AI made the sourcing funnel faster. The capital-allocation call still gets a generic answer.
Funds have adopted AI across deal sourcing and diligence as a tooling edge. The decision that actually moves returns, the thesis, the price, the hold-or-exit judgement, is the one most likely to be handed back fluent, confident and average.
Request a Strategic Briefing →Private equity, venture and asset managers have moved AI into the core of dealmaking: sourcing, screening, diligence acceleration, portfolio monitoring. Treated as a data and tooling edge in a crowded market, that is sensible. The trouble starts when the same engine is pointed at the decision that decides the return: the investment thesis, the entry price, the hold-or-exit judgement. Ask a commodity AI to assess a thesis and it tends to flatter the one you already hold, in fluent, confident, generic language. That is the Mirror Principle at fund scale: generic reasoning in, generic conviction out, on decisions measured in hundreds of millions. Gildoni installs the Havruta Methodology (formerly the Think Partner Methodology) into how investment leaders reason with AI, so the machine argues the thesis rather than confirms it. This is not a sourcing platform. It is the reasoning discipline behind the allocation.
The industry has already wired AI into dealmaking
This is not a forecast. It is where the money already is.
Deloitte's 2025 survey of corporate and private equity dealmakers found generative AI no longer at the edge of the process but inside it, concentrated in the pre-sign stages where deals are sourced and screened. Erik Dilger, a managing director at Deloitte Financial Advisory Services, framed the mood as one of conviction: dealmakers, he said, "are confident in GenAI's potential to recast the look and feel of dealmaking, and are investing accordingly."
The regulators see the same shift across the wider market. IOSCO, surveying capital-markets participants for its 2025 report, found AI moving out of the lab and into live operation, with a meaningful share of firms already running it in production, making real decisions rather than sitting in pilot.
The adoption is not in question. The numbers are emphatic.
of corporate and private equity leaders now use generative AI in dealmaking, most heavily in target screening and due diligence (35% each).
of capital-markets firms have adopted AI, and 41% already run it in production rather than in pilot.
of wealth and asset managers have scaled AI to multiple use cases, yet 86% were surprised by the regulatory and compliance complexity.
The tooling edge is real. It is also the easy part.
Dealmakers are confident in GenAI's potential to recast the look and feel of dealmaking, and are investing accordingly.
Why a faster funnel can still make worse calls
If AI is already this deep in the process, why is the decision that matters no safer? Three reasons.
The investment has gone to the funnel, not to the judgement. The wins are visible and measurable: more targets screened, diligence rooms read faster, monitoring dashboards that refresh themselves. The capital-allocation call, the thesis and the price, sits above all of that, and it does not get faster just because the funnel does. It gets noisier.
AI flatters the thesis you already hold. Put a deal memo to a commodity model and it will reason towards the conclusion you have signalled, in confident prose, without ever arguing the other side. The diligence work that AI does best, the data-heavy reading, is precisely the part that consumes the most analyst time and reveals the least about whether the deal is mispriced. As one M&A lawyer put it of AI in diligence, it "can't mirror the core functions performed by an attorney", and the same holds for the partner who has to own the number.
And speed reads as conviction. A thesis assembled quickly, with AI confirming every leg of it, feels more defensible than one fought over in the room. It is not. It is the same generic read, arrived at sooner. The discipline of pricing under uncertainty is old; AI has not rewritten it, but it has made it easier to skip.
Sourcing faster is the easy part
Here is what the tooling story leaves out. Every platform, vendor and data provider now promises the same thing: more deals seen, diligence compressed, portfolios monitored in real time. Almost none of them touch the question that decides the fund's return: how an investment committee actually reasons through the thesis and the price.
That is the real gap, and it is not a data gap. Put the allocation question to a commodity AI and it hands back a fluent, confident memo that agrees with the conviction it can sense in the prompt. At the altitude where the cheques are largest, that is the Mirror Principle at its most expensive: if the reasoning going in was generic, the conviction coming out is generic, however polished the memo reads. A better sourcing engine can fill the top of the funnel. It cannot make the judgement at the bottom of it any good. That is a different discipline.
AI compresses the wide top of the funnel, sourcing, diligence, monitoring, which is where the tooling money goes. At the narrow point sits the capital-allocation call, and that is where a commodity AI returns a confident, generic read that flatters the thesis already held. The red is the distance between a faster funnel and a better decision.
