Every quote you have been given for automation work is a different number, and none of the vendors will explain why. That is not because the answer is secret. It is because the number depends almost entirely on things they have not looked at yet.
Here is what actually moves the price, so you can work out your own range before anyone quotes you.
The model is the cheapest part
People assume the AI is the expense. It rarely is. Running a language model over a few thousand emails a month costs less than a phone line. What costs money is everything around it.
The four things that actually drive the number
1. How many systems it has to touch
One system is a script. Two systems is an integration. Five systems is a project. Each connection is a set of credentials, a data format, an error path and a thing that can change underneath you without warning. If you want a rough mental model: cost climbs with the number of connections, not with the cleverness of the automation.
2. How much judgment has to be encoded
"Move this number from here to there" is cheap and stays cheap. "Decide whether this is a change order" is expensive, because somebody has to sit with you and extract a rule that currently exists only as your instinct. That extraction is the work. The software is the easy half.
3. How clean your data already is
If the same customer exists three times under slightly different names, that gets handled before anything can be connected — or the automation faithfully propagates all three. Data cleanup is invisible on a proposal and very visible on a timeline.
4. What happens when it goes wrong
A build with no monitoring, no error handling and no human approval step is cheaper on day one and more expensive on day ninety. Anything that touches a customer or moves money needs a person in the loop, and that design work is part of the price. A vendor who has not mentioned failure has not priced it.
Why we will not publish a number on this page
Because we would have to invent it. A flooring contractor connecting two systems and a professional-services firm rebuilding an intake process are different orders of work, and publishing a single range would mislead both of them.
What we will publish is the honest shape of it: the cost is dominated by connections and judgment, not by AI; a first project should be small enough to prove itself in about 90 days; and the cheapest good outcome we regularly find is a setting inside software you already own, which costs nothing at all.
Estimate your own range in an afternoon
- Count the connections. Write down every system the work touches between the first contact and getting paid. That count is the biggest single driver.
- Count the "it depends" moments. Walk the process out loud. Every time you say "well, it depends," that is judgment somebody has to encode.
- Check your data. Search your CRM for a customer you know well. If they appear more than once, add cleanup to the plan.
- Price the alternative. Hours per month currently spent on the manual version, times a loaded hourly cost. That is not what the build costs — it is what doing nothing costs, and it is the only number that tells you whether to proceed.
If step four is a small number, the honest answer is do not automate this. We tell operators that regularly, and it is a better outcome than a system nobody needed.
If you would rather have somebody map this with you, the Architecture Assessment is 45 to 90 minutes and ends with a written report. It is $250, credited 100% toward any build — and if we find nothing worth building, we say so. Book it here.