3 min read
“AI consultant” went from a rare title to a common one in about eighteen months. Some of the people using it build working systems. Some sell a subscription to a tool they resell. From the outside, the pitch sounds the same — so it helps to know what the job actually is.
The job, in one sentence
An AI integration consultant connects AI models to the systems a business already runs — the CRM, the inbox, the website, the booking system, the product catalog — so the model does useful work inside them, under rules about what data it can see and what it is allowed to do. See AI integration.
The model is the easy part. The work is the plumbing around it: permissions, data mapping, error handling, logging, review steps, documentation and training. That is also why the job overlaps so heavily with marketing operations, CRM administration and web development, and why it tends to go badly when it is treated as a software purchase.
What you should get
- A map of how work moves today, and where it leaks.
- A short, ranked list of opportunities, each with a baseline and the ones not worth automating named plainly.
- A written data and use policy before anything touches client data — see AI governance.
- The build itself, usually on tools you already pay for.
- Documentation and training so your team can run it, change it, and turn it off.
- A measurement plan at 30, 60 and 90 days, against the baseline.
Why standards are making this cheaper
Connecting AI to business systems used to mean custom code for every pairing. Open standards are changing that. The Model Context Protocol, introduced by Anthropic in 2024 and now governed under the Linux Foundation, gives AI assistants a common way to use tools and data; it reported more than 97 million monthly SDK downloads by December 2025. In commerce, the Universal Commerce Protocol and the Agentic Commerce Protocol do something similar for shopping. Less glue code means more of a consultant’s time goes on the parts that are actually specific to you.
Five questions to ask before you hire one
- What will you measure, and what is the number today? If there is no baseline, there will be no result — only a feeling.
- Where does our data go, and who can see it? The answer should name the tools, the accounts and the retention settings, not reassure you in general terms.
- What happens when it’s wrong? Listen for review steps, limits, logging and an off switch. See human-in-the-loop.
- Who owns the accounts, prompts and workflows if we part ways? The right answer is you, in writing.
- What would you tell us not to automate? Someone with no answer to this is selling, not advising.
Red flags
- A guaranteed return on investment, or a specific revenue multiple.
- A new platform proposed before anyone has looked at your current one.
- An agent that will “run the business,” or “replace your team.”
- No written policy on client data, especially in health, legal or finance.
- A demo built on sample data that never touches your real systems.
How we got here
Found First started in AI search visibility — making sure a business is the one named when a customer asks an assistant who to hire. That work is about making a business readable by machines. AI integration is the same idea turned inward: making the business’s own operations readable and workable by machines, with people still in charge of the decisions. The standard is the same either way — built to be read by machines, run by your team.
Sources: MCP blog, MCP joins the Agentic AI Foundation, 9 December 2025 · Google, Universal Commerce Protocol, 11 Jan 2026 · Stripe, Agentic Commerce Protocol, 29 Sep 2025.
Definitions: AI integration, AI agent, Model Context Protocol, AI governance. Or see what Found First does — every engagement starts with a map, not a tool.