Pricing
What AI Consulting Costs
The cost of an AI project is decided almost entirely before any model is called, and the largest variable is one that rarely appears in a proposal: the state of your data.
Two businesses can ask for the same assistant. One has documents in a consistent format with an owner, and the build takes weeks. The other has the same material spread across shared drives in four formats with three versions of each file, and the same build takes months, because the real work is sorting that out.
An honest quote establishes which of those you are before quoting.
The staged shape, and why it is staged
Discovery. Map the process, score the opportunities, assess whether the data supports them. Fixed scope, ends in a decision. This stage regularly concludes that the highest-value item is not the one that prompted the enquiry.
Proving one case. Build the single highest-value opportunity end to end and measure it against the manual baseline. Fixed scope against a written specification.
Building out. Only after the proving case has produced a measured result. Priced per use case, because each has different data requirements.
The staging is not a sales structure. It exists because the alternative, committing to a platform programme before anything is proven, is how these projects lose their sponsor at month nine.
Where budgets get wasted
Four patterns account for most of it.
Building before the data is ready. The most expensive one, because the build gets completed and then cannot go live. Data readiness is assessed in discovery for exactly this reason.
Automating the wrong step. Usually the visible step rather than the expensive one. Measuring the process first is what prevents it.
Paying for the largest model by default. On retrieval work, a cheaper model usually scores the same against an evaluation set. Not measuring is what makes running costs several times higher than necessary.
No metering. An unmetered integration calling a paid model in a loop is the most common way a small project produces an unexpected invoice.
Running costs, which are separate
Build cost and running cost are different conversations and should be quoted separately.
Running cost is driven by usage volume, how much context each request sends, and which model answers. All three are measurable, and everything I build meters them from the first day with a hard daily cap and an off switch.
That means you can see the real running cost in week one and decide whether it is worth it, rather than discovering it at the end of a quarter.
Related
Talk about your situation
The first conversation is short and mostly questions. Get in touch and tell me what you are trying to fix.
Frequently Asked Questions (FAQs)
How much does AI consulting cost?
It depends far more on the state of your data and processes than on the technology, which is why the engagement starts with a fixed-scope discovery rather than a quote for a build. Discovery ends with a scored opportunity list and a specification, which is what makes the build quotable with any accuracy.
What is the difference between build cost and running cost?
Build cost is the one-off work to specify, build and harden the system. Running cost is what it costs per month once people use it, driven by volume, context size and model choice. They should be quoted separately, and any supplier who quotes only the first has left out the number that matters over a year.
Can we start small?
That is the recommended route. Discovery is a defined piece of work that ends in a decision, and proving a single use case is deliberately sized to produce evidence rather than a platform. Building out only follows a measured result.
What if the discovery says we should not build anything?
That is a legitimate and reasonably common outcome, and the discovery is priced as a standalone deliverable so it is available. Finding out that the process should be fixed rather than automated, or that a product already solves it, is worth considerably more than a build nobody uses.