Should you hire an AI consultant or do it yourself?

Do both. Just know which is which.

The Federal Reserve found that access to AI tools is about the same for a two-person shop as for a large company. So this is not a question of whether you can do AI yourself. You can, today, for the price of a subscription.

The real question is which kind of AI work is in front of you, because there are two kinds and they have almost nothing in common.

The first kind helps a person work faster. Drafting emails, summarizing a contract, researching a competitor, cleaning up a proposal. If that is the work, do it yourself. Nobody needs a consultant to use ChatGPT, and anyone selling you one should make you suspicious.

The second kind does a job so a person does not have to. Leads researched before you walk in. Invoices reconciled without an afternoon of copy-paste. Intake sorted before you open the inbox. That is not a prompt. That is software: it has to connect to the systems where the work lives, survive messy real data, and keep running when one of those systems changes.

Why the DIY route stalls on the second kind

MIT looked at company AI projects and found only about 5% ever reach daily use. The pattern behind that number is consistent. The demo works. Then real data breaks it, or it never gets connected to the tools where the job actually happens, or the one person who understood it gets busy and it quietly rots.

The hard part was never getting a smart answer out of the model. The hard part is the plumbing and the upkeep, and that is plain software and operations work. It takes time you do not have, because you are busy running the business.

What you are actually paying for

In the same MIT research, projects built with outside help reached production about twice as often as internal builds. Not because outsiders have better models. Because making software hold up inside a real company is a skill, and it is the whole job rather than a side project squeezed between everything else you do.

So pay for the build, not the talk. A few things keep you safe:

• Ask to see something they built that is running in a business today.
• Start with one workflow, not a roadmap. It should ship in weeks and either work or not.
• Ask who maintains it after launch, and what that costs. An automation with no owner has a shelf life measured in months.

If you want a sense of real numbers before talking to anyone, we wrote up what AI actually costs a small business.

A simple way to decide

If the AI helps a person type faster, do it yourself and keep the money. We say the same thing in our guide on just using ChatGPT: use it, it is worth it.

If the AI is supposed to do a job while nobody watches, treat it as a build and staff it like one, whether that staff is you or someone you hire. If you are not sure which side of the line your idea sits on, that is exactly what the assessment is for.