How to choose an AI consultant: a 2026 checklist

A real AI consultant ships a measured change into your business and proves people use it; a deck-seller hands you a strategy slide and a bill. The difference shows up in six things you can check before you sign: a named implementation they actually shipped, adoption numbers after go-live, vendor and method credentials, a scoped plan tied to your tools and data, honest pricing drivers, and a willingness to tell you not to hire them when an off-the-shelf tool already does the job.

We have run AI adoption inside large teams — including one enterprise program that put AI in front of 4,000 staff — so this checklist is written from the rollout side, not the sales side. Work through it in order.

The 2026 checklist: six things to verify before you sign

1. They can name an implementation they shipped — not just advised on

Ask one question: "Walk me through the last AI project you took from idea to people using it daily." A consultant who has done the work will describe the messy middle — the integration that broke, the workflow they redesigned, the team that resisted and why. A deck-seller will pivot to a framework. Strategy is cheap and AI strategy ages in months; shipped work is the only proof that holds.

What good looks like: a specific tool, a specific team, a before-and-after you can picture. If every answer is a maturity model, you are buying a workshop, not an outcome. This is the line between advising on AI adoption and implementation and merely presenting on it.

2. They measure adoption after go-live — and will show you numbers

Most AI projects don't fail at launch; they fail at week six, when the novelty wears off and people drift back to old habits. The consultants worth hiring know this and instrument for it. Ask what they track and what counts as success.

Concrete benchmarks help. In our enterprise program, weekly adoption sat at 72% and daily use at 46% at six months, with 98% training satisfaction. You don't need those exact figures — you need a consultant who talks in numbers like these rather than in "engagement" and "buy-in." If they can't tell you how they'd measure success, they can't deliver it.

3. Their credentials are real signals — necessary, not sufficient

Badges don't ship software, but they screen for seriousness. The ones that mean something in 2026:

• Vendor partnerships — a Claude partner relationship or OpenAI program membership means the firm has direct access, early features, and a support path when something breaks. (LOKAL holds a Claude Certified Architect credential and is part of the Claude partner network and the OpenAI Champions Network.)
• Change-management certification — Prosci and similar methods matter because adoption is a people problem, not a model problem.
• Operations and CRM tooling — HubSpot and automation-platform fluency, so the AI lands inside the systems you already run.
• Professional bodies — for context, we sit within PMAP and HRAP and use Prosci on the change side; bodies like these signal a firm operates by a standard, not a vibe.

Pair every badge with point 1. A certificate plus a shipped project is a signal; a certificate alone is a brochure.

4. They scope to your tools and data — before quoting a number

A consultant who quotes before understanding your stack is guessing. Good scoping starts with your reality: which tools you run, where your data lives, how clean it is, and which one workflow would pay back fastest. The plan should name the use case, the integration, the rollout, and the success metric — in that order.

Watch for the opposite tell: a fixed "AI transformation package" sold the same way to every client. Your bottleneck isn't the same as the next company's, so the plan shouldn't be either.

5. They're honest about what drives cost — and tell you to check current pricing

AI pricing has two layers, and a straight operator separates them for you. There's the consultant's fee, and there's the tooling you'll run afterwards. On tooling, the costs are knowable but they move, so check current pricing for each: Microsoft 365 Copilot is a paid add-on, ChatGPT and Claude have paid plans on top of their free tiers, and automation platforms like Zapier, Make, and n8n run from a free self-hosted option up to paid plans that scale with usage.

On the consulting fee itself, a consultant who won't talk openly about what makes a project cheaper or dearer — clean data and a narrow scope make it cheaper; legacy integrations and a big rollout make it dearer — is hiding the ball. We scope and quote on a call rather than name a number here.

6. They'll tell you when not to hire them

This is the strongest signal of all, and the rarest. An operator who turns down work they can't justify is one you can trust with the work they can. We say no when a client doesn't yet have a problem worth automating, or when an inexpensive off-the-shelf subscription already covers it. (More on exactly when to walk away below.)

Consultant vs in-house vs freelancer — at a glance

Three routes solve different problems. Most teams blend them over time: a consultant ships the first wins, a freelancer handles a one-off build, and an in-house hire takes over once the workload is steady.

Cost drivers are integration complexity, data readiness, and rollout size — not a published figure. We scope and quote on a call, and you should check current tool pricing before budgeting.

Adoption is the number that decides if it worked

Everything above ladders up to one outcome: people using the thing six months later. If you remember nothing else from this checklist, ask every consultant on your shortlist how they'll prove that — and compare their answer to these.

When NOT to hire an AI consultant

The honest answer most firms won't give you: sometimes you shouldn't. Skip the consultant — for now — if any of these is true:

• You don't have a clear problem yet. "We should do something with AI" isn't a brief. Name the workflow that's slow or expensive first, then decide who fixes it.
• An off-the-shelf tool already covers it. If an inexpensive off-the-shelf subscription solves the need, run a two-week pilot yourself before paying anyone to advise on it.
• Your data is too messy to act on. No consultant can build reliable AI on data you can't trust. Sometimes the right first project is cleaning the data, not adding AI.
• You only need one bounded build. A single script or integration is a freelancer job, not a consulting engagement.

Pilot first, then call someone when the gap is real — integration, change management, or scale. That's the point where a consultant stops being overhead and starts paying for itself. If you'd rather skip the off-the-shelf detour and have a team set up the automation cleanly from the start, that's a fair reason to bring one in early.

What this looks like in the Philippines and Australia

The checklist is the same in both markets; the buying context differs. In the Philippines, AI work usually pairs with a training spine, because the bottleneck is fluency as much as tooling — that's why our AI consulting in the Philippines bundles rollout with hands-on enablement.

In Australia, the questions skew toward governance, data residency and integration into established stacks. The two extra things to check on an Australian engagement: that the consultant works to a recognised change-management standard, and that they're comfortable with your compliance and privacy obligations before any data moves. If you want a build done cleanly and locally, our AI automation in Australia work and our AI consulting in Australia service are scoped for exactly that.