Two companies we met with this week have a similar issue when it comes to time. They want to quickly introduce new technology across their organization to remain competitive externally but they first need to slow down and get their team on board internally:

“The masses are on Copilot, and then maybe 20-25 people on Claude. Nobody has done any training. That's a glaring gap that I've got on my list…And that goes beyond just the AI stuff. We typically don't train with new tools.”

- VP of AI, Chemicals Company

“We have all of our agents built on one person’s machine because we don’t want to, like, let anyone else use this until it’s safe for us and for them.”

- Director, Real Estate Company

Last week’s blog focused on healthy debates between the AI skeptics and enthusiasts. Talking is easy. This time we see the difficulties of simply kicking off an AI rollout, and the campaigning that goes on behind the scenes. 

Covering a full AI rollout would take pages, so we’re focusing on just the outset. We’ve seen this occur with companies handing out their first Claude licenses or releasing a database of admin-managed agents. It’s all the same.

In the beginning, developing trust is the only thing that matters and yet very hard. If the technology doesn’t work perfectly on an employee's first attempt, they’re as good as gone. And it’s not a fair fight because they will be looking to find flaws.

The obvious solution would be to make every tool bullet proof before launching, but it’s impossible to perfect without live testing. Something will inevitably break for people rooting for that exact outcome to happen so they can go back to “their way” of doing the job. That’s why the steps that come before this happens are so key.

We’ve seen three successful ways to manage expectations:

1. Townhalls: open forums to vent concerns and questions are far more compelling than emails.

2. Dashboards: starting with promptless tools minimizes the potential outcomes and ensures familiarity.

3. Template Queries: pre-set searches point employees in a direction that will not only help them but trust the results.

None of these have anything to do with improving the tool. Putting in the work upfront is meant to plant the seed. Ultimately it’s up to the end user to decide how much patience they will have while experimenting themselves.

So here’s the game plan: 

• Write out the top 5 prompts you want your team members to test. Ideally these will be prompts that aren’t commonly used today and get folks thinking about how to tweak their inputs to enhance AI output. Of course, review the outcome yourself to ensure that the query responses are accurate and helpful.

This is not your get out of jail free card. In fact, the effort needed to develop trust within new technology may never stop. But failing to do so at the start all but guarantees chaos.