Why build prescriptive alerts instead of another dashboard?

The first time the veterans saw it, they shrugged. "Yep, I know this already." Until it started to flip.

A multibillion-dollar building products distributor deployed AI to more than two thousand territory managers and returned roughly a quarter of a million hours a year, doubling rep productivity inside six months. When it turned to branch managers in 2026 it refused to build a dashboard, on the grounds that consolidating systems into one view encodes a human judgment about what matters and then inherits that bias permanently.

Industry

Building products distribution

Who uses it

Territory and branch managers

What it does

Prescriptive operational alerting

Measured

First six months, annualized

"The initial iteration is like, yep, I know this already, I know this already," said the company's chief digital officer. "Until it started to flip."

That flip, the moment an experienced operator stops dismissing what the system tells him and starts paying attention, is what two years of engineering have been aimed at. Not accuracy. Not coverage. The flip.

The company's first deployment with InstaLILY went to more than two thousand territory managers, and the results were the kind that get put on a slide. Roughly a quarter of a million hours a year returned to the business, rep productivity doubled inside six months, double-digit reduction in overhead.

In 2026 the company turned to a harder job: the branch manager.

A branch manager runs a small business. Profit and loss, labor scheduling, inventory, safety, fleet, deliveries, customer relationships. All of it, all day. When the company ran internal interviews with branch managers across the country, the pattern that emerged was not a missing report. It was overload.

The managers described logging into one system for speeding data on the trucks, another for heavy equipment, another for parts, and going through each one looking for anything that might be wrong. They described learning that a P&L line had gone sideways at month end, when it was already too late to do anything but explain it. Items were not being missed because nobody cared. They were being missed because everything was competing for the same attention.

The obvious product to build here is a dashboard. One screen, every system, all the numbers in one place.

The company did not want to build it. It had seen this movie before.

His argument went further. Consolidating five systems into one view, he said, is an act of human judgment disguised as technology. Someone decides in advance which numbers matter, and the system inherits that bias permanently. "AI creates value by surfacing unknown patterns. When we create a single view of something, we're putting our own human bias and saying this is important."

The principle he kept returning to was how to surface an insight the manager can act on, at the moment acting still helps. Not a to-do list. Not a prettier report. "The focus should be the learning loop," he said. "Or else you might as well develop a Power BI report."

So Lily for Operations was built to be prescriptive rather than descriptive. It flags a revenue anomaly on a key account while there is still time to call the customer, instead of surfacing it at month close. It projects who is tracking toward overtime by Friday based on hours logged Monday through Thursday, and shows where the manager can shift work. It notices when a nearer branch could make a delivery more cheaply and drafts the email to arrange it.

And it is built to be told when it is wrong. Thumbs up, thumbs down, iterate. Pick three to five insights, put them in front of people who know the business better than the system does, and let them sort it out.

Which brings it back to the shrug. An expert saying "I know that already" is not a failure signal. It is the baseline every AI system starts from when it meets someone who has done the job for twenty years.

The work is getting to the other side of it.