AI Pilot or Discovery Sprint: Which Should You Buy First?
A discovery sprint costs about $2,500 and a pilot about $5,000 in our agency data. See what each one answers, and how to pick the right first step.

What matters most
- The middle published price for an AI discovery sprint is $2,500 across 18 agencies, and two weeks is the common length.
- The middle published starting price for a paid AI pilot is $5,000 across 35 offers, with a middle length of three weeks.
- A discovery sprint answers what to automate, and a pilot answers whether one specific automation works on real data.
- Only 24 of 66 paid pilots say success criteria are agreed before work starts.
- Four of 18 audits and 16 of 66 paid pilots credit the fee toward the full build.
Two agencies, two proposals. The first wants $2,500 for a two-week discovery sprint to "find where AI fits." The second wants $5,000 for a three-week pilot that builds one working automation. Both say their approach is the safe first step. You only have budget and patience for one.
The two are not substitutes. They answer different questions, and buying the wrong one first is how companies spend a quarter and end up with a report or a demo, but no decision. We compared what agencies publish for each, so you can tell which question you actually have. The scenario above is an illustration, not a real client.
Here is what matters most:
- In our agency data, the middle published price for an AI audit or discovery sprint is $2,500, across 18 agencies. Two weeks is the common length.
- The middle published price for a paid AI pilot is $5,000, across 35 offers priced in US dollars. The middle length is three weeks.
- A discovery sprint answers "what should we automate?" A pilot answers "does this specific automation work on our real data?"
- Credit toward the build is uncommon in both: 4 of 18 audits and 16 of 66 paid pilots say the fee comes off the full project.
- Only 24 of 66 paid pilots say success criteria are agreed before work starts. Without them, a pilot cannot give you a clear yes or no.
Where these numbers come from
This combines two of our studies. On November 21, 2026 we read 204 company pages on what AI automation agencies charge and 239 pages on AI pilots and proofs of concept. In both, we asked AI assistants with US web search switched on the questions buyers ask, collected the pages they cited, used AI to pull out prices and terms, and checked every quote we kept against the page text.
The audit figures come from 18 agencies that publish a US dollar price for a first paid step before any build, which we covered in what an AI audit or discovery sprint costs. The pilot figures come from 100 companies that describe their own pilot offer, 66 of which say it is paid. Small agencies are over-represented, and large firms mostly quote privately, so treat these as reference points.
What each one costs and how long it takes
Discovery sprint or audit. Across 18 agencies, the middle price is $2,500, the middle half runs from about $1,150 to $3,400, and the range is $200 to $7,500. Two weeks is the most common length, and one agency commits to ten working days. You get a document: a ranked list of what to automate and, at its best, a fixed price for the first build.
Pilot or proof of concept. Across 35 paid offers priced in US dollars, the middle starting price is $5,000. The middle half runs from about $2,500 to $7,500, and the range is $500 to $50,000. Thirteen start under $3,000 and 7 start at $10,000 or more. Of 71 companies that state a length, the middle is three weeks, and 54 are four weeks or shorter. You get working software on a slice of your real work, and a measured result.
So a pilot costs about twice as much and takes about a week longer. That is not the useful comparison, though. The useful one is what you know at the end.
What each one answers
Buy a discovery sprint first if you do not yet know what to automate. If the honest answer to "which process?" is "we have a list of ten complaints," a pilot is premature. You would be picking the pilot's target by guess. A sprint is the cheaper way to rank the options by time saved and cost to build, and to rule some out.
Buy a pilot first if you already know the process. If you can name the workflow, the people who do it and the hours it takes, a sprint mostly tells you what you already know. A pilot tests the part you do not know: whether the automation copes with your real documents, your real systems and your edge cases.
Skip both if the job is small. A single connection between two tools, or one simple workflow, can cost less than a sprint. In our study of business process automation consulting, the middle published starting price for a project is $2,000. Paying $2,500 to study a $2,000 job makes little sense.
Some agencies sell the two in sequence. One describes a proof of concept that takes four to eight weeks "after a 1–2 week discovery sprint." That order is sensible for a large project, as long as the sprint ends with a fixed price for the pilot.
The terms decide whether either one is worth it
Whichever you buy, three terms matter more than the price.
Credit toward the build. Four of the 18 audits say the fee is credited to the build or the first month. Sixteen of the 66 paid pilots say the fee comes off the full project, and 48 do not say. Ask for it. If the agency treats the first step as part of its sales process, crediting the fee is fair.
Success criteria agreed up front. This matters most for a pilot. Only 24 of the 66 paid pilots say the pass mark is agreed before work starts. Only 4 say both that the pass mark is agreed up front and that the fee is credited. One agency puts it simply: "We define success metrics before the pilot begins, not after." Without a written pass mark, a pilot produces an impressive demo and no decision.
A refund or guarantee. Five of the 66 paid pilots mention one. It is rare, which makes it a useful sign of an agency that is confident in its work.
For a sprint, the equivalent test is what you keep. A report you could hand to another agency or your own team is worth paying for. One that only makes sense to the agency that wrote it is a sales document.
A simple way to decide
Answer two questions:
- Can you name the process and the hours it costs each month? If no, buy a sprint. If yes, go to question 2.
- Is the build likely to cost more than about $10,000? If yes, buy a pilot with a written pass mark before committing. If no, ask for a fixed price for the full build and skip the pilot.
If an agency offers only one of the two whatever your situation, that tells you more about how it sells than about what you need.
Our page on what AI automation costs explains how we price a first step, and our AI consulting page covers how we help companies decide what to automate.
FAQ
What is the difference between an AI pilot and a discovery sprint?
A discovery sprint studies your work and recommends what to automate, usually in about two weeks. A pilot builds one automation on your real data and measures whether it works, usually in about three weeks. The sprint answers what to build. The pilot answers whether a specific build works.
How much does an AI discovery sprint cost?
In our sample of 18 agencies that publish a price, the middle is $2,500, with a range of $200 to $7,500. Four of the 18 credit the fee toward the build or the first month.
How much does an AI pilot cost?
Across 35 paid pilot offers priced in US dollars, the middle starting price is $5,000, with a middle half of about $2,500 to $7,500. The middle length is three weeks.
Do I need a discovery sprint before a pilot?
Only if you do not yet know which process to automate. If you can name the workflow and the hours it costs, go straight to a pilot with success criteria agreed in writing.
Not sure which first step you need?
If you have a sprint or pilot proposal in hand, book a free strategy call. Tell us the process and we will tell you which first step fits, and what to ask the agency to put in writing.


