AI in the Investment Committee: 12 Questions GPs Should Ask Before Approving a Deal

What AI questions should an investment committee ask before approving a deal? The twelve questions in this article share one property: each forces the deal team to quantify something a traditional technology audit never measured. How much of the target's thesis does AI disrupt, how much of its margin is exposed to AI-driven cost and revenue shifts, and how technically defensible is the product against AI-native substitutes. The AI investment committee conversation has moved decisively beyond the classic IT-diligence checklist of infrastructure, cybersecurity posture, and licence compliance. Those checks still matter, but they no longer answer the question that determines whether a multiple is defensible over a hold period.

The scale of the shift explains why. industry analysis estimates that AI puts $4.7 trillion of corporate profits at stake between 2025 and 2035, structurally transforming 71% of sectors, compared with 41% during the internet era, and reaching industries the internet largely bypassed such as healthcare delivery, industrial manufacturing, and professional services. For PE investment committee members and deal captains, this means almost no target sector sits outside the blast radius.

Importantly, the disruption is uneven, and the IC's first task is to locate the target on industry analysis's map. Only a small slice of that $4.7 trillion total-just $0.3 trillion-sits in industry analysis's 'revolution' cluster, where the core service itself shifts to AI, as in customer support and test prep. A far larger $1.3 trillion sits in the 'augmentation' cluster, where the business model survives but the race for AI-powered productivity gains decides who keeps margin, and 'rewired' sectors such as enterprise software face a wide-open competitive race. In other words, most targets face augmentation economics, not extinction, but in augmentation the laggards lose share to competitors using the same publicly available tools.

• From technical audit to commercial question: the IC should ask what AI does to revenue quality, cost structure, and defensibility, not only whether the IT estate is sound.
• From binary to clustered: diligence should place the target in a disruption cluster and size the exposure accordingly, because a revolution-sector target and an augmentation-sector target justify completely different underwriting.
• From static snapshot to hold-period view: AI capability is compounding, so the question is not where the target stands today but what its thesis looks like in year three and year five of the hold.

The twelve questions below are grouped into five clusters: thesis impact and defensibility, financial exposure, data assets and technical dependencies, management credibility, and open questions for the memo. For a fuller pre-deal framework, see the AI due diligence checklist for private equity value creation.

Q1: Does AI strengthen or weaken the investment thesis?

This is the anchoring question, and it should be answered before any others. AI can strengthen a thesis by widening a moat, compressing service costs, or opening an upsell, and it can weaken one by eroding a pricing premium, commoditising a workflow, or arming a competitor. The IC needs an explicit, quantified directional call, because 'AI is a tailwind' stated without numbers is not diligence. Evidence to demand: a revenue-line-by-revenue-line map of AI impact across the hold, customer interviews on willingness to pay for AI features, competitor adoption benchmarks, and the target's own churn and win-rate trends since deploying AI. Red flag: the thesis memo mentions AI only in the value-creation section and never in the risk section, which usually means the analysis has not actually been done.

Q2: Which of the target's revenues are most AI-exposed?

industry analysis is explicit that in sectors such as enterprise software there are no predetermined winners: a leader with deep customer integrations, high switching costs, and defensible data moats has a structural advantage only if it treats AI as a board-level priority and ships features customers will pay for. The IC should test whether the target's core workflow could be rebuilt from scratch by an AI-native competitor or collapsed into a frontier model. Evidence to demand: switching-cost data by customer cohort, net revenue retention over at least three years, an analysis of which workflow steps a general-purpose model already performs adequately, and the product roadmap's response. Red flag: the core workflow is a thin wrapper over an off-the-shelf model, with no proprietary data, integration depth, or domain logic that a substitute could not replicate within a hold period.

Q12: What would invalidate the investment thesis over the hold period?

Every IC memo carries risks, but AI moves fast enough that the committee should require a written invalidation test, not a generic list of mitigants. Name the two or three observable events that would falsify the thesis: a named competitor shipping an AI-native equivalent, model capability crossing a threshold that automates the target's core deliverable, or customer concentration shifting to buyers who demand AI features at no premium. Evidence to demand: trigger points with owners and review dates, a monitoring plan tied to public signals, and pre-agreed portfolio actions if a trigger fires. Red flag: the invalidation discussion is deferred to a 100-day plan rather than resolved before approval, leaving the committee to approve a thesis it has not stress-tested.

