What does an AI consultant do, and do you need one?

An AI consultant helps a company decide where AI is worth using, proves it on a small piece of real work, and leaves the company able to run what was built. A good one delivers a decision, a tested pilot and a handover. A weak one delivers a slide deck and a list of tools.

Whether you need one depends less on how much AI you want and more on who will run it afterward. This guide is for owners and IT leaders who are deciding whether to hire help or do the work with their own team.

What should an AI consultant actually deliver?

Most AI projects do not fail on the model. They stall between a promising demo and daily use. S&P Global's 451 Research found that the average organization scrapped 46% of its AI proofs of concept before they reached production (a survey of 1,006 IT and business professionals, published October 2025). That is an average share of projects, not the share of companies that failed.

BCG points to where the trouble usually sits. In its October 2024 study of 1,000 executives, it traced about 70% of AI challenges to people and process, about 20% to technology and only about 10% to the algorithms. Tool selection alone does not address the people and process work.

So judge an engagement by what you hold at the end:

• A short list of use cases ranked by business value and readiness, including the ones rejected and why.
• A data and access review: which documents and systems the AI may use, who can see what, and what stays out.
• A pilot scope with a baseline, acceptance criteria and a named decision-maker.
• A working pilot on real work, not a demo on sample data.
• Training for the people who will use it and for the people who will run it.
• A handover pack with a named owner for each part.

When can your internal team do the work?

More often than consultants admit. If the goal is to get staff using AI that is already in software you pay for, your own team can usually lead it.

Take a 120-person firm that already licenses an AI assistant inside its office suite. It needs an acceptable-use policy, a short training session per department, a few shared prompts for common tasks and someone who answers questions for the first month. That is ordinary rollout work. Paying a consultant to design it is rarely the best use of the budget.

The policy matters more than it looks. IBM's Cost of a Data Breach Report 2025, based on 600 breached organizations, found that one in five reported a breach due to shadow AI: AI tools staff used without approval. An approved tool and clear rules are work an internal team can do well.

Your team can usually handle it when:

• The use case sits inside one department and one system.
• Someone has the time, not just the interest, to own it for the first few months.
• Your documents are in reasonable shape, and someone knows where they are.
• Mistakes are cheap: a wrong draft, not a wrong payment.

Bring in help when the work crosses systems, touches sensitive data, needs actions rather than answers, or nobody inside has time to own it.

When should you hire a consultant, an implementer or an MSP?

These are different jobs, though many firms offer more than one.

There is some evidence that outside help raises the odds of reaching deployment. MIT NANDA's The GenAI Divide: State of AI in Business 2025 (July 2025) reported that tools built with external partners reached deployment about 67% of the time, against about 33% for internal builds. Read that with care. The figures come from interviews with 52 organizations, the outcomes are self-reported, and the authors say the result is a correlation that does not prove cause. Companies that hire partners may differ in budget, risk appetite or skills. The link goes to a copy hosted by a third party, because MIT restricts access to the original.

What belongs in the first pilot's scope?

One problem, one group of users, one owner. A good first pilot is small enough to finish and real enough to matter.

Here is an illustrative IT example. Employees ask in Teams or Slack how to get approved software. The assistant answers from the company's software policy and shows the source. If the employee still needs the software, the request becomes a ticket and a named approver says yes or no. After approval, a tested automation installs it on the employee's device. If anything fails, the ticket goes to a person on the IT team, who sees what the assistant already tried. Success is measured against the weeks before the pilot: how long employees waited, how many requests IT handled by hand, and whether the software worked afterward.

Use this scope-and-handover checklist before you sign, and again before the pilot ends:

• Problem and baseline. The one problem, who has it, and how it is measured today.
• Users. The group in the pilot and the channel they will use.
• Sources. The documents and systems the AI may read, and who keeps them current.
• Actions. Each action the AI may take, the system it touches and the account it uses.
• Approvals. Which actions need a person to say yes, and who that person is.
• Handoff. When the AI must stop, who takes over and during which hours.
• Acceptance. The result that counts as success, seen by the business, not just a log showing the automation ran.
• Exit. What happens if the pilot misses its target: change it, extend it or stop.
• Owner after launch. The person or provider who runs the service, updates sources and retests actions.
• Handover pack. Configuration, credential ownership, documentation and training for that owner.

A fuller working version is in the AI implementation playbook.

What evidence should you request before signing?

• A finished project similar to yours, or a past client you can call.
• Real deliverables from a past engagement, redacted: a scope, an acceptance plan, a handover document.
• The named person who will do the work, and their hours on your project.
• How they handle your data: where it goes, which AI services see it and what is kept.
• What they would not recommend for you. A consultant who says yes to everything has not looked closely.
• Any stake in the tools they recommend. Referral fees and resale margins are common; ask for them in writing.

Who owns the service after launch?

Decide this before the pilot, not after. AI assistants and automations need upkeep: documents change, applications change their screens and APIs, staff leave, and approvers go on holiday. Someone has to notice and fix those things.

On September 29, 2026, Gartner predicted that 70% of enterprises will abandon agentic AI built by vendors' forward-deployed engineers by 2028. Press coverage of the forecast gives the reason: those companies cannot run or change the systems themselves. It is a forecast, not a measured result. The lesson applies to any outside builder: a good engagement ends with your team, or a provider you choose, able to own and change what was built.

In practice the owner is your internal IT team, the MSP that already supports you, or the consultant on an ongoing contract. Any of them can work. What fails is leaving the question open.

What determines the cost?

There is no reliable standard rate. These are what move the price:

• Scope. One workflow in one department costs less than a company-wide program.
• Data condition. Scattered or outdated documents add cleanup before anything else.
• Integration. Answering questions costs less than taking actions in other systems.
• Risk and review. Sensitive data, approvals and security review add time.
• Pricing model. Hourly, fixed-fee pilot, monthly retainer or outcome-based.
• Running costs. Software, AI usage and the owner's time after launch. Ask for these in the proposal.

Where does NetZen fit, and where does it not?

NetZen AI is a service desk, RMM and automation platform with AI built in. It fits when the pilot is IT help for employees. The AI answers from your company's knowledge and shows its sources, runs automations in a browser, through an API or on a device, can require approval from a person you name before acting on a ticket, and hands the ticket to a person when it cannot finish. Employees reach it in Teams, Slack or a portal, and it connects to ChatGPT, Claude and Copilot Studio. The list price is US$100 per technician per month.

It does not fit model development, company-wide AI strategy, or data-loss prevention and compliance enforcement. Adopting it also means moving your service desk to NetZen; it is not a layer on top of another ticketing system. If your IT team will run it, see NetZen for internal IT.

Questions buyers ask

For AI consultants: the NetZen Partner Program

If you are a consultant, these are the questions your clients will ask you. The NetZen Partner Program is for consultants who lead a client's AI decision and want a named operator running the result. You lead the client; NetZen supports the demo, the proposal and the first rollout. Partners earn 15% of the NetZen subscription each client pays, for 24 months. See how the work is scoped and handed over in the partner playbook.

Research sources are linked beside the claims they support. Survey and interview figures describe the organizations studied and are not NetZen results. Gartner figures are forecasts.