AI Project Cost Estimator — What AI Automation Costs to Build

How much does it cost to build an AI automation system?

The dominant cost driver is not the AI — it is how reachable your existing systems are. A single workflow against a system with a modern, documented API is a short project. The same workflow against a legacy system with no API can be several times the effort, because an integration layer has to be built before any automation is possible. Scope, number of channels, whether voice is involved, governance requirements and whether models must run in your own environment each add on top. This estimator sizes engineering effort in days and converts it at a rate you set.

Methodology

How this is calculated

Published in full, so you can disagree with it. A tool that hides its model is a lead form.

A base effort is set by scope: 18 days for a pilot, 40 for one production workflow, 85 for several connected workflows, 160 for a platform.
Integration effort = systems × 5 days × an API-maturity factor (1.0 modern, 1.35 partial, 1.8 legacy, 2.3 no APIs). This is usually the largest line.
Each additional customer channel adds 6 days. Voice adds 22. Heightened governance adds 15. Self-hosted models add 18.
The total is presented as a range of 0.85× to 1.35×, because discovery routinely changes the integration picture in both directions.
Elapsed time assumes a small focused team, not one engineer working alone.

Assumptions and limits

This is a planning aid, not a quotation. No estimate produced without seeing your systems should be treated as a price.
Running costs — model usage, telephony, infrastructure — are excluded and are typically a small fraction of build cost in year one.
Ongoing tuning and monitoring after launch are excluded. Budget for them: an automation left unmonitored degrades.
The effort figures come from our own delivery experience on comparable work, not from an industry benchmark.

Worked examples

One workflow, two modern-API systems
Around 50 engineering days at the midpoint — a focused project with a clear measurable outcome, which is the right way to start.

Platform scope across five legacy systems
Well over 200 days. We would advise against starting here regardless of budget: prove one workflow first.

FAQ

Questions about this tool

Because calling a model is straightforward. Getting reliable, permissioned, real-time access to the systems that hold your data — and handling what happens when they fail — is the engineering. Vendors who lead with model capability are usually the ones who have not done the integration.

If it targets the riskiest unknown, yes. A pilot that proves the integration works and the extraction is accurate de-risks the whole programme. A pilot that demonstrates a chatbot answering questions proves nothing you did not already know.

Then narrow the scope rather than cutting quality. One workflow, one channel, one system, done properly, will teach you more than a broad build that is never trusted enough to switch on.

It reflects how we think about effort, but our actual proposals come out of discovery where we can see the systems. We publish it so you can sanity-check any quote, including ours.

Related tools

If you want this built

Turn the number into a plan.

Send us the workflow behind your result. We will come back with how we would automate it, what stays human, and what it takes to build.

Find What I Can Automate Talk about a workflow