Why will most AI consultancies not publish rates?

Because pricing is often set against perceived budget rather than scope. We would rather lose an enquiry early than spend two calls discovering we are an order of magnitude apart on expectations.

Do you charge hourly or fixed fee?

Fixed fee for defined deliverables such as a strategy engagement or a scoped pilot, because you should not carry the risk of our estimate being wrong. Ongoing support and open-ended build work runs monthly.

What is the smallest sensible engagement?

A focused assessment on a single use case. If your question is whether a specific workflow is worth automating, that is answerable in a couple of weeks and does not require a full strategy engagement.

What drives cost most?

Data readiness, more than anything technical. A client with clean, accessible, well-documented data can move three times faster than one whose source system requires reverse engineering. The second most common driver is the number of stakeholders who must approve the output.

Pricing

What AI Consulting Actually Costs

Almost nobody in this field publishes numbers. Here are ours, with the caveats that make them meaningful.

Typical ranges

Engagement	Range	What that covers
AI strategy and roadmap	$25,000 to $60,000	Six weeks or so. Readiness assessment, use-case scoring, technology recommendation and a phased roadmap with a business case. Fixed fee.
Scoped pilot	$40,000 to $120,000	One use case built against real data and put in front of real users. Range depends mostly on data accessibility and integration count.
Production deployment	$120,000 to $400,000+	A system in production with monitoring, evaluation and handover. On-premise deployments sit higher because of infrastructure and validation work.
On-premise infrastructure	From $15,000 in hardware	Separate from services. A single professional GPU serving a small team is modest; multi-GPU capacity for a large model at concurrency is materially more.
Ongoing support	$5,000 to $25,000 per month	Monitoring, model updates, evaluation and capacity planning. Optional. Many clients take handover and run it themselves.

These are the ranges our engagements have fallen into. They are indicative, not a quote. Scope, data condition and integration count move the number substantially in both directions.

What moves the number

Data readiness, more than anything else

This dominates. A client whose data is clean, accessible through a documented API and reasonably structured can move several times faster than one whose source system needs reverse engineering before any AI work starts. When an estimate comes in high, data condition is usually why.

Integration count

Each system the AI must read from or write to adds work, and the ones without an API add far more. Two integrations is a different project from seven.

Regulatory validation

Where output feeds a regulated decision, validation and documentation can be a third of the effort. This is real work, not overhead, and it is what makes the deployment defensible.

Stakeholder count

Underrated. A project with one decision-maker and one user group moves quickly. One requiring sign-off from four departments spends most of its calendar time waiting, and calendar time costs money on any engagement with people assigned to it.

What we will tell you for free

Whether the problem you have is worth solving with AI. That conversation is a 30-minute call and there is no charge for it, including when the answer is that a rules engine, a better integration or hiring one person would serve you better. We give that answer often enough that it is worth saying out loud.