All insights
EN · AI Strategy & Transformation

How to evaluate an AI consulting ROI business case before hiring

Test the baseline, attribution, adoption, lifecycle cost, scenarios, and measurement plan before accepting an AI provider's ROI case.

An executive analysis compares the baseline, total investment, and verifiable benefits before approving an AI proposal.
The AI Value Case Proof-8 tests evidence, lifecycle cost, uncertainty, and benefit realization before signing. · Generated with OpenAI

Do not approve an AI engagement because the proposal shows a high ROI, a short payback period, or thousands of hours saved. A defensible business case connects an observable baseline to a specific operating change, separates released capacity from realized cash, includes full lifecycle costs, makes adoption and uncertainty explicit, and names who will measure the result after launch. If the consulting firm cannot make that chain auditable, treat the return as a hypothesis—not a commitment.

Require three comparable scenarios—conservative, base, and expansion—and a decision that remains rational when adoption, volume, model cost, or timing deteriorates. The purpose is not to prove that AI always wins. It is to identify the conditions under which it creates value, the evidence that reduces uncertainty, and an economic stop point.

The AI Value Case Proof-8: 32 points before signing

Score each dimension from zero to four: zero is absent or contradictory; one is unsupported; two is partial; three is sufficient evidence with controllable gaps; four is reproducible evidence with an owner and update plan. As an illustrative threshold, require 24 of 32, no zero, and at least three for baseline, lifecycle cost, and measurement. Set weights before seeing finalist ROI figures.

  • Decision and outcome — problem, population, unit of value, horizon, and genuine alternative.
  • Baseline — current performance, variability, period, source, data quality, and owner.
  • Counterfactual and attribution — what happens without the project and what change AI can claim.
  • Benefit mechanics — how technical improvement becomes capacity, avoided cost, revenue, risk, or working capital.
  • Adoption and change — process owner, training, incentives, exceptions, and usage ramp.
  • Lifecycle cost — discovery, data, integration, evaluation, security, cloud, models, support, change, and exit.
  • Uncertainty — ranges, dependencies, sensitivity, downside case, and stop point.
  • Measurement and governance — metric, source, frequency, owner, review, decision, and learning.

Start with the baseline and counterfactual

A percentage improvement without a denominator is not an estimate. Record current volume, time, error, conversion, cost, or loss, including seasonality and team variation. Define the counterfactual: natural trend, already-funded automation, planned hiring, non-AI redesign, or status quo. Comparing the future with an unusually weak historical period inflates value.

Specify attribution before launch. Use a staggered rollout, comparable group, controlled experiment, or pre-defined time-series analysis when feasible. Otherwise disclose limitations and triangulate evidence. A post-hoc dashboard produced only by the paid provider is not an evaluation design.

Do not turn every saved hour into cash

Released time can create more throughput, shorter queues, higher-value work, or genuine cost avoidance. Only the last category automatically changes cash. To monetize capacity, show eligible demand, the bottleneck, a headcount or budget decision, and the internal owner who will execute it. Do not count the same hours as labor savings, incremental revenue, and headcount reduction.

Incremental revenue requires eligible volume, uplift, margin, cannibalization, operating capacity, and attribution. Risk reduction requires exposure, frequency, severity, and control effectiveness. Qualitative benefits can remain important, but keep them separate from monetized flows and define observable indicators.

Include full cost and the time value of money

Include data access and cleanup, integrations, continuous evaluation, observability, security, privacy, model consumption, licenses, infrastructure, human review, training, change management, support, rework, upgrades, and transition. Apply the Finance-approved discount rate and keep nominal or real cash flows internally consistent.

Separate sunk, incremental, and allocated costs. Show cash flow by period, not only a total. Payback, net present value, and ROI answer different questions. OMB A-94 and the Green Book provide official structures for alternatives, costs, benefits, discounting, and uncertainty; neither supports choosing from one headline ratio.

