Five places AI help starts at a benefits agency

1. Agency discovery and systems mapping

Agency discovery replaces a vague feeling that everyone’s busy with a one page map of where the hours go. Before, the owner knows the team is slammed but can’t say which handoff is eating the week.

In a 2 to 3 hour session, I look at who’s on staff, how the data flows, who touches the information, what you do to bring business in, and how that’s structured. I draw it live on a whiteboard, with cards for people, tools, and data and lines for the handoffs. Within one week you get the map, the one or two connections worth making, and a straight answer on whether I can help.

What it gives back is a number. Manual data entry is what turns up most. “It’s shocking how many agencies are still typing things in by hand.” More in [business process mapping](/articles/business-process-mapping).

2. Software selection and setup

Software selection means picking the tools that fit how your agency already works, then connecting them with limits. Before, a new tool gets bought, half set up, and abandoned.

I build most things myself, and I’ll still set up an off the shelf product when it’s the right answer. Setup is where the snags hide. When Claude gets connected to a Google Workspace Gmail account, the admin console flags it as a “risky app,” and a yellow alert shows on the profile photo. I warn the admin during setup so nobody reads it as a break in. More in [insurance agency software selection](/articles/insurance-agency-software-selection).

3. AI training for teams

AI training for teams teaches people to hand AI a job instead of a question. Before, the team treats it as “a glorified Google search.” They upload documents into the chat, copy the answer, paste it into another system, and edit it there. That’s the biggest learning curve I see.

After, they tell their agent to go work on something on the computer. Every client gets the AI Basics lesson library at dobetter.plus/concepts: 30 lessons of about 2 minutes each, from Chat vs Agent to What Not to Paste In. The first 3 are open to anyone. A small team workshop runs 2 to 3 hours on your team’s own tasks. More in [AI training for employees](/articles/ai-training-for-employees).

4. AI use policy

An AI use policy is a short set of written rules for what goes into AI and what a person checks before it goes out. The 2026 Big “I” technology survey found 55% of independent agencies don’t have one.

I haven’t written one for a client yet. I’d write yours from the rules I run on myself: agents draft, and a person sends. Nothing gets saved to long term memory without my approval. The last chapter of the lesson library covers the same ground: what not to paste in, who can see the chat, and when to stop and ask a human. More in [AI acceptable use policy](/articles/ai-acceptable-use-policy).

5. Ongoing support and builds

Ongoing support keeps what was built working and adds the next build when you’re ready. Before, the person who set it up leaves and nobody knows how it works.

While I’m retained, I’m available to fix anything that breaks, and it rarely happens. What I’d prefer is to teach someone at your agency to have their agent fix what breaks. A connected workflow project includes testing, training, and 30 days of stabilization. More in [AI support after the build](/articles/ai-support-after-the-build).

Where should I start if I can only pick one?

Start with discovery. Every other service depends on knowing where the retyping happens and who does it. If you already know the problem is your team using AI like a search engine, start with training. For a broader starting point, read [what should my agency automate first?](/articles/what-should-my-agency-automate-first)

Quick answers

Do I need new software first?

No. Discovery starts with what you pay for today.

Can training happen without a build?

Yes. The workshop and lesson library stand on their own.

What does discovery cost?

$1,500, half credited to month one, full refund if I can’t help.

Is support hourly?

Never.