Negotiating an AI engagement: discuss value before price

Negotiating starts with framing

An engagement negotiation rarely starts with a number. It starts with a simpler question: what exactly are we talking about? Until the scope, the expectations and the actual role are on the table, any rate discussed rests on an assumption — yours on one side, the client's on the other, and the two almost never match.

That explains a familiar situation: an engagement accepted at a satisfying rate that becomes uncomfortable six weeks in, because the real workload, the stakeholders or the deliverables bear little resemblance to what was described. The problem was not the price. It was the missing frame.

A good negotiation is not only about a better rate. It is about ending up with an engagement whose scope, expectations and conditions are clear enough for you to succeed.

This guide is written for an AI professional talking to a company, a consultancy, a services firm or an intermediary. The logic is the same in all four cases; what changes is the number of people between you and the decision-maker, and with it the amount of information lost on the way.

Understand the scope before discussing money

An engagement title describes an intention, not a responsibility. “Set up an internal assistant” can mean a few weeks of prototyping, the industrialisation of an existing system, or picking up an abandoned project nobody wants to own. Those three engagements do not carry the same workload, the same risk or the same value.

So before quoting anything, reconstruct the engagement behind the title. A handful of elements is enough, as long as they are explicit:

• The objective, stated as an expected outcome rather than a technology to use.
• The deliverables: what must exist at the end, and in what form — a service in production, an evaluated prototype, documentation, knowledge transfer.
• The level of autonomy: do you own the architecture, or apply decisions already made?
• Your role in decisions: consulted, prescribing, or executing a framing done elsewhere?
• The stakeholders: who validates, who arbitrates, how often.

This is not an interrogation. It runs as a working conversation, and it has a useful side effect: it shows you think like someone who will have to deliver, not like someone selling a service.

What the company's AI maturity tells you

Two engagements with the same title demand very different work depending on the environment around them. AI maturity shows in concrete signs, and those signs drive a large share of your real workload.

• Data: does it exist, is it accessible, who owns it, how long does an access request take?
• Infrastructure: is there an environment to deploy into, or must it be built before anything ships?
• Dependencies: how many teams must move for one technical decision to take effect?
• History: is this a first AI project, a restart, or an extension of a system already in production?
• The implicit expectation: is delivery enough, or will you also be expected to bring teams up to speed?

A low-maturity company is not a bad client. It is a context where a significant share of the time goes into getting access, aligning stakeholders and arbitrating expectations. That share exists; say it out loud, or it will be absorbed silently — by you.

One question usually locates maturity: “what has already been tried on this, and why did it stop?” The answer says more than any role description.

Rhythm, duration and visibility

Working conditions weigh as much as scope. A six-month engagement announced as “renewable” with no renewal criteria gives less visibility than a three-month one with a scheduled decision point. In the same way, two mandatory on-site days a week change what you can commit to elsewhere.

• The stated duration, and above all what will trigger an extension or an end.
• The expected rhythm: full time, shared time, fixed days, recurring meetings.
• Location: remote, on-site presence, travel, and whether those rules are negotiable.
• Notice period, both ways, when it makes sense for the duration involved.
• The real start date: an engagement beginning “once access is ready” has no start date.

What makes an engagement genuinely worth taking

“Interesting” is not a vague feeling; it breaks down. The criteria below do not carry the same weight for everyone, which is exactly why making them explicit helps — they make two superficially similar offers comparable.

• The problem: is it real, articulated, and will it still matter in six months?
• Responsibility: will you produce defensible decisions, or only deliverables?
• Learning: does the engagement add something you can demonstrate afterwards?
• Evidence: will you be able to talk about it later, at least in anonymised form?
• Conditions: are the rhythm and duration compatible with the rest of your activity?
• Outcome: can the engagement actually conclude, or does it depend on a prerequisite that does not exist?

An engagement can pay less and still be a good choice. It can also pay well and cost you dearly: shifting scope, impossible trade-offs, unusable results. Negotiation is what makes that comparison possible.

Discussing the rate without quoting a benchmark

A rate is far easier to discuss once the scope is set. “At that level of responsibility and over that duration, my rate is X” is a defensible position. The same sentence without framing forces the other side to compare your number against a reference they pick alone.

There is no need to invoke market averages: they vary with sector, company size, intermediaries, urgency and scarcity, and an average quoted from memory tends to turn against whoever cited it. What holds instead is your own consistency: a stable rate, explained by the content of the engagement, and owned.

Negotiating on something other than price

When the budget really is fixed, there is still room in places that cost the company nothing and change a lot for you:

• Scope: narrow the first phase to a reachable objective, then revisit.
• Duration: a longer engagement at the same rate sometimes beats a higher rate over three weeks.
• Rhythm: fixed days, remote work, availability windows — valuable if you run several engagements.
• Practical terms: payment terms, monthly invoicing, travel covered.
• Responsibility: securing a decision-making role rather than an execution one, which changes what the engagement is worth afterwards.
• Visibility: the right to mention the work, even anonymised, once it is over.

These levers work best when offered — not claimed at the end of the conversation, once everything else has been settled.

Signs of a badly framed engagement

Some signals are visible before signing. None is disqualifying on its own; an accumulation of them is.

• Nobody can say which problem the engagement should solve, only which technology to use.
• The deliverable changes from one conversation to the next.
• No stakeholder can describe the state of the data or how access works.
• You have never met the decision-maker, and the intermediary cannot answer substantive questions.
• The announced duration is short while the objective implies several months of work.
• Urgency is the only argument offered, including as a way to avoid pinning down scope.
• You are asked to commit to an outcome that depends entirely on things outside your control.

Questions to ask before accepting

Around ten questions are enough to surface what matters. They fit naturally into one or two conversations.

• Which problem should this engagement solve, and for whom?
• What has already been tried, and why did it stop?
• What does success look like at the end?
• Which data is available, and how soon can I access it?
• Who decides the architecture, and who signs off the deliverables?
• Which teams will I work with, and how often?
• What is expected in production, and what stays exploratory?
• What happens if the evaluation shows the planned approach does not fit?
• How long is it, and on what criteria would it be extended?
• Who is my day-to-day contact when something is blocked?

The answers matter, but so does how easily they come. A company that answers in ten minutes has already done its framing. A company promising to come back to you on every point is telling you how much framing work is waiting for you.

Declining, or asking for a reframe

Turning down an engagement is not a failed negotiation: sometimes it is the only decision consistent with what the conversation revealed. A reasoned decline, phrased without judgement, keeps the door open — and proposals do come back a few weeks later, better framed.

Between acceptance and refusal sits an underused option: proposing a reframe. A short first phase with a precise objective — a review of the data, an evaluation of one approach, a measured prototype — then a joint decision on what follows. That phase protects both sides: it gives the company a basis to decide, and it saves you from committing to twelve months of work in an environment you barely know.

A well-negotiated engagement shows on day one: you know what you have to produce, who you talk to, and how the result will be judged.

• Freelance
• Market