Nobody Rolled It Out

ASA now holds six Applied AI use case records from member companies. Not one of them describes a rollout.

The records cover a manufacturer automating a monthly rep agency scorecard, a distributor putting product and spec answers in front of counter staff, an outside sales group trying to get ahead of quiet account churn its CRM was not surfacing, a manufacturer replacing Friday morning CRM entry, and a technical knowledge assistant still sitting in test. Different segments, different functional areas, different tools. In every case the work started small enough that you could name the people involved.

The records contain no company-wide launch, no go-live date, no training curriculum, no phased deployment schedule. Where a record says how the work began, it began with volunteers you could fit around one conference table. One describes a pilot group of about eight people. Another describes a single person and a single recurring task.

The scorecard

The clearest example is a manufacturer's marketing team assembled a monthly scorecard for its rep agencies by opening a stack of Excel files and pulling the numbers out by hand. Three to six hours, every month, forever. No review step, because the person doing it was the only one who understood the assembly.

The workflow owner was hesitant. Not resistant to AI. She did not know where to start and had no room to find out.

What moved it was not a tool decision. Her manager took other projects off her plate to make space for the work.

The record says it plainly: asking employees to adopt AI on top of everything else is unlikely to succeed.

Read that as a staffing statement rather than an AI statement. The binding constraint was never model capability, licensing, or integration. It was one person's calendar, and somebody senior enough had to change it.

Where the tools stop

The second pattern is about the boundary, and it is the same boundary in every record.

The human keeps the customer. AI drafts, retrieves, flags, and assembles. A person validates it, prices it, quotes it, and has the conversation. One distributor's assistant gathers information and hands off, and the order and the customer contact stay with the counter rep. One outside sales implementation informs reps and branch managers and touches no customer directly. The automated scorecard carries a mandatory human review before anything material leaves the building.

Nobody in the channel has put AI in front of a contractor. Not because they cannot. They decided not to yet, and they said so on the record.

The position

If your AI plan has a go-live date and a user count, you have a software project. What the channel's own record shows is that adoption is a staffing decision, and the scarce input is not budget. It is somebody's Tuesday.

That runs against how the money moves. A platform decision is a line item. Somebody can approve it in a quarter and it reads cleanly in a capital request. Five volunteers and a manager clearing a calendar is not a line item and never will be. So the license shows up in March, the first real use of it shows up in September, and in between nobody gave anyone the time.

Where small starts fail

Starting small is not a strategy by itself, and two of the six records show the gap.

One knowledge assistant is built, loaded, and running in test. The member has not yet decided who verifies an answer before it reaches a customer, or who escalates when the answer is wrong. Another record documents meeting notes flowing into a CRM automatically, and does not say whether reps review what got written.

A small pilot buys a real answer about whether the thing works. It does not excuse anyone from writing down who owns the output. That step gets skipped most often, because it feels like paperwork right up to the morning an unverified spec answer reaches a contractor.

What ASA is doing with this

We are collecting these records, not grading them. Nobody gets a score, and the repository is not a vendor list. A distributor deciding what to try next quarter should be able to read what other member companies actually did, including the parts still unresolved.

If you are running something, we want it in the repository. If you are deciding what to run, the applied AI session at the ASA Innovation Summit sits on the same seam. Steve Lerch is covering how distributors move from experimentation to execution, and where experimentation is running without operational grounding.

The Summit runs November 9 and 10 at Caesars Palace in Las Vegas, immediately ahead of NETWORK2026 in the same hotel. It is $250 added to a NETWORK2026 registration and $500 on its own, with a full refund for written cancellations through October 9. Details at asa.net/Innovation-Summit.

Bring the person who would run the pilot, not only the person who would approve it. On the evidence so far, the approver's real job is clearing three hours a week off the other one's calendar.

By Nils Swenson