AI came for the reference call, not the deck
Diligence automation went after the half of the raise founders do not control. What that changes about how you prepare.
A company called DiligenceSquared raised $5 million in March. The product is an AI voice agent that phones your customers.
Not your customers, exactly. The customers of whatever company a private equity firm is about to buy.
TechCrunch covered it on 5 March 2026. The founders came out of Blackstone, BCG and Google. Their pitch is a price comparison. They say a firm pays McKinsey, Bain or BCG somewhere between $500,000 and $1 million to interview dozens of corporate customers. They say they do the same work for $50,000. That is their claim, not an audited figure.
I build voice and CRM automation for a living. I read that and did not think about private equity.
I thought about the founders I sit next to every week.
This is not new. That is the point.
Bridgetown Research raised $19 million for a version of the same idea. Accel and Lightspeed co-led it. Lightspeed announced it on 26 February 2025.
So the pattern is about eighteen months old, not eighteen days.
Lightspeed’s own post says the product gets an investor a diligence report in 24 hours instead of weeks. That is an investor describing their own investment, so weigh it accordingly.
The speed is not what interests me. The direction is.
Diligence has two halves
Half one is reading what the founder handed over. The deck, the model, the data room, the metrics page.
Half two is finding out what the founder did not hand over. What customers actually say when nobody is selling to them. Why the churned accounts left. What the last VP of sales thinks now.
Founders prepare for half one. They rehearse it for weeks.
Automation went after half two first. That was not an accident, and it was not about which half is harder.
Mechanism
Why the second half moved first
-
Step 01
Follow the cost, not the difficulty
Reading a deck was already cheap. An associate does it in ten minutes. Primary research was the six-figure line item on the diligence invoice. Products get built where the invoice is.
-
Step 02
Buyers with an existing budget moved first
Private equity firms already paid consultants to interview customers. They had a cost to replace, so they had a reason to buy. Seed funds never had that line item, which is why seed diligence looks much the same as it did.
-
Step 03
The work was repetitive, not subtle
Forty similar calls running off one question set is the exact shape automation handles well. Judging whether a founder can carry a market is not that shape, and it has not moved.
Three things change for the founder
None of them are about your deck.
The sample gets bigger
You used to give five references. Five is a curated number. It fits inside a human being’s calendar.
An agent does not have a calendar. The limit on how many customers get called is budget, not time. Budget just fell.
Your five best accounts still get called. So do the other thirty.
You stop choosing who speaks
A reference list is a selection. For most founders it is the last piece of control they hold in a diligence process.
When the caller works from a customer export, a review site or an expert network, that selection moves to them.
Nobody asks your permission before calling an account that churned last year.
It happens earlier
Primary research used to be confirmatory. You ran it after you already liked the company. It cost six figures, so nobody spent it on a maybe.
At the prices these companies quote, it can sit at screening instead.
That reorders the raise. The customer call can now land before the partner meeting, not after it.
What changed
The reference call, before and after
The old shape
Five calls, chosen by you
- Limited by a human calendar
- You picked who spoke
- Ran late, once a partner already liked you
- Cost meant it was reserved for real candidates
The shape now
Forty calls, chosen by them
- Limited by budget, and budget fell
- A customer export replaces your list
- Can sit at screening, before the meeting
- Cheap enough to run on a maybe
What the agents get, and what they miss
I have built the interview end of this. Not for diligence, for our own sales work at Qubit. So I will be specific about both sides.
What they get right is coverage. Call forty people and you see a distribution instead of an anecdote. A human on call thirty-eight is tired and starts leading the witness. An agent asks question four the same way every time. That consistency is real and it is worth something.
What they miss is the pause.
The “well, honestly” that arrives four seconds after the official answer. The thing a customer says once they think the call is over. A person hears hesitation and follows it somewhere unplanned. Agents follow a script tree.
They are getting better at this. They are not there.
Here is the version of that I ran into. We had an agent qualifying inbound conversations. It logged one founder as a clean fit on every field we asked about. Stage, sector, cheque size, timeline, all green.
A colleague listened to the recording later. The founder had said “we are figuring that out” twice, in a tone that meant something.
The transcript said yes. The call said no. Only one of those was in the CRM.
So the output is wide and shallow. Wide beats narrow when someone is screening a pipeline. It does not replace the one call a partner makes to a person they already trust.
Where this does not reach yet
I want to be careful about who this actually touches today.
The buyers here are private equity firms and later-stage funds. They are diligencing a company before writing a very large cheque. At that size, $50,000 of primary research is a rounding error.
Nobody is spending that to diligence a $500,000 seed cheque. The arithmetic does not work yet.
So if you are raising pre-seed or seed right now, your process probably has not changed. A partner still forms a view from a conversation, a warm intro and a quick look at your numbers.
What has changed is the direction of travel. The price of this work has fallen once already. Products aimed at PE budgets tend to walk down-market once the build cost is paid.
Plan for the version of diligence that reaches you in two rounds, not the one you got last time.
Coverage
What the second half of diligence now reaches
- Current customers The accounts you would have put forward, and the ones you would not have.
- Churned accounts Why they left, in their own words, without you in the room.
- Former employees What the last VP of sales says about the pipeline you inherited.
- Rival customers Buyers who looked at you and chose someone else, and what decided it.
- Sector experts Sourced through expert-network partnerships rather than your intro list.
What to do about it
This is not a reason to panic. It is a reason to change what you prepare.
Know your churn story before someone else does
Call the accounts that left. Ask why, and let the answer be uncomfortable.
You will not enjoy some of it. Better you hear it in April than a partner hears it in July.
Treat references as a distribution
Stop thinking about your five best customers. Think about what your median customer would say to a stranger.
If the median answer is weak, the fix sits in the product. It does not sit in the reference list.
Write down the question you are afraid of
Every founder has one. Say it out loud, then go find out what customers actually answer.
That is a two-week piece of work. It used to be optional.
Do not coach your customers
This is the obvious bad idea, so I will name it.
Briefing five references works because five is a small sample and a human is reading it. Briefing forty does not work, and it fails in a way you cannot see.
Coached answers cluster. They land on the same phrases and the same enthusiasm. Across forty transcripts that shows up as an anomaly, not as strength.
The people building these products are looking for exactly that pattern. It is the cheapest thing to detect.
The deck is still read by a person. That part has barely moved.
One vendor published an analysis of AI deck triage inside institutional funds. It puts working deployments below 12 percent. The basis is conversations with operators at roughly twenty funds.
That is anecdote-grade evidence, so hold it loosely. It matches what I see, which is not the same as it being true.
The half that moved is the half you were never in the room for.
Prepare for that half.