Skip to content
Dustin's AI Lab
Go back

What I only learned after taking on enterprise AI training—prep time, the client liaison, instructor flexibility, and how to pick a consultant

Four things I learned after running a handful of enterprise AI training sessions. Prep time is 3-5x the session itself. The client liaison decides everything. The instructor has to handle wildly different skill levels. And if you're looking to hire a trainer, the signals are like picking a pop star.


Four things I only learned after running a handful of enterprise AI training sessions:

1. Prep time will be 3-5x the session time, easily more

Especially if it’s not group work but “every individual gets a custom hands-on task,” that multiplier climbs. Per-person customization means:

So when you’re negotiating the contract, lock down the in-class exercise format up front and price prep time into it. If the client asks for “every person on a different task,” your quote needs to clearly separate “teaching hours” from “preparation hours.” The latter can have a lower unit rate, but it can’t be free.

What I do now: the contract states “prep time ≤ 4x teaching time,” with anything beyond that billed separately. Otherwise a 1.5-hour class can sink 8-12 hours of customization and cut your margin in half.

2. The client liaison determines whether the project flows

I got lucky—my first engagement was with a smart, open-minded, accommodating CEO. She reviewed the lesson plan herself, discussed it with me without friction. Aligning needs didn’t have to go through “I teach the liaison → liaison relays to the C-suite → C-suite disagrees → it filters back to me to redo.”

But apparently most engagements aren’t like that. They’re more often the kind where the client wants to pay money in exchange for an outcome, doesn’t want to be in the process, and only notices at delivery that it doesn’t match what they had in mind. So managing client expectations early is a job in itself.

Specifically:

3. The instructor has to handle wildly different skill levels and questions

The spread of skill levels in an enterprise class is much wider than in a public course. One session might include:

The capability requirement: you need to simultaneously hold “too easy is boring” and “too hard is unreachable.” My approach is tiered tasks: a base task everyone can finish, an advanced task that lets the high-skill folks stretch. For the mid-level manager’s question, I redirect with “let’s look at a market signal first” rather than debating their anxiety in front of the room.

4. If you want to hire an AI consultant, the signals look a lot like picking a pop star

If you’re the client looking for a trainer, here’s a contrarian tip:

Pick the one with good articulation, sharp thinking, enthusiasm for teaching, patience with students—but not much experience, not much fame, and not many active engagements.

No experience but high enthusiasm means they’ll burn extra time to solve your problems. They want to build milestones with these early engagements, so they’ll be more careful, more thorough.

It’s a bit like what a friend who’s into fandom told me—

If you want to follow an idol, find one who’s just debuted with no fame. They’ll dote on fans. The ROI is the highest.

The reverse holds too. If you’re a trainer just starting out and want to land enterprise gigs, your competitive edge is “I’ll spend three times the hours other people would on this one.” Put those hours into Precall, customization, and post-class follow-up. After three sessions, word of mouth will hand you the next three.

These are the four things I most wish someone had told me before I took my first enterprise engagement. AI training is currently a seller’s market, but seller’s markets don’t last forever. What lasts is the reputation of “this person is worth booking again.” That’s worth more than any single high-margin invoice.


Share this post on:

Previous Post
Two MCP servers for Taiwan public data—what mcp-taiwan-legal-db and Twinkle Hub are each good for
Next Post
How to tune Opus 4.7's effort, and why I still keep non-coding tasks on Sonnet 4.6 medium