Skip to content
← All guides
Guide

How long does an AI consulting project take?

Short answer

A first AI consulting engagement for a small or mid sized business typically takes eight to fourteen weeks. That runs from the intro call to a first automation in production. Discovery takes two to four weeks, the roadmap about a week, and the first build four to eight weeks. Projects run long when the inputs are messier than expected or nobody inside the business owns the exceptions.

Anyone who gives you a timeline before they have seen your business is guessing. With that said, first engagements in small and mid sized businesses fall into a fairly consistent range.

A typical first engagement

Intro call, 30 minutes. You describe what you are trying to achieve, what success looks like and the timeframe. The consultant tells you honestly whether they can help and, if so, scopes discovery.

Discovery, two to four weeks. The consultant sits with the people doing the work and sees a full cycle of it. They map where the hours go and pull history from your systems. Two teams in scope takes longer than one. Shorter than two weeks is rarely enough to see the exceptions.

Roadmap, about a week. The candidates found in discovery are scored, cut down to two or three, sequenced and priced. You get the shortlist and the reasons things were cut.

First build, four to eight weeks. One workflow against your real systems, tested against your own history before it goes live. It ships with a queue for the exceptions and someone in your business trained to run it.

Embedding, ongoing. The first month live is where the rules get tuned and trust gets built. The consultant must still be on call.

Add it up and a first automation is typically running eight to fourteen weeks after the intro call.

What makes it run long

Messy inputs. Orders that arrive as photos of handwritten lists, drawings that vary by architect, invoices from suppliers with no consistent format. Each variation costs time to handle. Discovery must surface this. Sometimes it does not surface all of it.

More systems than the brief said. The workflow was meant to touch the accounting system. It turns out to need the job platform and a supplier portal too. Each connection is work.

No baseline. If nobody knows the error rate of the manual process, the consultant must establish it first. Only then can the automation be judged against it.

No owner. The automation is built, works, and sits unused because nobody inside the business was named to run it and handle its exceptions. This is the most common reason a finished project fails, and it is entirely avoidable by naming the person on day one.

What should not take long

Deciding whether to proceed. A consultant can tell you after a 30 minute call whether discovery is worth doing, and after discovery whether a build is worth doing. If they cannot, they are not being straight with you.

How to hold a consultant to a timeline

Buy in stages, each fixed price with a defined end. Ask at the end of each stage whether the next one is still worth doing. And name the person inside your business who will own the result before the build starts.

Our four steps are laid out on the home page. What each stage typically costs is in our pricing guide.

Related questions

Questions

How long does AI discovery take?

Two to four weeks for a business with one or two teams in scope. The consultant needs enough time to sit with the people doing the work, see a full cycle of it, and pull history from your systems. Shorter than two weeks usually means they are guessing.

How long does it take to build one AI automation?

Four to eight weeks for a single workflow against one or two systems, including testing against your own history. Longer when several systems are involved or the inputs vary a lot between customers or suppliers.

Why do AI projects run late?

Three reasons account for most of it. The inputs were messier than anyone admitted. More systems were involved than the brief said. And nobody inside the business owned the exceptions. The automation was built but never fully trusted.

How quickly will we see a return?

A well chosen first build usually starts returning hours within the first month of being live. Payback on the whole engagement within six to twelve months is a reasonable expectation. Treat faster claims made before discovery with caution.

Next step

We map where your hours go and price the work that is worth automating.