What an AI consultant actually does (and what they don't)

A good AI consultant does three things: identifies where AI will actually move the numbers in your business, builds or selects the right tools for the job, and owns the implementation end-to-end. Anything short of that is a strategy deck, and strategy decks rarely ship.

The short version

A good AI consultant does three things. They find where AI will actually move the numbers in your business, they build or select the right tools for the job, and they own the implementation end-to-end. That is it.

Everything else is a strategy deck. Strategy decks are cheap, they look smart, and they almost never ship. If the person you are talking to cannot tell you what will be running in production by the end of the engagement, you are buying a deck.

What a good consultant actually does

The work breaks down into four phases, and the shape of each phase matters more than the name of the framework.

1. They understand the business before they touch the tech

The first week is not about AI. It is about understanding how the business actually works — what the customer journey looks like, where the money is made, where the money leaks, which decisions take the longest, and which humans spend the most time on repetitive work. Most AI projects fail because they skip this step and build the tool someone asked for instead of the tool the business needs.

You should expect pointed questions about things that have nothing to do with technology: your margin, your churn, your hiring plan, your data hygiene. If a consultant starts with "which LLM should we use?", walk away.

2. They identify the real opportunity

Out of that first-week conversation, three to five opportunities usually surface. The job is to rank them by two things only: how much they will move the number, and how realistic they are to ship. The winner is almost never the one everyone got excited about in the brainstorming session.

This is the step where a good consultant earns their fee. Saying "no" to a project that sounds cool but will not ship — or that will ship and change nothing — is worth the entire engagement on its own.

3. They build or buy, deliberately

There is no shortage of AI tools. The question is always the same: build something custom because off-the-shelf cannot serve the specific need, or pick the best available product and configure it properly. Both are valid. Building for the sake of building is not.

A consultant who only ever recommends custom builds is running a build shop. A consultant who only ever recommends off-the-shelf is running a reseller business. A good one chooses deliberately, based on what the business actually needs.

4. They ship it — and make sure someone else can run it

The end state is not a deck. The end state is a tool that is live, documented, and something your team can operate without weekly calls to the consultant. If you cannot fire the consultant after 90 days without the project falling over, the engagement was a subscription, not a project.

What they don't do (and what you should not pay for)

A few things consultants often sell that you should push back on.

They don't write you a 60-page AI strategy

The 60-page strategy is the oldest trick in the consulting book. It looks thorough, it costs a lot, and it sits on a shared drive untouched for 18 months. A short decision memo — one page of opportunities ranked by impact and effort, one page of a 90-day plan — beats it every time.

They don't promise transformation

"AI transformation" is a phrase designed to sell six-figure retainers. Real AI projects are scoped, shipped, and measured in weeks, not transformation cycles. If someone uses the word transformation in their pitch, it usually means the engagement is too vague to fail cleanly.

They don't need to own your data forever

If the implementation requires your consultant to stay involved indefinitely so the tool keeps running, the tool is not really yours. Good engagements include a handover — credentials, documentation, and a runbook so your team can maintain it. The consultant should be leaving you stronger, not more dependent.

In our experience

Across the engagements we have run at pdconsults., the pattern holds: the biggest wins come from the projects nobody found exciting when we started. The most expensive failures come from the projects everyone got very excited about in the first meeting. Excitement is not a validation signal — it is noise.

The shortest path to a shipped AI project is also the most unglamorous one: understand the business, identify the one thing that will move a real number, build or buy the right tool, and hand it over working. Four phases. Measurable outcome. No deck required.

How to tell if you actually need one

Before you hire anyone, ask yourself three questions:

1. Do you already know what problem you want AI to solve, or do you need help finding it?
2. Do you have the internal capacity to implement the solution once it is designed, or do you need someone to ship it for you?
3. Can you clearly describe what "better" looks like in numbers after the project is done?

If you have clear answers to all three, you probably do not need a consultant — you need a builder. If you are unsure about any of them, that is the gap a good consultant fills. The assessment at the bottom of this page is a 10-minute way to find out where you actually sit.