Q2: Which of the target's revenues are most AI-exposed?

Not all revenue carries equal AI risk. Recurring revenue tied to a workflow AI can automate is far more exposed than revenue tied to regulated processes, physical assets, or proprietary data access. The IC should require revenue to be segmented by AI exposure, because the answer changes both the multiple paid and the covenants negotiated. Evidence to demand: a contract-level view of which revenue lines depend on billable hours, per-seat licences, or information asymmetry, all of which AI compresses; customer-level concentration within exposed lines; and pricing history showing whether AI features have been monetised or given away. Red flag: management cannot segment revenue by AI exposure at all, which signals the leadership team has never run this analysis on its own business.

Q3: Can AI materially change the cost base, and who captures the value?

industry analysis's Technology Report finds that leading companies scaling AI across core workflows have delivered 10% to 25% EBITDA gains over the last two years, while most companies remain stuck in experimentation mode with modest productivity gains. That gap is exactly the point for the IC: the upside case assumes the target becomes a leader, and leadership requires budgeted operational expenditure, not aspiration. The committee must also ask who captures the value, because in competitive markets productivity gains flow to customers as price cuts rather than to shareholders as margin. Evidence to demand: a bottom-up cost-model showing which functions AI changes, the run-rate cost of the tooling and talent required, and the competitive dynamics that determine whether savings are retained. Red flag: a forecast assumes leader-level EBITDA improvement while the model budgets little or none of the AI opex needed to achieve it. For a deeper framework, see the analysis of AI value creation diligence and EBITDA uplift EBITDA uplift diligence.

Q9: What AI-related capex and opex is required across the hold?

AI spending is recurring, not a one-off project cost. industry research's 2026 survey finds that roughly one in five respondents report that AI-related operating costs, including token costs, constrained their AI use, even as per-token prices fell, because consumption grew faster than unit costs declined. Evidence to demand: a hold-period cost curve covering compute, model licences, data infrastructure, and AI talent; the split between capex and opex; and sensitivity cases if usage grows with scale. Red flag: the financial model treats AI as a single transformation budget line in year one with no recurring run-rate, which historically understates cost by the widest margin.

Q5: What differentiated data does the target own?

Proprietary data is the most defensible asset in an AI market, and the IC should treat it with the same rigour as customer contracts. The question is not whether the target has data but whether it has exclusive, legally clean, workflow-embedded data that improves model performance in a way competitors cannot buy. Evidence to demand: a data provenance audit covering sources, consent, and usage rights; volume and freshness metrics; evidence that data quality translates into product performance; and confirmation of contractual rights to use customer data for model training. Red flag: the target's training data was scraped or licensed without clear rights, or its 'proprietary dataset' is largely derivable from public sources. For the evaluation framework, see the guide to AI moat due diligence.

Q8: What third-party model dependencies exist?

Most targets depend on external foundation models, hyperscaler infrastructure, or specialist AI vendors, and each dependency carries pricing, availability, and strategic risk. The IC should require a dependency map before approval, not discover it during a post-close integration. Evidence to demand: the full vendor stack with contract terms, pricing mechanisms, and exit provisions; an assessment of switching costs between model providers; in-house capability for evaluation and fallback; and any concentration exposure to a single provider. Red flag: the entire product runs on one external foundation model with no contractual protection against repricing and no tested alternative, a dependency explored in depth in the AI infrastructure exposure framework.

Q6: What AI assumptions are already embedded in management forecasts?

By the time a deal reaches IC, management's plan may quietly assume AI-driven revenue uplift, cost savings, or both, without labelling them as AI assumptions. The committee cannot price what it cannot see. Evidence to demand: a line-item reconciliation separating base-case performance from AI-attributed uplift; the timing assumptions behind each; and whether pricing, retention, or productivity assumptions changed after the target began deploying AI. Red flag: forecast margin expansion that management attributes to 'efficiency initiatives' but cannot decompose when asked, which usually means speculative AI upside is baked into the numbers.

Q7: How credible is management's AI roadmap?

