AI consulting fees are difficult to compare because apparently similar engagements can purchase very different things.
One “AI strategy” may be a two-day workshop and a use-case list. Another may include economic baselining, workflow analysis, opportunity prioritisation, technical discovery and a tested path to investment. One “pilot” may be a supplier demonstration. Another may be a working capability used inside real operations.
The useful buying question is not simply “What is the day rate?” It is:
What decision, evidence or capability will the business own at the end—and what uncertainty will have been removed?
Buy an outcome from the engagement, not a promise about the whole transformation
No responsible consultancy can guarantee an uncertain AI opportunity will create a specific financial result before the assumptions are tested.
It can be accountable for the quality of the work it produces at each stage.
For example:
Strategy stage
Buy:
- a material outcome and baseline;
- a focused opportunity portfolio;
- explicit value hypotheses;
- prioritisation logic;
- ownership, measures and a path to proof;
- clear decisions about what not to fund.
Do not accept a long use-case inventory as the main deliverable.
Opportunity assessment stage
Buy:
- a view of the actual workflow and constraints;
- an evidence-based opportunity score;
- the economics and assumptions behind the case;
- a recommendation to back, sharpen, defer or stop;
- a bounded Build and Prove design.
Do not accept a generic maturity score disconnected from a specific value decision.
Build and Prove stage
Buy:
- a working Minimum Viable Capability;
- representative real-world testing;
- observed user and workflow evidence;
- a measured update to the value case;
- known data, control and integration requirements;
- a scale, revise or stop recommendation.
Do not accept technical feasibility alone when the investment decision depends on operating value.
Embed and Scale stage
Buy:
- a redesigned end-to-end workflow;
- named ownership and decision rights;
- technology, data and control integration;
- role and capability change;
- benefits-realisation measures;
- a business able to own and improve the capability.
Do not treat licences deployed or users trained as proof of transformation.
Match the commercial structure to the uncertainty
Different stages justify different pricing models.
Fixed fee for a bounded decision
A fixed fee can work well when the scope, access, outputs and decision window are clear—for example, an opportunity review or strategy sprint.
The agreement should define:
- the outcome and questions in scope;
- required access and inputs;
- the outputs;
- assumptions and exclusions;
- the decision at the end.
Stage-based fee for Build and Prove
A prototype or MVC contains discovery. Stage-based pricing can protect both parties from pretending the path is fully known.
Tie each stage to evidence and a choice about the next increment. Avoid a structure that makes continuation the default regardless of the result.
Time and materials for genuinely variable work
Time and materials may be appropriate where the client controls priorities, the environment is uncertain or specialist support is required flexibly.
It should still be governed by outcomes, a backlog and regular decisions. Hours are the charging unit, not the definition of value.
Retainer for ongoing senior judgement
A retainer can be useful when leadership needs continuing access to an experienced operator across opportunities, suppliers and decisions. Define availability, response expectations and the substantive work charged separately.
Do not use a retainer to hide undefined delivery.
Value-linked elements
A value-linked component can align incentives when the baseline, attribution, timing and client-controlled dependencies are clear. It is less suitable where many factors outside the consultancy's control determine the result.
Avoid artificial success fees based on activity measures or modelled benefits.
Compare the full cost, not the headline fee
A lower consulting fee can create higher total cost if it produces weak requirements, technical debt, unsupported change or a pilot that must be rebuilt.
Include:
- internal leadership and subject-matter time;
- data preparation;
- software, model and infrastructure cost;
- integration and security;
- human review and operations;
- change, training and communications;
- parallel running;
- ongoing evaluation and improvement;
- switching or dependency risk.
The economic question is cost per improved decision or realised capability, not cost per consultant day alone.
Ask who will actually do the work
The credibility of the pitch team and the delivery team may differ.
Clarify:
- named team and senior involvement;
- which experience belongs to those people;
- who will inspect the workflow;
- who can build;
- who will work with risk and technology;
- who owns the value model;
- what client capability is required;
- how knowledge and ownership transfer.
A smaller senior team can be more valuable than a large pyramid when the work depends on judgement and speed. A larger team may be appropriate when the direction is proven and scaled delivery is required.
Define acceptance criteria before signing
For each deliverable, agree what “done” means.
Examples:
- Leadership can choose one priority and explain why.
- The value hypothesis contains an agreed baseline and conversion logic.
- The MVC operates on a defined set of representative cases.
- Human review effort is measured.
- Scale, revise and stop thresholds are agreed in advance.
- The client owns the code, data, documentation or operating process as specified.
- The new workflow has a named owner and benefits measures.
Acceptance criteria improve the engagement even when the commercial model is time based.
Procurement questions worth asking
- What decision will this stage enable?
- What observed evidence will we own?
- What assumptions remain outside scope?
- What access and client effort are required?
- What would cause the engagement to recommend stopping?
- What becomes reusable capability inside our business?
- What production, integration or change cost is not included?
- How are intellectual property, data and model outputs treated?
- How will value be measured and validated?
- What must be true before the next stage is funded?
The standard to use
A good AI consultancy fee purchases clarity where the decision is uncertain, evidence where the value is unproven and capability where the opportunity has earned investment. The commercial model should make it easier to focus, learn and stop—not reward the volume or duration of activity.