In 2019, Hisco installed Proton, an order recommendation and reorder engine that put AI-driven suggestions in front of our customer service reps. I’ve written about that implementation before. Here’s the part I haven’t.
We wrote standard work for it.
Content: what the rep looked at and what the tool was allowed to feed them. Sequence: the steps from recommendation to customer. Timing: where it sat in the day and in the call flow. Expected outcome: what a correct result looked like, so a rep and a supervisor could both tell the tool working from the tool guessing.
Same template we used for receiving, picking and returns. Nobody called it AI enablement.
The tool was new.
The management discipline wasn’t.
I bring it up because last week a company many times Hisco’s size did the same thing, and most of the industry read it as a software story.
What Uber Actually Did
On October 22, OpenAI announced that Uber had expanded ChatGPT Work from a 1,000-person pilot to more than 30,000 employees. OpenAI credits three things with carrying the rollout: an internal AI skills library, approved integrations, and dedicated employee training.
Strip the vendor language off that list and you’re looking at a standard.
The skills library is the content. OpenAI describes a collection of reusable skills that teams in operations, finance, marketing, and product pull from. Someone at Uber wrote down how the work gets done and turned it into assets the whole company can use. We’d call that standard work. The tech industry calls it a skill.
The approved integrations are the content boundary. ChatGPT Work connects to Google Drive, Gmail, Google Calendar, and Slack. Somebody decided what the tool could touch before anyone used it, and by implication what it couldn’t.
The training is how a standard gets from the binder to the floor.
And the 1,000-person pilot is the pilot. Uber ran it, learned from it, then committed the company.
The software was the headline.
The standard was the story.
The Quiet Assumption
Most distribution leaders treat an AI rollout as a purchasing decision. Pick the tool, negotiate the seats, send the announcement. Six months later you read the usage report and see what you already suspected.
No operator I know would launch a new pick process that way. You’d write the standard first. Then you’d train to it. Then you’d audit it.
AI gets a pass on that discipline. That pass explains most of the stalled rollouts I see.
A license is a tool. A standard is a system. Distributors keep buying the first and expecting the second.
What It Looks Like Without One
Picture a regional electrical distributor with 280 employees across six branches. This is a composite built from patterns I’ve watched repeat, not a named account.
In March, leadership approves Copilot for everyone. The announcement goes out on Teams. IT turns on the licenses. The CFO signs off because the per-seat number looked reasonable next to the headcount.
Nobody writes down what a good quote email looks like.
Nobody decides whether the tool is allowed to read from the ERP.
Nobody trains the inside sales team beyond a lunch-and-learn.
By September, three people on the quote desk do something useful every day. A dozen more paste customer price files into whatever chatbot is open in another tab. The branch managers never touched it. The CFO asks what the subscription is for, and nobody has a clean answer.
The tool worked fine. The company never gave it a standard.
I’ve watched this movie with warehouse management systems, with CRM, and with the first wave of e-commerce. Technology changes. The failure pattern holds.
Why AI Gets the Pass
The reason is structural, and it has little to do with competence.
Buying seats is a visible decision with a clear owner. It fits on a purchase order. It shows up on the board deck as “AI initiative launched.”
Writing the standard is invisible. It sits between IT, operations, and sales. It requires someone to sit at the quote desk for two weeks and write down what a 20-year veteran does without thinking. Nobody gets credit for that in the quarter it happens.
There’s a second reason, and it’s a belief problem. Leaders assume AI is too fluid to standardize because every prompt and every answer comes out different. Customer calls come out differently every time too, and we wrote standards for those. You standardize the inputs, the sequence, the checkpoints, and the definition of done. You leave the conversation to the human.
So, the organization does the visible part, skips the invisible part, and blames adoption.
What ASCM Found
ASCM and the Digital Supply Chain Institute released a report this month titled The People Paradox in Supply Chain AI Integration. DSG covered it on October 22. Three findings line up with what Uber did and what we did at Hisco.
83% of supply chain leaders say their employees are unprepared to use AI effectively. Only 14% describe their organizations as fully prepared for an AI-augmented future.
