Often you don't. That's worth saying first, because a lot of what businesses want from AI right now is a paid seat on a tool and somebody spending an afternoon showing the team how to use it.
Here's where outside help genuinely pays, and what a straight engagement looks like when you're reading one.
When you don't need one
Three situations, and between them they cover most businesses asking the question.
- You want your team using AI day to day. Buy business seats and give somebody two hours to show everyone what it's good at. That's a lunch, not an engagement.
- You want one document summarized or one email drafted. Open it and do it.
- You haven't found a job worth handing over yet. Nobody outside your business can find that for you as cheaply as you can. Watch your own week for two weeks first.
Paying for advice before you have a specific problem gets you a report. What you want is a change in how a job gets done.
When outside help pays for itself
- It has to run without anybody driving it. Once it happens every time rather than when somebody remembers, it's software, and software is a build.
- It touches money, records or a legal obligation. The check-the-answer design is the whole job here, and it's the part that's easy to get wrong invisibly.
- Two systems have to talk and they don't. This is ordinary integration work and it's rarely a good use of your evenings.
- You tried and it half works. Extremely common and a good position to be in, because the hard question is already answered.
What to ask before you sign anything
Five questions. The answers tell you more than any credentials will.
- What does this job cost us today? If they haven't asked, they can't tell you whether the work is worth doing. This should come before any talk of technology.
- What's the fixed price, and what's it based on? A build has a number. Anything that can't be priced until after a paid discovery phase is worth pushing on.
- Who owns the code and the data at the end? The answer should be you, including the accounts. Get it in writing.
- What happens when it's wrong? If there's no answer, they haven't thought about the part that matters.
- What would you tell us not to build? A straight answer to this is the strongest signal there is.
How to tell advice from a pitch
The clearest tell is what they lead with.
Advice starts with your process. What happens today, who does it, how long it takes, what it costs when it goes wrong. Technology comes up late and often the answer turns out to be ordinary software or a setting you already have.
A pitch starts with the technology and looks for somewhere to put it. You'll recognize it by the vocabulary: transformation, journeys, readiness, roadmaps. None of those name a job in your business.
The other tell is whether they'll talk you out of anything. Somebody who says every idea you have is a good one is selling hours.
What it should cost
Rough shapes, so you can tell whether a number is in the right range.
- Getting your team started. A few hours of somebody's time, plus $20 to $25 US per person a month for the seats. This should not be a project.
- Working out what's worth doing. Days, not weeks, and it should end with a shortlist and a number against each one. Be careful of anything that ends with a strategy document and no build.
- Building something. A fixed price agreed before it starts. Custom software here starts at $5,000, automation on its own is often less, and larger or more custom projects are quoted separately.
The prototype is what to insist on. Seeing it work on your own data answers more than any proposal, and it's usually ready within a week.