Making the same work faster is not enough

Making the same work faster is not enough if your team still has to work the old way afterward. We ran into that at Berry when AI evaluated more than two hundred sales meetings a month, but nobody wanted to read all the reports, so managers kept using sales numbers to decide who needed help.

We had made the evaluations faster without changing how managers used them. The bigger gain came from deciding what work they could stop doing and what we still needed them to do with the salesperson.

The reports were done. The manager's work wasn't.

The first version did what we asked it to do. AI read the sales call transcripts and scored each one against a rubric with roughly twenty-five points. By the end of the month we had so much information that it was hard to see who was doing well and who was not.

So managers were still looking at sales numbers, finding the lowest performers, and working with them first. That was our previous method, and it took too long to fix problems because we were finding them through the sales results.

I would not call that resistance to AI. We were asking managers to read all those reports and work out who needed their attention, and the system had not made that part useful yet.

Being able to evaluate all those calls was useful, of course. But we still needed to turn the evaluations into something the manager could act on, otherwise we had done one part faster and left them with all the work that came after it.

If you are the owner, that part is yours to work out. At Berry, we had to change what the manager received and what they did with it, because asking managers to read more reports was not going to solve the problem we had created.

We changed what the manager received.

We did not throw away the detailed analysis. The rubric still ran on the calls, and those evaluations stayed in the system as input.

What changed was the next step. We had AI read through the detailed reports and pull out a short coaching list: which salesperson the manager should work with, and which points to work on with each one. The opportunities that looked easiest to get results from sat at the top, highest gain first, in the order we cared about.

The manager no longer had to start with the whole pile and hunt for priorities. And instead of waiting for low sales numbers to tell us someone needed help, we could catch what they were doing in the calls much sooner. With AI, the feedback was almost real-time.

For example, we had a series of qualification questions salespeople were supposed to ask. Some were not asking them properly or in the right sequence, so they were losing the script and the ability to take the potential client through the conversation the way we intended. That hurt their ability to close the sale.

We could catch that much faster and help them practice the questions. A low sales number told the manager who was struggling, but this gave them a specific part of the conversation to work on.

Getting to that useful list took trial and error. We still had to work out whether the suggestions were right before asking managers to rely on them.

Checking was part of getting there, not the final job.

For a while the managers still checked. They took the first ten results on the list and went through the suggestions with each salesperson involved. Is this correct? Did this really happen? People's feedback helped us refine the process, and we adjusted from what we heard.

In the end they pretty much stopped checking on their own. The evaluations were correct ninety-nine percent of the time, and the managers used the list to start conversations and run coaching from there. It became very useful.

If we had kept asking managers to check every evaluation, we would still have left them with a lot of the work we were trying to take away. As the evaluations became dependable, that part of their job largely went away.

The manager still had to help the person improve.

Even after the list was trusted, the manager's job did not disappear. One thing is knowing what went right or wrong in sales technique. Another thing is connecting that finding to our way of working at Berry, and then communicating it in a way that helps and motivates that salesperson instead of causing more problems or hitting the wrong spot.

That was the hardest part, and it stayed with the manager. Knowing that someone had missed qualification questions gave us a place to start, but we still had to help them use those questions properly in a conversation.

We used simulations and role-play, and we gave salespeople a script so they could practice with each other. The AI finding could lead to someone practicing the part of the call they needed to improve, rather than another report for the manager to read.

That is where we drew the line in this process. We largely stopped asking managers to repeat the evaluation, and kept them involved in helping the person change how they worked.

Decide what people will stop doing.

If you have introduced AI into a process, ask the person receiving its output to walk you through what they do next. Are they using it to act sooner, or are they still doing the old work and now have an extra report to read?

At first, our managers still depended on the sales numbers to figure out who needed help. The evaluations existed, but they had not replaced that slower way of finding problems. That is the difference I would look for before asking the system to produce more.

And this took validation and refinement. If people still need to correct the evaluations, you cannot simply take checking out of their job, and a different kind of work may need continued review. Working out that balance is still hard.

So work through the change with the person who has to use the result. Decide what they can stop reading or rechecking once it is dependable, and what you need them to do with the information instead. In a coaching process like ours, that includes making room for the conversation and the practice, because finding a mistake earlier only helps if someone works on it.