Hourly vs fixed price for AI automation: why fixed usually wins

When you hire someone to build AI automation, the billing model shapes the project more than the rate does. Hourly vs fixed price for AI automation work is a choice between two sets of incentives, and one of them quietly works against you from the first invoice.

Hourly pays for slowness

Hourly billing pays the vendor more for taking longer and nothing for being clever. That is arithmetic, and I do not think it says anything about vendor character. A UK contractor at £800 a day who needs eight days bills £6,400. One who solves the same problem in three days bills £2,400. The market punishes the sharper builder, and over time everyone responds to that pressure, mostly without noticing they are doing it.

In AI automation the effect is worse than in general software, because this field is full of shortcuts that only experience reveals. Knowing that a prebuilt integration exists, or that a whole category of parsing work disappears if you change where the data enters the system, can erase days of work in a single decision, and on hourly billing every one of those decisions costs the vendor money. On fixed price it is margin, so the vendor goes looking for them. You want to hire the looking.

Hourly also puts all the risk on you. The estimate said five days. It is now day twelve and the integration still fails on edge cases. You are paying for the vendor's learning curve and their misjudgment of the problem, with no ceiling and no recourse except walking away with half a system.

Fixed price forces the scope conversation

The strongest argument for fixed price is something other than cost certainty. A vendor cannot quote a fixed price without first defining the scope, and vague scope is one of the five causes behind the 42% AI abandonment rate.

They rarely die because the technology could not do the job. They die because "automate our intake" meant four different things to four people and nobody wrote down which one was being built. Hourly work can start without resolving that; the meter runs while everyone figures it out. Fixed price cannot start until it is resolved, because the vendor is the one holding the risk. It drags the hard conversation to the beginning of the project, where it is cheap.

Six things a fixed scope must contain

A fixed price is only as good as the document behind it. Before signing, check the statement of work for:

1. Named workflows with a trigger and an outcome. "When a form is submitted, the system qualifies the lead, writes it to the CRM and notifies sales within two minutes." The phrase "lead automation" on its own scopes nothing.
2. The systems touched: which CRM, which calendar, which inbox, and whose name is on each account.
3. Edge cases handled and edge cases excluded. Both lists. Ours now includes two callers booking the same slot in the same second, because one of our voice assistants did exactly that before anyone had thought to write it down. The other ways a voice agent breaks on live calls are on the list by default now.
4. An acceptance test, agreed before the build starts, so delivery is a demonstration and no longer a debate.
5. Warranty. What happens when a bug appears two weeks after handover, and who pays. Ours is 30 days.
6. A change process, so new requests get priced and scope creep becomes a decision.

If a vendor offers a fixed price without most of this, the fixed price is fiction. It will be renegotiated mid-project, from a position where you have already paid.

"Fixed" is sometimes the worse deal

Fixed price is wrong for some work, and a vendor who says otherwise is selling. When the job is true research, nobody has done it before and the deliverable may turn out to be "this is not feasible yet", a fixed quote only means the vendor padded the number to cover the unknown, and you pay the padding whether or not the risk shows up. The same applies to untangling an undocumented legacy system, where scoping the work would cost as much as doing it. The better pattern is a small paid discovery phase, hourly or flat, that turns the unknowns into a scoped fixed build. That is why our rescues begin with a $500 written diagnostic that argues fix or rebuild and why our audit costs $2,500: somebody has to be paid to find out what the work is before anyone can price it.

My view is that hourly should be the exception a vendor has to justify.

How we structure it

Our builds are fixed price, $3,000 to $15,000 depending on scope, half up front and half on delivery. The split matters. The first half commits the client and funds the work; the second half means we are not fully paid until the acceptance test passes, so the risk stays shared. Every build carries a 30-day bug warranty and the client owns every account. The hydrogeology client whose document generator we built (168 documents produced and 349 boreholes on file as of August 2026) came back for a second fixed build at EUR 3,300, and scoping the second one took a fraction of the time, because the first had already answered most of the questions.

The scope document has to come from somewhere. Our AI Opportunity Audit is $2,500, and the whole fee comes off the build price if you go ahead within 90 days. What it produces is the statement of work described above, with a price on it.