Value of advice
AI Consulting ROI Framework
Evaluate advice through changed decisions, credible attribution and the value of resolving uncertainty.
A consulting engagement can improve a decision without owning the entire result that follows. An advisor may help stop a weak project, narrow an unsafe scope or identify a prerequisite. Operations, engineers and employees then create much of the realised value. Measuring consulting ROI requires a credible account of what changed because of the advice and what would probably have happened without it.
For a Saudi finance director or transformation sponsor, this distinction prevents two errors: crediting a consultant with every subsequent saving, and declaring useful advice worthless because it did not directly generate revenue. Use this framework to assess the next advisory purchase and review one already completed.
Start with the decision the advice can change
Write the decision before the engagement starts: whether to fund a pilot, which two opportunities to prioritise, whether to renew a vendor, or how much authority an agent may receive. Record the current preferred option, the evidence behind it, remaining uncertainty and the person authorised to decide. This becomes the baseline for evaluating the advice.
A deliverable count is an activity measure. Ten workshops or a hundred-page report may be useful, but neither demonstrates an improved decision. Specify the change you expect: eliminating an unsupported benefit assumption, resolving a data-access constraint, comparing feasible alternatives or establishing a test that prevents premature investment. The transformation business case evaluates the wider programme; this page evaluates the contribution of the advisory work.
Separate four kinds of value
| Category | Possible effect | Evidence boundary |
|---|---|---|
| Avoided commitment | A weak project is stopped before further spend | Show that the spend was likely, avoidable and actually cancelled |
| Improved option | A different scope produces better net outcomes | Compare against a credible alternative, including extra delivery cost |
| Earlier decision | A dependency is resolved sooner | Identify the critical-path change and the value of the time gained |
| Organisational capability | Teams can repeat the decision process themselves | Observe later use and quality; do not assume a cash return from training attendance |
Keep these categories separate where they describe the same underlying benefit. A cancelled project cannot simultaneously generate the revenue it would have produced and the full value of avoided spending. Nor should the same reduction in staff effort appear under both improved productivity and reduced operating expenditure unless those effects are independently demonstrated.
An attribution ladder for honest claims
Begin with direct observations: the advisor found a missing access dependency, the investment committee changed its decision, and a purchase order was not issued. Then examine the causal link. Was the dependency already known? Would procurement have rejected the purchase anyway? Did another team produce the decisive evidence first? A useful evaluation invites these alternatives rather than defending the consultancy's narrative.
The UK government's Magenta Book methods annex describes contribution analysis as a way to test a reasoned account of an intervention's contribution against evidence and other explanations. It does not turn that account into definitive proof or a precise effect size. Here it is a methodological reference, not a Saudi regulatory requirement.
Report the strength of the claim: observed decision change, supported contribution, estimated attributable benefit or causally evaluated impact. Do not jump from a before-and-after improvement to a precise consulting ROI. Seasonality, a new manager, changed workload and parallel data remediation may explain part of the movement.
Worked example: the value of resolving uncertainty
Consider an illustrative group choosing whether to launch a document-review programme. A SAR 600,000 commitment is under consideration. Before investigation, management assigns a 35% probability to a blocking data-rights problem. A narrowly scoped review costs SAR 60,000. The first question is whether better information could change the funding decision, not whether the consultant can promise savings.
Under a deliberately simplified model, assume the review perfectly identifies that blocker, the entire commitment would otherwise be lost if the blocker exists, and management will cancel it when identified. Expected avoided loss is 600,000 × 35% = SAR 210,000. After the fee, the modelled net benefit is SAR 150,000. Those strong assumptions make this an optimistic illustration, not a forecast.
Make the model more realistic. If the review detects the problem 70% of the time and only SAR 400,000 is genuinely avoidable, expected avoided loss becomes 400,000 × 35% × 70% = SAR 98,000. Subtract SAR 60,000 advisory fees and SAR 12,000 internal evaluation cost: modelled net benefit is SAR 26,000. The corresponding illustrative return on the SAR 72,000 decision cost is 26,000 ÷ 72,000 = 36.1%.