What the Havruta Methodology installs at the investment committee
The Havruta Methodology is that discipline. It changes the default behaviour of the machine from agreeing with the thesis to arguing it, which is precisely what an allocation decision needs.
The Flip
The Flip puts the machine on the other side of the deal. Instead of confirming the thesis, it argues against it: where is the model optimistic, what would have to be true for this price to be wrong, what is the bear case the committee is not voicing. The thesis gets stress-tested before the holding period does the testing.
Ground Truth
Ground Truth keeps the reasoning anchored in this deal's real numbers, the actual cash flows, the specific cohort data, the genuine comparables, rather than the generic sector language an AI produces by default. A diligence read built on a plausible average is worse than no AI at all when the cheque is this size.
Decision Velocity
And Decision Velocity lets the committee reach a defensible position at the speed a competitive process demands, compressing the path from data room to conviction without surrendering the judgement, or the price, to the machine.
The fuller account of how all of this works is on the methodology page.
What this is not
This is not a deal-sourcing platform and it is not diligence software. It is not portfolio analytics, not a data room, not a CRM, and it is not AI training or general AI literacy. The tools that screen targets and read data rooms are a separate market. This is the thinking that turns their output into a defensible allocation.
It changes how the investment committee reasons about the call it already owns: the thesis, the entry price, the hold-or-exit judgement, the decision the fund answers for.
Where an investment team starts
The methodology is installed along a ladder, and an investment team enters at the rung that fits.
Most begin with the Eye-Opener Workshop, a half-day in which the team sees the shift on a live deal or thesis of its own.
An investment team embeds the practice through the Havruta programme, taking the discipline across the partners and the deal teams.
A single high-stakes question, a contested thesis, an entry price, a hold-or-exit call, can be worked through Advisory Havruta.
How a portfolio or division leader reasons with AI
For the leader allocating capital across a portfolio specifically, the role page takes the same discipline to the seat that owns the allocation across holdings. A Strategic Briefing is how to decide where to begin.
Go to the division-leader pageInvestment-committee questions about AI
Can AI improve our deal sourcing and diligence?
Yes, and most funds have already proved it. AI widens the top of the funnel, screens more targets and reads data rooms faster, which is why adoption across dealmaking is now near-universal. That is a genuine tooling edge. The caution is that the funnel is not the decision. Faster sourcing and faster diligence do not, on their own, make the thesis or the price any better. The judgement at the bottom of the funnel is a separate problem, and it is the one that decides the return.
Where does AI go wrong in a capital-allocation decision?
It agrees with you. Put a thesis to a commodity model and it reasons towards the conclusion it senses you already hold, in fluent, confident prose, without arguing the other side. This is the Mirror Principle: generic reasoning in, generic conviction out. On a screening task that is harmless. On a decision priced in hundreds of millions it is dangerous, because a confident memo that flatters the thesis feels like diligence when it is really an echo.
Can AI replace the judgement in due diligence?
No. AI excels at the data-heavy reading, the part that consumes most of the team's hours, and that is real value. But it cannot weigh the quality of a management team, the durability of a thesis, or whether a price is right. Practitioners reviewing AI in diligence make the same point: it cannot mirror the core judgement a professional brings. The reasoning that turns a fast diligence read into a defensible allocation is a discipline the committee has to practise, not a function the tool performs.
Is this a deal-sourcing platform or diligence software?
No. It is not a sourcing platform, not diligence software, not portfolio analytics, and it does not touch your data stack. Those address the funnel: more targets, faster reading, live monitoring. This addresses the thinking at the narrow point of the funnel, how an investment committee reasons through the thesis and the price so the conviction is genuinely theirs, anchored in the real numbers, and stress-tested before the holding period tests it for them.
How should an investment committee reason with AI well?
Stop asking the model to confirm the thesis and make it argue against it: where is the case optimistic, what would have to be true for the price to be wrong, what is the bear view no one is voicing. Anchor every read in this deal's real cash flows and comparables, not generic sector language. Then decide at the speed a competitive process demands without handing the price to the machine. The tools fill the funnel; this is how you reason at the point the cheque is signed.
Where should we start?
With a Strategic Briefing, or with the Eye-Opener Workshop, where an investment team sees the difference between instructing AI and reasoning with it on a live thesis of its own. From there the path depends on whether you are embedding the practice across the deal teams or working a single contested allocation, an entry price, or a hold-or-exit call.