A number nobody believed
The company was a founder-era manufacturer doing about $50 million a year with roughly 120 people. The founder was still active, but no longer in the day-to-day. He could see the company needed an operating structure it had never had, and had been resisting. He also knew newer technology, especially AI, could add real value, even if he wasn't sure yet what that looked like. Mostly, he knew too many decisions were landing on his desk, often without him being close enough to the work to make the best call.
On-time shipping was reported at 68 percent, and nobody trusted it. Customer calls about late orders bounced between leaders. Sales, operations and delivery needed a referee most days. There was no clear escalation path except the founder, and going straight to him usually skipped the people who could actually fix the problem.
Expedite freight, overtime, and friction
When people don't trust the system, they work around it. People who had stopped believing the numbers took orders into their own hands, which left the whole company blind to what was open, what had been promised, and what had actually shipped.
Expedited freight was constant, and it ate margin. Unnecessary overtime eroded it further. But the biggest cost was the friction it created between sales and operations.
Three hours, one whiteboard
We put operations leadership, shipping, sales, and production in the same room for about three hours: fifteen people from six functions. Each person walked through exactly what their function did, and we mapped it on the board in real time. Most people were surprised to learn some of that work was being done at all.
Then we looked hard at the number itself. The calculation was correct. The gut feeling that the number was wrong was also correct, and we had to prove it before anyone would stop being suspicious. The data feeding the calculation had problems:
- Orders and delivery dates were entered several different ways.
- System defaults had never been updated.
- Delivery dates that got extended were never changed in the system.
- Some people bought material and created orders their own way, outside the system.
- Old orders hung around as anchors that dragged the number down no matter what.
Clean data first, then a weekly number
A plan both sides signed. Sales leadership and operations leadership committed to a specific set of changes, and we agreed to review the work again in two weeks. At that review, the number had moved into the low 80s. From there we found what else wasn't being handled and fixed it in the system, and within two months the true number was around 90 percent.
One weekly report. Every week, the actual on-time shipping number went out. The real win was that we now knew exactly how to calculate it correctly, and when we missed the target, we knew which areas to go fix.
The process behind the number. Getting from 90 to 97 took every function, from the minute an order was taken until it went out the door. We found the internal bottlenecks, the paths that led to rework, and the steps where delays tended to happen. Much of it lived in order intake and entry.
Tools that catch problems early. We built AI tools that flagged missing information in the sales process, the gaps that would otherwise surface weeks later as a late shipment. With each fix, the number moved a little more, until it held in the 97 percent range.
Sales owns the number
Once the process was fully in place, sales owned the on-time number. They were responsible for leading the effort to get information right at the moment an order was placed, which is where most late shipments actually started.
Because the company could now give customers a true delivery date, far fewer orders needed to be expedited, and expedited shipping costs came down with them. Customers got dates they could plan around.
Off his desk
- Fielding escalations that belonged with the function that could fix them.
- Making delivery calls without being close enough to the work.
- Refereeing between sales, operations, and delivery.
- Wondering whether the on-time number was real.