Across 13 core supply chain functions, average AI adoption scored 1.94 on a five-point scale, where 1 means no use and 5 means fully autonomous execution. Most companies have barely left the starting line.
The finding that matters most here: 17% of organizations with an AI-ready workforce report a 71% project success rate, against 28% for everyone else. The report’s authors attribute the difference to learning by doing.
Standard work is how learning by doing becomes repeatable. Without it, every rep learns alone and the organization never learns at all.
The Four Elements, Applied to a Quote Desk
You’ve written hundreds of these. The template doesn’t change because the tool did.
Content. The ten tasks the desk repeats every day: build a quote from an emailed bill of materials, write the follow-up when a quote goes quiet, check a cross-reference, draft the response when a customer disputes a lead time, summarize the account history before a call. Each one written as a reusable instruction with the examples your best person would use. Content also means the systems the tool may read from and the shorter list it may write to. Decide that before anyone logs in. Deciding it afterward is how price files end up in a public chatbot.
Sequence. The order of steps, including where the human checks. The tool drafts, the rep reviews, the rep sends. Or the tool pulls account history, the rep reads it before the call, the rep decides what to use. Write the human checkpoint into the sequence or it disappears under volume.
Timing. When it happens and how long should it take? Quote follow-ups get drafted first thing, before the phones start. A quote from a clean bill of materials takes minutes, and if it takes longer, something in the content is wrong. Timing is what makes the standard auditable.
Expected outcome. What good looks like, written so a supervisor can check it without the rep in the room. Correct pricing tier. Correct lead time from the ERP, never from the model’s memory. The customer’s name spelled right. A follow-up that references the actual quote number. If you can’t describe the correct output, you can’t train to it and you can’t audit it.
This is also where your veterans earn their keep. The person who has worked at the counter for 20 years holds the context the tool lacks. The standard is where that context gets written down, tested and shared. AI brings the information. Your people bring the meaning. The standard is where the two meet.
Where Standards Decay
Even distributors who write the standard hit three second-order problems, and every one of them is familiar from the warehouse.
Nobody audits it. The standard gets written once. Six months later the pricing logic changed, the vendor renamed a product line, and the skill quietly produces wrong answers. Layered audits catch this on the floor. They catch it on the quote desk too. Someone must own the AI standard the way someone owns the price book.
The pilot never ends. One desk runs well, leadership calls it a success, and the expansion to the next desk never gets scheduled. Set the expansion date when the pilot starts.
Content creeps. The approved system list grows one exception at a time until the tool can touch everything. Review it quarterly, the way you review user permissions in the ERP.
[Inference] I’m drawing these three from the pattern of prior process rollouts in distribution, not from the Uber announcement. OpenAI has not published outcome data yet; its post says the two companies plan to share adoption and outcome insights as the rollout progresses.
What Changes Monday Morning
Pick one desk. Order entry or quotes, whichever carries more volume. Skip the whole company launch and skip the department with the most enthusiastic manager.
- Spend the first two weeks with the desk, not with the vendor. Write down the ten tasks that desk repeats every day. Get the longest-tenured person on that desk to review each one and add what the written version missed.
- Write the standard for each task on the template you already use on the floor: content, sequence, timing, expected outcome. Test every standard against last month’s real work.
- Decide, in writing, which systems the tool can read and which it can write to. Keep the write list short.
- Train the desk to the standard. Measure the desk’s output against the expected outcome.
- Put the standard on the audit calendar and set the date for the second desk before the first one goes live.
Login counts measure activity. Expected outcomes measure intent. Most AI dashboards only show the first.
Uber ran this with 1,000 people and then 30,000. We ran it at Hisco with a handful of reps and one tool. A 300-person distributor runs it with eight people and then 40. The structure is identical. Only the scale changes.
Before you sign the next renewal, one question: does your AI have a standard, or just a license?
Distributors looking to evaluate AI tools and their applications can attend Distribution Strategy Group’s Applied AI for Distributors, June 22-24, 2027, at the Renaissance Schaumburg Convention Center Hotel outside Chicago. The conference will bring together distribution leaders, technology providers, and industry practitioners to discuss implementation and measurable results. Register here.
Share this article:
