What Questions Should You Ask an AI Consulting Firm Before You Sign? A 12-Question Due Diligence Framework

TLDR: AI vendor due diligence is what separates enterprises that execute from those that pay for strategy documents and stalled pilots. Most companies evaluate consulting firms through demos and proof-of-concept engagements, neither of which reveals the signals that predict delivery success. This guide provides 12 questions, organized across delivery, governance, and exit dimensions, that no credible firm should struggle to answer.

Best For: COOs, VP Operations, and Chief Transformation Officers at mid-to-large enterprises who are evaluating AI consulting firms or implementation partners before committing to a multi-month or multi-year engagement.

AI vendor due diligence is the structured process of evaluating an AI consulting firm or implementation partner against production-grade delivery standards before contract signing. Unlike a traditional software vendor selection, where feature comparison provides a meaningful signal, AI consulting due diligence focuses on the hardest question to answer from a demo: can this firm actually ship something that works in your operating environment, sustain its performance after go-live, and leave your team capable of running it without them? The answer is not visible in a case study slide or a reference webinar. It is visible in how the firm answers 12 specific questions.

Why Most Enterprises Skip AI Vendor Due Diligence

Most enterprises skip AI vendor due diligence because the sales process feels like evaluation. Multiple workshops, proof-of-concept engagements, reference customer webinars, and pricing negotiations create the impression that thorough vetting has already occurred. In practice, none of these activities address the questions that predict whether an implementation will reach production and sustain value over 18 months.

The research is hard to ignore. RAND Corporation's 2025 analysis found that 80.3% of AI projects fail to deliver their intended business value. MIT NANDA research from August 2025 puts it more starkly: 95% of enterprise AI pilots deliver no measurable P&L impact, and only 5% of custom tools ever reach production. The technology is rarely the problem. Most of these failures trace back to partner selection, scope definition, and contract structure — things a proper pre-contract evaluation would have surfaced.

The Production Track Record Problem

The most important indicator of a partner's ability to deliver is a verifiable production track record. Yet most enterprises evaluate AI consulting firms through case study slides and reference webinars where the vendor controls the narrative. These formats are designed to demonstrate competence, not to reveal failure modes, recovery timelines, or the post-go-live support model. Production deployments that survived 18 months of organizational change, data drift, and system integration pressure are a fundamentally different signal than a successful 90-day pilot, and they are almost never discussed in a standard sales process.

When Consulting Deliverables Are Not the Same as Delivery

A structural problem in the AI consulting market is that the deliverable of many engagements is documentation: strategy reports, roadmaps, use case prioritization frameworks, and technical architectures. These outputs are real and sometimes valuable, but they are not AI deployments. Consulting engagements that produce strategy documents without transferring operational capability create a dependency cycle: the client cannot sustain or extend what the consultant built, leading to either ongoing consulting spend or project abandonment. Your evaluation process must explicitly probe whether the partner's business model is oriented toward delivery and capability transfer, or toward ongoing engagement extension.

How the Evaluation Standard Has Shifted

For most of the 2020s, enterprise AI consulting evaluation looked a lot like software vendor selection: check features, compare pricing, evaluate support terms. That worked when AI meant point tools with defined outputs. It breaks down when the engagement involves cross-functional workflow changes, data integration, and an operating model that needs to keep running after the consultants leave. The most rigorous buyers now treat AI partner evaluation the way they'd treat hiring an engineering partner for a complex infrastructure project: production references, defined accountability, and contract terms that protect you if delivery falls short. Gartner has estimated that roughly 85% of AI projects fail to reach production or deliver measurable business outcomes. Getting to the other 15% starts before you sign.

What AI Vendor Due Diligence Actually Covers

AI vendor due diligence covers four domains: delivery evidence, data and governance practices, contractual protections, and the post-engagement relationship structure. A well-designed evaluation process surfaces the firm's actual production track record rather than its marketing claims; its data handling and compliance posture rather than a general privacy policy; what the client will own after the engagement ends; and who specifically will be doing the work.

The global AI system integration and consulting market reached $11 billion in 2025 and is projected to reach $14 billion in 2026, with every major AI vendor now building formal partnerships with consulting firms to help enterprises deploy at scale. As the market has expanded, the gap between credible delivery partners and pitch-oriented consultancies has widened. Structuring your evaluation against all four domains ensures you are assessing delivery capacity, not sales capability.

AI Vendor Due Diligence vs. Software Vendor Selection

The distinction matters because the evaluation criteria are genuinely different. Software vendor selection centers on the product: does it do what we need, does it integrate with our stack, and is the vendor financially stable? The product performs the same way regardless of who implements it, which means evaluation risk is primarily about fit. AI consulting due diligence centers on the team and the methodology: does this firm have the specific production experience, organizational design, and accountability structure to execute in our environment? The same methodology applied by two different teams to the same use case will produce materially different outcomes. That makes the evaluation fundamentally about people and process, not features.

The 12 Questions to Ask Before You Sign

These 12 questions cover delivery, data and governance, and contract and exit dimensions. A credible firm answers each one substantively. When a firm struggles, deflects, or gives you a generalized non-answer, that's evaluation data too — often the most useful kind.

Delivery Evidence Questions (1 to 4)

1. Can you name three production deployments in our industry, and will you arrange direct reference conversations with those clients?

A firm with genuine production experience in your industry names clients without hesitation and facilitates direct peer conversations, not facilitated webinars. During those conversations, ask the reference clients the same questions you asked the vendor and compare the answers.

2. What percentage of your engagements reach sustained production within 12 months?

Most firms will not have tracked this metric precisely. A credible partner can give an honest range and explain what prevents some engagements from reaching that milestone. A firm that quotes a suspiciously high figure without supporting detail has almost certainly not measured it.

3. Who specifically will be on our team, and what is their individual production track record?

The senior partner who leads the sales process often does not lead the implementation. Ask for the CVs and production project histories of the specific individuals allocated to your account, and confirm what percentage of their time will be dedicated to your engagement versus shared across other clients.

4. What failure case in the last 18 months taught your team the most, and what changed as a result?

A thoughtful, specific answer about what went wrong, what the firm did to recover it, and what changed in their delivery process afterward is one of the highest-quality evaluation signals available. AI and ML monitoring usage in production climbed to 54% in 2025, up from 42% in 2024, reflecting that mature delivery teams are increasingly systematic about tracking what breaks. Firms operating at that maturity level have failure cases, and they learn from them.

Data and Governance Questions (5 to 8)

5. Walk me through exactly what happens with our data during the engagement.

Ask specifically where your data is stored, whether it is used for model training or fine-tuning, who at the firm has access to it, and what happens to it after the engagement ends. Rigorous AI vendor due diligence on data practices requires tangible documentation, not a reference to a general privacy policy. For enterprises in regulated industries, this question also surfaces whether the partner has operated within your sector's specific compliance framework before.

6. What governance and compliance review process do you follow before recommending that a use case go live?

A credible partner has a defined pre-deployment checklist that includes compliance review, testing under production load, and a go or no-go decision process that involves your operations leadership, not just the technical team. Enterprise AI vendor evaluation in regulated industries requires proof of governance controls, not assurances of general compliance. Ask who signs off on a deployment decision and what criteria must be met.

7. What does your data readiness process look like, and what happens if our data turns out not to support the recommended use case?

The four most common AI implementation failure modes are inadequate data foundations, absent governance, misaligned KPIs, and organizational resistance. A firm that does not conduct a data readiness assessment before committing to a use case recommendation is accepting scope risk on your behalf. Ask how they handle mid-engagement discovery that required data is unavailable or unusable.

8. What AI infrastructure or platforms do you recommend, and what are the exit terms if we want to change platforms in 18 months?

This question surfaces vendor alignment, lock-in posture, and commercial incentives. Firms with platform partnerships have economic incentives to recommend specific technology stacks. Vendor lock-in has become one of the top concerns in enterprise AI procurement for 2026, with sophisticated buyers asking for written exit terms, including data portability for prompts, embeddings, evaluation sets, and audit logs, before signing.

Contract and Exit Questions (9 to 12)

9. What does the engagement model look like after go-live?

Many AI consulting engagements are scoped to deliver a working system, not to sustain it. After the engagement ends, monitoring, retraining, and performance management often fall to the client's team without documentation, training, or support structure. Ask specifically whether post-go-live support is in scope, what the handoff process looks like, and what client team readiness is expected by go-live.

10. What does the standard exit deliverable include, and what do we own?

Your organization should own the models, prompts, code, architecture documentation, evaluation sets, and operational runbooks at the end of the engagement. Ask the firm to describe the standard exit deliverable package and confirm that ownership transfers to you in writing.

11. What is your approach to change management and employee adoption?

AI implementations that succeed technically but fail organizationally deliver no sustainable value. A serious delivery partner has a defined change management approach that includes manager enablement, frontline training, adoption tracking, and a feedback loop for surfacing resistance. If the firm's answer is that change management is the client's responsibility, you have identified a scope gap that predicts adoption failure.

