What does an AI implementation consultant do?

They turn an identified use case into a working system that people use every day. That covers understanding the process, preparing data, building and testing the solution, connecting it to existing systems, training users and handing it over. The emphasis is on delivery in production, where a strategy consultant stops at recommendations.

What is the difference between an AI strategy consultant and an AI implementation consultant?

A strategy consultant advises on where and whether to use AI and produces a plan. An implementation consultant builds and deploys a specific solution and is accountable for it working. Many engagements need a little of the first and a lot of the second.

How long does an AI implementation engagement take?

For a single well-chosen process, three to five months is typical: two weeks of discovery, four to eight weeks of pilot, six to eight weeks of rollout and a short handover. Anyone promising production use in two weeks is describing a demo.

How should I evaluate an AI implementation consultant?

Ask how they will measure success before they start, what happens when the system is wrong, who pays the running costs, and what you will own at the end. Ask for a reference from a client whose system has been in daily use for at least six months. Strong answers are specific and include things that went badly.

Should my agency hire an AI implementation consultant or partner with one?

Partner first if you have fewer than two or three AI projects signed, since a full-time specialist is expensive to keep idle. Hire once the work is steady enough to fill most of a year. Use the partnership period to train your own engineers alongside the specialist.

AI implementation consultant: what the role is and how to judge one

The title “AI consultant” covers everyone from a slide-deck strategist to a prompt hobbyist. The implementation consultant is the specific one: the person who makes a system work in a real business, with messy data, existing software and staff who did not ask for it.

I have hired these people, been one, and partnered with them from inside an agency. This page covers the role, the shape of a good engagement, what it costs, and how to tell a good one from a confident one. It is part of the AI hub.

What the role is

An AI implementation consultant owns the path from “this process could be improved with AI” to “this is how we do it now”. The work breaks down roughly like this:

Activity	Share of the engagement
Understanding the process and the people	15%
Getting, cleaning and understanding the data	20%
Building the solution itself	20%
Testing and evaluation	15%
Integration with existing systems	20%
Training, documentation and handover	10%

These shares are my rough experience across projects, and they shift from one to the next. The point they make holds steady: building the AI part is about a fifth of the job. The rest is the ordinary, difficult work of changing how a business operates. That is why strong implementation consultants tend to come from software delivery or operations, and why pure model expertise is rarely enough.

The role differs from a strategy consultant, who stops at recommendations, and from a contract engineer, who builds what they are told. The implementation consultant is accountable for the result being used.

What a good engagement looks like

Stage 1: discovery

One to two weeks. The consultant watches the process being done, interviews the people who do it, and looks at real examples, including the awkward ones. They establish a baseline: how long it takes now, how many errors, what it costs.

Output: a short document stating the use case, the baseline, the proposed approach, the measure of success, and the risks. A good discovery sometimes ends with “do not do this”, and a good consultant is willing to say so.

Without a baseline, nobody can prove the project worked. Insist on it.

Stage 2: pilot

Four to eight weeks. A working version runs against real data, measured on a test set the client helped choose. It is used by a small group, in parallel with the existing process.

Output: measured results against the pass mark, a list of failure cases with their causes, an estimate of running costs at full volume, and a go or no-go recommendation.

The failure list is the most useful thing in the pilot. It tells you what the system should refuse to handle and pass to a person.

Stage 3: rollout

Six to eight weeks. The system is connected to production systems, with access controls, logging, monitoring and a route for people to flag wrong output. Users are trained. The old process is retired in steps, with the option to fall back.

Output: the system in daily use, with a dashboard or regular report showing volume, accuracy and exceptions.

Rollout is where projects stall, usually for human reasons. Staff who fear for their jobs find reasons the tool cannot be trusted. A good consultant spends real time here and involves the team from discovery onward.

Stage 4: handover

One to two weeks. Documentation, a runbook for incidents, the test set and instructions for rerunning it, and training for whoever will own the system. All accounts, code and data are in the client’s name.

Output: a client who can run, monitor and make small changes to the system without the consultant.

A consultant who leaves the client dependent has failed at this stage, however good the system. Ongoing support should be a choice the client makes, usually as a retainer, with the option to walk away.

