The line I hear most from teams is, "Leadership says we have to start using AI." Right behind it: "We need AI. What's your plan?" It's a reasonable ask. It also fails in a predictable way. AI doesn't fix a weak foundation. It multiplies whatever it's given, good or bad.
Lately, many partners have asked me how AI should fit into their organization's processes. I've seen two approaches: some take the time to outline how things work now and how they should work with AI embedded for future growth. Others just jump in.
Here is what that looks like in practice.
A knowledge base nobody curated. When an agent searches for an article, either to help resolve an issue or to hand the requester something they can use to self-resolve, two things can go wrong. It finds overlapping articles, or it finds that none exists. Requesters get more to read and less help. Adding AI won't improve that experience. It runs into the same results. The fix isn't more AI. It's a hard look at the structure of your knowledge base.
A request catalog with one catch-all form. When there's only one option for requests, requesters leave it to the agent to figure out what they're asking for. It can also lead the agent to ask for details that a structured request type would have captured up front.
Categories that blur. Take a request like "my printer isn't working." It could be an application problem, a new setup, or something else, and each needs a different fix. When they all sit under one category, a human agent can use judgment to decide what to do next. An AI agent depends on the category to know.
Triage that goes in circles. Some processes look like triage but never deliver a resolution or tell the requester what happens next. The requester feels like they're going in circles, gets frustrated, and eventually gives up.
Routing is not deflection. Automation should move the process along, not add extra steps that nobody has defined. Deflection is about first-pass resolution, and that only works when intake is structured.
AI bolted on instead of built in. More adoption only makes a team more efficient when AI works as a layer in the process, not as an added step. Either way, the effect shows up in both cost and productivity.
What to do first
You don't have to rebuild everything. Here's a sequence I'd suggest. Your order may differ depending on what you already have in place.
- Clean and tag your knowledge base by request type, so search narrows instead of sprawls.
- Replace the catch-all with structured request types, so the AI knows what it's looking at.
- Define your categories clearly, so requests that look alike get the right path.
- Connect the agent to automation and give it tools, so it resolves requests and doesn't only describe them. Keep a human approval step on high-risk actions such as elevated access.
- Build an escalation path for when the agent can't finish, with full context handed to a named owner.
- Then scale AI usage, once answer quality is where you want it.
If a full rebuild isn't realistic, identify the highest-priority items that bring you closer to AI adoption and first-tier resolution. Work those, measure the results, and move down the list until you reach the outcome you're after.
The test
Before you go out and buy AI, ask: could a new hire do what I'm asking the AI to do, given how things are set up today?
If the answer is no, adding AI isn't the only thing you need to review.
If your team is being told to “just start using AI” and isn’t sure the foundation underneath it can support that, we’d love to talk. Praecipio helps organizations fix the knowledge base, request catalog, and categorization work that AI depends on, before scaling adoption on top of it.
This blog was written by Atiim Browne, Enterprise Solutions Architect at Praecipio.
