and why capital committed to this asset class should be capital you can lose entirely.
Read next
About this guide
Narracomm is a communications and content strategy team. We build and test prompt systems inside live client work and revise them as models and conditions change. [REQUIRED BEFORE PUBLISHING: named reviewer with genuine angel, syndicate or venture experience — with credential and review date shown. This page discusses capital allocation decisions where the base case is total loss, which makes independent review non-negotiable rather than advisable.]
Freshness
Last updated: 8 August 2026.
Changelog — 8 Aug 2026: first published. Diligence-hours and portfolio-construction figures verified against the sources below. Correlation-not-causation caveat added to the headline statistic — most write-ups of this research omit it.
Review cycle: every 14 days. Only bump the update date when something material changed — a corrected figure, a new prompt, a replaced source. Re-dating an unchanged page is explicitly flagged in Google’s helpful-content guidance.
Sources & further reading
- Wiltbank & Boeker — Returns to Angel Investors in Groups (primary source, PDF)
- Angel Capital Association — the one factor that improves returns — the diligence-hours finding
- Seraf — Angel investing returns: research and reality
- Allied Venture Partners — The power law of angel investing
- Angel portfolio construction and diversification — position-count guidance
- Why angel investors lose money: the portfolio maths
- SEC — Accredited investor definition
- Google — Creating helpful, reliable, people-first content
Scope: this guide is general information about a diligence process. It is not investment, legal or tax advice, does not evaluate any specific company or security, and does not assess your suitability for this asset class. Angel investing carries a high probability of total loss on individual positions and positions are illiquid for extended periods. Return figures cited are from published research on historical angel group performance and are not predictive. Verify your accredited or qualified status and the applicable exemptions with a securities lawyer in your jurisdiction. Last reviewed: 8 August 2026 · Next review due within 14 days.
