The signature that does not scale

The decision as it arrived

The client was the CEO and majority owner of a contract pharmaceutical manufacturer, 600 people filling and packaging sterile injectable drugs for the companies that own them.

Two of her three largest customers asked her the same question in one quarter. What is your AI plan? She had a good one. $14m, aimed at the paperwork the business runs on: the reports her staff write when something unexpected happens on a line, and the investigations that follow. Her quality director had tested a tool that drafted one of those in 11 minutes. Her people take four hours. The board wanted an answer at the next meeting.

The Read

Ammara followed one investigation from the first draft to the signature that closes it.

Seven people at the firm can sign, and she asked each the same question. If this volume doubled, what would you stop signing? Five gave the same answer, and it was not the investigations. They named the routine approvals for suppliers and equipment.

She then pulled two years of letters from the regulator. Both findings the firm had received concerned investigations closed too quickly. Neither concerned a backlog.

The Pressure Test

The operator who knew the industry best argued for committing. She had run quality at a larger manufacturer and put this exact tool into two sites. Drafting is where the hours go, she said, and a customer who asks about AI twice in one quarter is telling you something about the next renewal.

The second had bought the same class of tool for an insurance claims business. Drafting was not the bottleneck, he said. The signature was, and the tool pushed work toward it faster than it could move. Seven is not a number you buy your way out of in two years. These signers hold a standing the regulator grants by name, and finding three more takes 18 months when the market has them.

The third had joined a manufacturer the year after it did exactly this. Four times the volume went into an unchanged review layer, and reviewers who had read every line began reading the summary. The regulator did not find a backlog. It found closures that looked alike. The customers behind her question buy $29m a year, and one finding like that puts the site on notice with all of them.

The Action Brief

Do not commit. She had brought one purchase with a board date on it, and the answer to that purchase was no.

The idea was right and the order was wrong. Put the technology where no signature waits at the end: supplier data and stability trending, and the scheduling that drives annual product reviews into the same two months every year. That frees the signers the rest of her plan depends on, and she can show it to a customer this quarter.

Two conditions. Measure what the signers carry before anything is deployed, and make freeing that load the test for reconsidering the paperwork program. The smaller program costs a third of the $4m and produces nothing a customer can point to for a year.

The pilot proved the drafting works. Her constraint sat one step past the draft.

Where it landed

At the debrief she said the board approved the smaller program and one director called it timid. One customer pushed again at renewal, and after she sent the plan the contract renewed smaller than the year before. Her quality director, who had built the pilot, left eight months later.

Ammara followed the paperwork past the point the pilot stopped measuring, and put what she found to three people who had seen where it goes.