12. What happens if, after six months, the engagement is not delivering the outcomes we agreed on?

Ask whether the contract includes outcome-linked milestones, what the escalation and remediation process looks like if milestones are missed, and whether there is a termination clause that protects the enterprise if delivery falls materially short of agreed benchmarks. Gartner predicts that more than 40% of agentic AI projects will be canceled by end of 2027, with unclear business value as the leading cause. Contracting for accountability at the outset is the most cost-effective way to ensure ongoing alignment.

How to Interpret the Answers

Structuring your evaluation responses across three categories helps you reach a decision without waiting for certainty you will never have.

Green Flags That Predict Delivery Success

Named production references in your industry who agree to direct peer conversations. A specific, reflective answer to the failure case question with documented process changes. A defined post-go-live support model with clear responsibility allocation. A contract structure that includes outcome milestones and client-owned data portability. Deloitte's 2026 State of AI in Enterprise report, which surveyed more than 3,000 C-suite leaders, found that the organizations achieving the most AI value had deeply integrated delivery partners rather than vendor relationships managed at arm's length.

Yellow Flags That Require Follow-Up Before Signing

References available but only through facilitated webinars. Specific team members not yet named at the proposal stage. Change management described as a shared responsibility without a firm-side methodology. Data governance described in general terms without written documentation. None of these conditions are disqualifying on their own, but each requires explicit resolution before you sign.

Red Flags That Should End the Conversation

Inability to name production deployments in a comparable industry or use case. Vague data governance practices without written documentation. An exit deliverable described as "your team will take over" without supporting structure. Resistance to including outcome milestones or data portability terms in the contract. For a comprehensive view of warning signs across the full engagement lifecycle, the AI consulting red flags guide covers patterns that appear before signing and during delivery. The AI vendor selection scorecard provides a structured format for comparing multiple firms simultaneously.

What Skeptics Say (And What to Say Back)

"We've done reference checks before. They don't tell us much." Reference checks conducted through vendor-facilitated webinars reveal almost nothing. The value is in direct, unstructured peer conversations where you can ask what went wrong, not what went right. Structuring those conversations around the same 12 questions you asked the vendor creates a comparison data set that generic reference calls cannot produce.

"We don't have the expertise to evaluate the technical answers to these questions." You do not need technical expertise to run this evaluation. Your job is to assess whether the firm answers each question substantively and whether the answers are internally consistent across the sales process, the reference calls, and the contract. Credible partners give complete, specific answers. Partners who are not ready for enterprise delivery deflect, generalize, or redirect. That pattern is visible to any experienced operations leader.

"The firm has a strong brand. Isn't that sufficient?" Brand credibility does not predict delivery performance on any specific engagement. The global AI consulting market is projected at $14 billion in 2026 and every major firm has made substantial AI capability claims. The team you actually get, the methodology they apply, and the contractual structure you sign are the variables that determine outcomes. This evaluation process exists precisely because brand is not a delivery guarantee.

Before committing to a consulting partnership, also review how to evaluate AI vendors beyond the initial demo for guidance on structuring proof-of-concept evaluations, and what an AI vendor RFP should cover if you are running a formal procurement process.

Frequently Asked Questions

What is AI vendor due diligence?

AI vendor due diligence is the structured process of evaluating an AI consulting firm or implementation partner before contract signing, focused on production track record, data governance practices, contractual protections, and post-engagement capability transfer. It goes beyond feature comparison to assess whether a firm can deliver production-grade outcomes in your specific operating environment.

Why do most enterprises skip AI vendor due diligence?

Most enterprises skip AI vendor due diligence because the sales process creates the impression that evaluation is already complete. Workshops, demos, and pricing discussions feel thorough but do not surface the signals that predict delivery success: production track record, data governance posture, and post-go-live accountability structure. According to RAND Corporation's 2025 analysis, 80.3% of AI projects still fail to deliver their intended business value.

What is the single most important question to ask an AI consulting firm?

The single most important question is: Can you name three production deployments in our industry, and will you arrange direct reference conversations with those clients? A firm with genuine delivery experience in your sector names clients without hesitation and facilitates unstructured peer conversations. Vague references to adjacent industries or delays in arranging calls are the most reliable early signal of limited relevant production experience.

How do you evaluate an AI consulting firm's production track record?