Day rates, as worked examples

Rates vary by market and experience, so here is how they are built. Then you can check any quote against the arithmetic.

An independent consultant.

Income target: $180,000
Business costs: $30,000
Revenue needed: $210,000
Working days: 46 weeks x 5 = 230
Billable at 60%: 138 days
Day rate: $210,000 divided by 138 = about $1,520. Call it $1,500.

A consultant supplied by a firm.

Fully loaded cost of a senior consultant: $200,000 a year
Billable days: 180
Cost per billable day: $200,000 divided by 180 = about $1,110
At a 50% gross margin, the firm charges about $2,200 a day

The firm costs more per day and brings backup if the person is ill, a second reviewer, insurance and a contract with a company behind it. Whether that is worth $700 a day depends on how critical the system is.

A full engagement at the independent rate.

Stage	Days	Fee at $1,500
Discovery	8	$12,000
Pilot	25	$37,500
Rollout	30	$45,000
Handover	5	$7,500
Total	68	$102,000

Most good consultants will quote each stage as a fixed fee, with the client free to stop after any stage. That structure is a sign of confidence. An open-ended day-rate arrangement with no stages is a sign of the opposite. The reasoning behind fixed fees is in value-based pricing.

Compare the total to the baseline from discovery. If the process costs the client $400,000 a year in staff time and the system removes half of it, $102,000 plus running costs pays back in well under a year. If the process costs $60,000 a year, the project should not happen at this scale, and an honest consultant will say so in stage one.

How a buyer should evaluate one

Seven questions. The answers separate people who have delivered from people who have read about it.

“How will we know it worked?” Good: a baseline, a test set and a number. Poor: talk of transformation.
“Tell me about a project that went badly.” Good: a specific story about data or adoption and what they changed afterward. Poor: nothing ever went badly.
“What happens when the system is wrong?” Good: confidence thresholds, human review for costly decisions, a feedback route. Poor: assurances that it is very accurate.
“What will it cost to run each month?” Good: an estimate tied to volume, to be refined in the pilot. Poor: a shrug.
“What do we own at the end?” Good: all code, configuration, test data and accounts, in your name. Poor: a subscription to their platform with no exit.
“Where does our data go?” Good: a precise list of services and locations, and what is retained. Poor: vagueness.
“Can I speak to a client whose system has been live for six months?” Good: yes, here is the contact. Poor: only recent pilots.

Check the paperwork as well. The agreement should state acceptance criteria, data handling, IP ownership and a liability cap. If the consultant has no contract of their own, that tells you how many engagements they have completed. A sound starting point is a consulting agreement template.

Red flags

A fixed promise of accuracy before seeing your data
No discovery stage, straight to building
A proprietary platform you cannot leave
Day rates with no stages and no end date
A proposal that names more tools than business outcomes
No mention of the people who do the process today

When an agency should partner for this

If you run an agency, clients are asking you for this work. You have three ways to supply it: train your own engineers, hire a specialist, or partner with one.

Run the numbers on hiring first. A senior implementation specialist at a fully loaded $200,000 needs about 110 billable days at $1,800 just to cover their own cost, before any contribution to overhead. If you have one AI project signed and two hopeful conversations, you are betting on a pipeline that does not exist yet. An idle specialist is an expensive addition to the bench, a problem I cover in bench management.

Partnering fits the early stage. You keep the client relationship and the contract. The partner supplies the specialist delivery under your name. Your engineers work alongside and learn. When you have enough signed work to fill most of a specialist’s year, you hire, and by then you know what good looks like.

Partnering has costs. Margin is thinner, you depend on someone else’s availability, and you must be clear in the subcontract about IP, confidentiality and who talks to the client. The full comparison, with the break-even arithmetic, is in when to partner instead of hire.

Where this role is heading

Demand for implementation is growing faster than supply, because almost every mid-sized business has been told to adopt AI and almost none has the people to do it. The role will get more standardized as patterns settle, and prices for routine use cases will fall. The durable part is the stubborn middle: understanding a specific business well enough to change how it works.

If you want to move into the role yourself, the route is in how to become an AI consultant. If you want to build a line of business around it, see the AI consulting business. The reason this work is a natural next step for software firms is the subject of the pillar, is software engineering dead.