Five artifacts that should travel with the proposal

  • One-page value case — decision, reference option, horizon, scenarios, and stop point.
  • Assumption and evidence register — value, source, date, owner, confidence, and validation plan.
  • Benefits map — outcome, metric, mechanism, dependency, and realization owner.
  • Lifecycle cost model — volumes, unit prices, ranges, internal work, and third parties.
  • Measurement plan — instrumentation, frozen baseline, evaluation design, cadence, and decision forum.

Run an adversarial business-case defense

In a 90-minute session, give every finalist the same shock: adoption falls by half, volume doubles, quality requires more human review, model pricing rises, and launch slips one quarter. Ask the named team to recalculate scenario, architecture, operating model, and recommendation. Observe traceability, speed, transparency, and willingness to recommend stopping.

A strong team separates controllable assumptions from external dependencies, shows which evidence changes the decision, and versions the model. A weak team protects the original number, changes definitions, or removes costs. The session tests commercial, financial, technical, and operating maturity together.

Business-case red flags

  • Guaranteed ROI or payback without a range, assumptions, horizon, and alternative.
  • Percentage improvement without a baseline, denominator, source, or period.
  • Double counting across time saved, headcount, revenue, and risk.
  • Savings depend on adoption or process change with no funded owner.
  • Data, integration, evaluation, security, operation, or exit costs are omitted.
  • The provider exclusively controls the metric that determines its compensation.
  • There is no stop point when evidence deteriorates.

Connect the case to sourcing and payment

Use https://makinai.co/insights/en/practical-framework-prioritize-ai-use-cases to compare investments, https://makinai.co/insights/en/how-much-ai-consulting-services-cost to normalize cost, https://makinai.co/insights/en/fixed-price-time-materials-outcome-based-ai-project to choose commercial structure, and https://makinai.co/insights/en/ai-project-acceptance-criteria-payment-milestones to connect evidence to payments. Do not pay solely for a gameable proxy; combine deliverables, quality, adoption, and shared-control outcomes.

When to involve MAKINAI

MAKINAI can reconstruct the value case, normalize proposals, design measurement, and structure a first phase that buys evidence before scale. Explore https://makinai.co/services/en/ai-strategy-transformation-consulting. The framework is not financial, accounting, tax, or legal advice; it makes assumptions visible to the qualified people making those decisions.

Sources and references

  1. HM Treasury — The Green Book 2026 · HM Treasury

    Provides a current framework for comparing options through objectives, costs, benefits, risks, uncertainty, monitoring, and evaluation rather than a headline ratio.

    2026-09-08
  2. U.S. GAO — Cost Estimating and Assessment Guide · U.S. Government Accountability Office

    Defines practices for reliable lifecycle cost estimates, including assumptions, data, risk analysis, sensitivity, documentation, and updates with actuals.

    2026-09-08
  3. NIST — AI Risk Management Framework Core · National Institute of Standards and Technology

    Requires AI risks and impacts to be measured through documented methods and metrics, independent review where suitable, and ongoing monitoring.

    2026-09-08
  4. OMB Circular A-94 — Benefit-Cost Analysis · U.S. Office of Management and Budget

    Provides official principles for benefit-cost and cost-effectiveness analysis, including alternatives, baselines, discounting, uncertainty, and non-monetized effects.

    2026-09-08
  5. UK Government — The Sourcing Playbook · UK Government Commercial Function

    Connects sourcing decisions to outcomes, whole-life cost, risk allocation, delivery model, performance management, and supplier evidence.

    2026-09-08
Making connections

Continue exploring

AI Strategy & Transformation

How to define AI provider governance and performance management before hiring

Read insight
AI Strategy & Transformation

Boutique AI firm, global consultancy, or systems integrator: how to choose

Read insight
AI Strategy & Transformation

Buy, configure, or build a custom AI solution: how to decide

Read insight
Related capability

AI strategy & transformation

An AI transformation consultancy should answer four questions before recommending technology: where business value exists, which capabilities and data are required, how risk will be controlled, and who will operate the change. MAKINAI connects those answers in an executable plan with priorities, owners, metrics and scale decisions.

Explore this capability