TL;DR. An AI implementation consultant gets AI working in your real systems on your real data, and keeps it working. Hire for seven skills: test sets from real cases, integration, data handling, model choice on accuracy and cost, human review design, monitoring and handover. Ask what they will deliver, and ask questions only people who have shipped can answer. Strategy without these skills ends in a plan, not a system.

An AI implementation consultant is the person, or small team, who takes an AI idea and makes it work in production: inside your real systems, on your messy real data, with a person reviewing the right steps, and with monitoring so it keeps working after they leave. When you hire one, hire for the ability to test, integrate and hand over, not for the quality of the slides.

I run Skcript Applied Labs. Implementation is what we sell, so read this with that in mind. I've tried to write the job description I would want a buyer to use, even when it rules us out.

What does an AI implementation consultant actually do?

The work has a recognisable shape, whoever does it.

  1. Measure the task as it runs today. Time per item, volume, error rate, cost. Without a starting number, nobody can prove the system helped.
  2. Build a test set from real cases. A hundred or more of your actual inputs with the right answers attached, messy ones included.
  3. Test several models on it. Accuracy, cost per task and where data is processed, on one scorecard.
  4. Build the first working piece inside your tools. Connected to the systems where the work arrives and where the answer has to land.
  5. Design the review step. A person approves what matters. Cases the system was not tested on go to a person.
  6. Run it on real work and measure again, the same way as at the start.
  7. Set up monitoring and hand over. So your team can see when it drifts, switch it off, and fix it.

If a proposal skips steps one, two or seven, it is a prototype, not an implementation.

How is implementation different from AI strategy consulting?

Strategy decides what to do and why. Implementation makes one of those things work. Both have value, but they need different people.

A strategy engagement ends with a roadmap: which areas to focus on, how to organise, what to buy. An implementation engagement ends with a system people use, before and after numbers, and a team that can run it.

The common failure is hiring strategy people to implement. You get a good plan and a demo on sample data, then the real work starts with someone else. If you already know which task you want done, you need implementation skills from day one. If you don't know which task, read do you need an AI consultant before hiring either kind.

What skills should you hire for?

These seven separate people who ship from people who present.

1. Building test sets from real work. The most important skill and the rarest. Can they turn your messy cases into a set with right answers, and use it to decide? We explain our approach in how we test AI models on your work.

2. Integration engineering. Most of the effort in a real system is not the model. It is reading from email, an ERP, a document store or an old tool with no API, and writing back safely. Look for engineers who have connected to systems like yours.

3. Data handling. Knowing where data is stored and where it is processed, which are not always the same place. Using business plans that do not train on your data. Keeping personal data to what the task needs. Working inside your security rules.

4. Model choice on evidence. Comparing models from more than one provider, including open models, on your cases, on accuracy and cost per task. Someone tied to one vendor can't do this honestly.

5. Human review design. Deciding which steps a person approves, what happens to cases outside the test set, and how the system says "I'm not sure." This is what makes a system trustworthy enough to use.

6. Monitoring. Watching accuracy and cost on live work, catching drift when inputs change, and testing new models as they ship. Models change. Providers retire versions. Something has to notice.

7. Handover. A runbook, documentation and a named owner on your side who knows how to check it and what to do when something looks wrong. If the system only works while the consultant is around, it does not work.

What should an AI implementation consultant deliver?

Ask for these in writing, before you sign.

  • A working system inside the tools your team already uses.
  • A test set from your real work, which you keep and can use to check any future tool.
  • A model scorecard: accuracy, cost per task and processing location for each model tested.
  • Before and after numbers, measured the same way.
  • A review step for the cases that matter, and for cases the system was not tested on.
  • Monitoring, and a way to switch it off.
  • A runbook and a handover your team can own.
  • Clear ownership. The code, the test cases and the numbers should be yours.

Which questions reveal real implementation experience?

People who have shipped answer these with specifics. People who have not answer with process.

  • "How will you know it is accurate enough?" Listen for a test set from your cases and a line agreed in advance.
  • "What happens to a case the system has never seen?" Listen for a person getting it, not "the model handles it."
  • "Which models will you compare, and how?" Listen for more than one provider, on the same cases.
  • "Where is our data processed, not just stored?" Listen for how they find out and put it in writing.
  • "What broke on your last project, and what did you change?" Anyone who has shipped has an answer.
  • "How do we switch models next year?" Listen for a design that makes it routine.
  • "Who on this call writes the code?" The best answer is "me."

My guide to choosing an AI consultant has more questions and the red flags to walk away from.

Should you hire a consultant or a full-time engineer?

Hire a full-time engineer when AI is core to your product, or you have years of steady AI work ahead. Bring in an implementation consultant when you need one task working soon, want numbers before you hire, or don't yet know which skills your team will need. A good consultant leaves behind exactly the thing a new hire needs: a test set, a scorecard, a runbook and a working system to learn from. I compare consultants, agencies and in-house teams in detail in AI consultant vs AI agency vs in-house.

What are the warning signs?

  • Demos on clean sample data. Real inputs are messier. Ask to see it on yours.
  • One model for everything. Often a sign of a vendor relationship, not a test.
  • No starting measurement. Without one, success becomes a feeling.
  • No answer on handover. If they can't describe what you will run without them, plan to pay them forever.
  • Open-ended discovery. Implementation can start small and fixed. A long discovery with no end date is a cost, not a phase.

How do you measure whether the implementation worked?

The same way it was measured at the start. Time per item, volume handled, error rate and cost, before and after, on real work. Add cost per task for the model, and the share of cases sent to a person.

Then keep measuring. A system that was right on day one can drift as inputs change. The test set you built at the start is how you check, every time anything changes. What happens between a pilot and a live system is covered in the AI pilot to production playbook.

How should you start?

With one task, a fixed price and numbers at the end. Not with a hire for a year, and not with a programme. Pick the task that eats your team's week, ask for a defined first step that measures it, tests models on it and builds the first working piece, and agree the price to go live before you commit.

That is how our 10-day AI pilot works. Prices for every market are on our AI consulting pricing page, and what drives cost across the market is in how much does AI consulting cost in 2026. The full offer is on our AI consulting page.