What a $3,000 AI pilot actually buys you

Most quotes you will get for "an AI tool for our business" are a range with no scope attached. Somewhere between $15,000 and "let's start with a discovery phase." That is not a price, it is a way of not answering the question until you are already invested.

So here is the direct answer. A first engagement with me is a fixed price between $3,000 and $6,000, it runs two to three weeks, and it produces one narrowly scoped system that is running in production at the end of it. Not a prototype, not a deck, not a roadmap. Something your team opens on a Monday and uses.

Below is what that money actually covers, what it does not, and why the number is fixed rather than hourly.

What is in the $3,000 to $6,000

One workflow, end to end. The scope is deliberately a single job: the document that gets read and filed, the lead that gets scored, the report that gets assembled every Friday. Not "an AI layer for operations." One workflow, picked because it is the one your team complains about most, and taken all the way from input to the place the output has to land.

The integration into where the work already happens. A tool nobody opens is worth nothing, so this is not a standalone app with its own login unless there is a reason for one. It writes into the CRM, posts into the Slack channel, drops the file into the shared drive, updates the row in the sheet. The integration is not a phase two.

The unglamorous reliability work. Retries, rate limits, a cap on model spend, and a defined behaviour for every failure path. This is the part that separates a demo from a system, and it is the part most often quietly left out of a cheaper quote. If the model is down, or slow, or refuses, the tool still has to do something sensible rather than show your operations manager a stack trace.

The repository, in your account, from day one. Not at the end, not on final payment. You own the code, it runs on your infrastructure, and if we never speak again you keep a working system. There is no per-seat licence and no platform subscription to me.

A written handover. What it does, what it costs to run per month, where the API keys live, what breaks first as volume grows, and what the next sensible thing to build is. Two pages, plain English, written for whoever inherits it and not for me.

Direct access to the person building it. There is no account manager layer, because there is no account manager. You are talking to the developer, which is the main reason a two week timeline is possible at all.

What is deliberately not in it

Custom model training. I build on frontier model APIs. If your problem genuinely requires a fine-tuned or self-hosted model, a pilot is the wrong shape of engagement and I will say so before you pay anything.

A multi-tenant product. A pilot builds a system for your company, not a SaaS you could resell. Those are different builds with different costs, and conflating them is how a $5,000 project becomes a $50,000 one.

Unlimited revisions. Two rounds of feedback are in scope. Past that, we are designing a different system, and the honest move is to price it separately rather than absorb it and resent it.

Data cleanup. If the CRM has four years of duplicates, that is real work and it is not two weeks of it. I will tell you what shape the data needs to be in, and that is usually work your team can do faster and cheaper than I can.

Ongoing operation. A pilot ends. If you want someone watching it, changing it, and building the next three things, that is a retainer, and it starts at $4,000 a month. But nothing about the pilot obligates you to it, and that is the point.

Why it is fixed price and not hourly

Hourly billing punishes the person who is fast, which is a strange thing to select for when you are buying software. It also makes every clarifying question feel expensive to you, which is exactly the conversation that determines whether the build is any good.

More practically: an hourly quote transfers all the estimation risk onto you. If I misjudge the work, you pay for my misjudgement. Fixed price puts that risk on me, which is where it belongs, because I am the one who is supposed to know how long this takes. The trade is that the scope has to be genuinely narrow, and I will hold that line during the project.

Why there is a floor, and why it is not free

I do not do free discovery calls that produce a written proposal. That sounds like a small thing but it is the single biggest filter in this business. Free scoping work attracts people who are collecting quotes to benchmark an internal hire, or who wanted a free consultation and framed it as a project. Both are real, both are expensive, and neither ends in a build.

A small paid pilot filters for the buyer who has an actual problem and a budget line, and it shortens the sales cycle from months to about a week. It is also, frankly, the fairest version: you risk $3,000 to find out whether working together produces anything, instead of risking $40,000 on a proposal that reads well.

How the number moves inside the range

The $3,000 end is one clearly defined input, one clearly defined output, and a system you already have an API for. Scoring inbound leads against a rubric and writing the result back to HubSpot sits here.

The $6,000 end is where there are multiple input formats, a system with no usable API that has to be driven another way, or a compliance constraint that changes how data is allowed to move. Document intake from email attachments in four different vendor formats sits here.

If it does not fit in $6,000 and three weeks, I will tell you that before you pay, and I will tell you what the smaller first version looks like. Splitting a large build into a pilot plus a follow-on is almost always better than compressing it, because the follow-on gets designed with three weeks of real usage data behind it instead of guesses.

The honest caveat

I am a solo studio. That means you get the person who wrote the code on every call, and it also means I take a limited number of these at a time and I am not the right choice if you need twelve people staffed by Monday. If what you actually need is a team, say so early and I will tell you plainly that this is not the right fit rather than stretch to take the work.

If you want to see what a scoped build looks like before talking to anyone, the Scope it live tool on the homepage will take a problem description and hand back the architecture, the sequence and the stack, for free and without a form. That is the same thinking that goes into a pilot, minus the building.