Evaluate production track record through direct reference conversations with named clients in your industry, not vendor-facilitated webinars. Ask references what percentage of the firm's commitments were met on time, how the firm handled scope problems mid-engagement, and whether they would retain the firm for a follow-on project. Cross-reference the answers with what the vendor told you during the sales process.

What does a green flag look like in AI vendor evaluation?

Green flags in AI vendor due diligence include named production references who agree to direct conversations, a specific and reflective answer to the failure case question, a defined post-go-live support model, client-owned data portability written into the contract, and outcome-linked milestones. These signals indicate a partner oriented toward delivery accountability rather than engagement extension.

What are the red flags when evaluating an AI consulting firm?

Red flags include inability to name production deployments in a comparable industry, vague data governance practices without written documentation, an exit deliverable described as "the client takes over" without structure, and resistance to outcome milestones or termination clauses. For a comprehensive view of warning signs across the engagement lifecycle, review the AI consulting firm red flags guide.

What should be included in an AI consulting firm's exit deliverable?

The standard exit deliverable should include all models, prompts, code repositories, architecture documentation, evaluation sets, and operational runbooks, with documented ownership transferring to the client. Ask the firm to describe their standard exit package in writing during the proposal stage, not after the contract is signed. Firms whose business model depends on ongoing retainer relationships to sustain what they built will often resist this question.

How do you evaluate an AI consulting firm's data governance practices?

Evaluate data governance by asking specifically where your data is stored, whether it is used for model training, who at the firm has access to it, and what happens to it after the engagement ends. Request written documentation, not a verbal reference to a privacy policy. For regulated industries, confirm the firm has operated within your sector's specific compliance framework before and can provide evidence of that experience.

What is the best way to conduct reference checks for an AI consulting firm?

The best reference checks are direct, unstructured peer conversations structured around the same 12 questions you asked the vendor. Avoid vendor-facilitated webinars where the reference client is pre-selected and the conversation is moderated. Ask what went wrong during the engagement, how the firm responded, and whether the reference would hire them again for a similar scope.

How do you know if an AI consulting firm will build internal capability or create dependency?

Ask directly what the client's team will be able to do independently after the engagement ends and what the standard exit deliverable includes. Firms oriented toward capability transfer describe specific training programs, documentation standards, and handoff milestones. Firms oriented toward ongoing engagement will describe the post-go-live relationship in terms of continued retainer work rather than client independence.

What should an AI vendor contract include?

An AI vendor contract should include outcome-linked milestones, a defined escalation and remediation process for missed milestones, a termination clause with specific triggers, full data portability terms, and a clear description of what the client owns at engagement end. Contracts that describe deliverables only in terms of effort (hours, sprints, workshops) rather than outcomes transfer engagement risk entirely to the client.

How long should AI vendor due diligence take?

For a significant multi-month engagement, AI vendor due diligence should take two to four weeks, including at minimum three direct reference conversations, a review of written governance documentation, and negotiation of contract terms. Firms that pressure buyers to compress this timeline to close within a quarter are often managing their own revenue schedules rather than your implementation risk.

What is the difference between evaluating an AI consulting firm and an AI software vendor?

Software vendor selection centers on the product: does it perform as specified, does it integrate with your stack, and is the vendor financially stable? AI consulting evaluation centers on the team and methodology: does this specific team have the production experience, governance posture, and accountability structure to execute in your environment? The same methodology applied by two different teams to the same use case will produce materially different outcomes.

What happens when an AI consulting engagement does not deliver?

Without outcome milestones and a defined remediation process in the contract, the client has limited recourse when delivery falls short. Gartner predicts more than 40% of agentic AI projects will be canceled by end of 2027, primarily due to unclear business value and absent accountability structures. The most cost-effective protection is outcome-linked contractual terms negotiated before signing.

How do you evaluate a consulting firm's change management capability?

Ask the firm to describe their standard change management methodology, including how they identify resistance, train frontline employees, enable managers, and track adoption against defined milestones. A firm with a defined approach can describe it in specific terms. If the firm's answer is that change management is the client's responsibility, you have identified a scope gap that predicts adoption failure in most enterprise deployments.

Should you run a formal RFP for an AI consulting firm?

A formal RFP is appropriate when you are evaluating four or more firms simultaneously and need a structured comparison framework. For smaller shortlists of two or three firms, the 12-question evaluation process described in this guide is faster and produces more useful differentiation data than a standard RFP format. What an AI vendor RFP should cover provides guidance on structuring a formal procurement process when one is warranted.