This second calculation still ignores false alarms, delay and the upside of projects incorrectly cancelled. Add those effects before using it for a real decision. If the blocker is already certain or the board will proceed regardless of the findings, the value of this particular information may be very small. Advice creates value through a changed action, not through the existence of an additional report.
Evaluate realised value without claiming hypothetical cash
After completion, distinguish prevented budget from booked savings. A cancelled future purchase may preserve cash but may also remove expected benefits. A better prioritisation decision may move staff to higher-value work without reducing payroll. Report these as different outcomes, with finance agreeing which measures belong in the financial accounts, capacity plan or investment narrative.
For a continuing programme, track decision cohorts. Compare cases reviewed with advice against similar cases handled under the prior process where feasible, and explain differences in complexity or risk. If comparison is impractical, maintain an evidence chain: original option, advice, changed action, subsequent outcome, alternative explanations and confidence. An honest range can be more useful than an unsupported single percentage.
A reusable consulting value record
- Decision ID and date: connect the value claim to an actual authorised choice.
- Baseline option: capture what the organisation expected to do before the advice.
- Information delivered: identify the finding, its evidence and unresolved uncertainty.
- Changed action: record the decision owner's response, including a decision not to change.
- Outcome window: state when benefits and costs can reasonably be observed.
- Contribution assessment: list other teams and changes that affected the result.
- Finance review: classify realised cash, capacity, risk reduction and unverified estimates separately.
- Next review: name the owner who will confirm or revise the claim.
Agree the record before a consulting engagement, not after a supplier asks for a success story. Include the cost of buyer participation and necessary follow-up in the denominator. For retainer arrangements, assess a portfolio of decisions over an agreed period rather than assigning invented savings to every meeting.
Use the findings to buy the next piece of advice
If the review finds that advice was accurate but ignored, examine authority and incentives before purchasing more analysis. If it arrived after the investment was irreversible, change timing. If it answered a question no decision maker owned, change scope. The strategy diagnostic can help isolate a wider execution problem; the transformation cost guide covers the full investment beyond advisory fees.
Open for AI's agentic AI consulting can be scoped around a defined investment or operating decision. Discuss an evidence-led advisory evaluation with the proposed decision, current preferred option and what new information could change it. That is a firmer starting point than asking a consultant to guarantee a return they do not wholly control.
FAQ
Your questions, answered.
How do we calculate ROI on AI consulting itself?
Define the decision affected, estimate the incremental benefit attributable to better advice, subtract all advisory and buyer participation costs, then divide the net benefit by those costs. The difficult part is credible attribution, not the formula. Report confidence, alternative explanations and the period measured; do not simply divide the whole transformation saving by the consulting fee.
Can stopping an AI project count as consulting value?
Yes, when the project would plausibly have proceeded, the advice materially changed that decision and the avoided commitment is documented. Deduct the cost of the review and account for benefits forgone by stopping. A cancelled budget is not automatically a realised accounting saving, and a project that was already certain to stop creates little additional attribution.
How can we measure advice when no control group exists?
Use a documented contribution account: what would have happened, what evidence the advisor added, how the decision changed, what followed and which competing explanations remain. Obtain the decision owner’s and finance team’s challenge. This can support a proportionate value judgement, but it does not establish the same causal confidence as a well-designed comparative evaluation.
How should consulting value be assessed when only part of the recommendation is adopted?
Record which recommendations were accepted, modified or rejected and why, then trace outcomes only to the actions actually taken. Compare against the credible baseline and include the cost of the adopted changes. Separate weaknesses in the advice from execution choices or later constraints. Do not claim the full forecast benefit, or assign every shortfall to the advisor, when the organisation implemented a materially different approach.
How should we evaluate a continuing AI advisory retainer?
Review a defined set of decisions over the period, the capacity actually used and the evidence that changed choices. Include useful challenge and capability transfer, but avoid attributing a fictional saving to each meeting. Renewal should depend on remaining decision needs, observed contribution and whether internal teams can now perform the work.
What makes an AI consulting ROI claim unreliable?
Warning signs include no baseline decision, all programme benefits credited to one advisor, omitted internal costs, ignored implementation costs, duplicated benefits and a precise percentage based on untested assumptions. Ask what would disprove the claim. If no evidence could change it, treat it as a promotional assertion rather than an evaluation.