Roadmaps are cheap to write and expensive to execute. The IC should assess management's record, not its slideware. Evidence to demand: documentation of previous technology rollouts and their realised versus promised results; named owners and milestones for each AI initiative; hiring plans and the actual technical bench in place; and evidence of governance, including how leadership reviews AI investment decisions. Red flag: AI revenue projections with no named executive owner, no dated milestones, and no historical precedent of the team shipping a comparable transformation, a credibility pattern covered in the AI transformation playbook for deal teams and portfolio companies.

Q10: What evidence supports the AI value-creation case?

industry analysis's executive survey found that while roughly 80% of generative AI use cases at adopting companies met or exceeded expectations, only about 23% of all respondents could tie AI initiatives to new revenue or lower costs, and about a third of disappointed companies said pilots worked but did not scale. The IC should therefore ask for measured results, not testimonials. Evidence to demand: customer-level proof that AI features are contracted and paid for; before-and-after productivity data from internal deployments; and reference calls with customers using the AI capabilities at production volume. Red flag: the value-creation case rests on a pilot with a handful of friendly customers, no pricing attached, and no scaling plan, a gap pattern explored in the analysis of high-conviction investment decisions high-conviction AI evidence.

Q11: Which material questions remain unanswered at IC?

A disciplined IC memo does not pretend to have every answer. It separates what is verified, what is estimated, and what is unknown, and it tells the committee exactly where the evidence stops. Deloitte's 2026 M&A pulse study, which found 90% of organizations now using GenAI in M&A processes, also stresses that because AI outputs are not highly visible, mechanisms to verify them are essential, and that human review remains the top requirement for high-stakes GenAI use. That principle applies to the memo itself: AI-generated analysis should carry source citations the committee can check. AI findings should surface in three places: thesis exposures that adjust the return model, risks with named mitigants, and open questions the committee must actively engage with rather than wave through.

Q11 belongs at the front of the committee's agenda, not the end. Unanswered material questions should each carry an owner, a date, and a pre-agreed condition, such as a completion mechanism, an escrow, or a post-closing workplan, so that approving the deal is an informed judgement about known gaps rather than an assumption that diligence was complete.

Answering twelve questions across thousands of data-room documents is a coverage problem before it is a judgement problem, and this is where Plausity supports IC preparation. Data Room Ingestion connects to virtual data rooms and processes PDFs, spreadsheets, contracts, and financial models within minutes, so analysis starts on full corpora rather than sampled folders. The AI-Analysis Engine reads, cross-references, and reasons over those documents to produce source-grounded diligence analysis, and Risk Radar evaluates findings by materiality, financial impact, legal exposure, and deal relevance to surface the anomalies a committee actually needs to see. For IC-ready synthesis, Report Builder drafts structured, investor-ready deliverables with full source traceability, which is what turns AI output into something an IC memo can cite. The firm's AI for M&A workspace ties these capabilities together for deal teams.

• IC-ready synthesis: consolidated findings with citations the committee can verify, structured for memo sections rather than free-form chat.
• AI Impact DD: analysis oriented to the questions in this article, including thesis exposure, revenue exposure, and cost-base impact.
• Evidence Gap Detection: identification of where the data room does not answer a material question, so open questions are surfaced before the meeting rather than during it.

One boundary deserves emphasis: Plausity does not make the investment decision. The platform's role is coverage and traceability, ensuring every claim in the memo traces back to a data-room source the committee can open, while the judgement on whether to deploy capital remains with the investment committee itself. Deal captains preparing their next IC discussion can arrange a walkthrough of the Plausity workspace with the team.

Plausity is an AI-native due diligence and deal intelligence workspace that helps M&A advisory firms, VC and PE funds, corporate development teams and investment-banking teams structure evidence, findings and questions across a data room. Plausity supports evidence extraction, source grounding, findings management and IC preparation — it does not replace human analysts, advisers or investment professionals, does not provide legal, tax, audit, regulatory or investment advice, and does not make autonomous investment decisions. All findings require human review. Built for today's investment and deal teams. Trusted by >200 firms.

To explore the underlying capabilities, see the Plausity AI analysis engine, the findings and risk intelligence and evidence gap detection product pages, and the IC memo product page. For team-level workflows, see how VC and PE funds and M&A advisory firms use Plausity across live deals, and how AI Impact due diligence and value creation workstreams support the analysis.

1. deloitte.com