This website uses cookies

Read our Privacy policy and Terms of use for more information.

Scalatris
ENGINEERED TO ELEVATE
ISSUE #10

THE FACTORY FORWARD  ·  04 AUGUST 2026  ·  ~3 MIN READ

The promoter asked "why this number?" Nobody could answer.

WHAT'S INSIDE THIS WEEK

▸ ONE FIELD NOTE
The pilot that ran fine - until the promoter walked over and asked the operator why the AI picked that number.

▸ ONE TOOL TO TRY
A free tool that makes your own SOPs and manuals answer "why" - and points to the exact line the answer came from.

▸ ONE NUMBER
3. The three things people do when handed an AI output they can't explain - all three fatal on a shop floor.

We've spent the last two issues on whether your plant manager will run the pilot. This week, assume they will. It's running.

The AI spits out a production plan, a reorder quantity, a supplier flag. Then the promoter walks over and asks one question: "Why this number?"

And your operator can't answer.

That silence kills more pilots than any technical failure. Not because the tool was wrong - because nobody on the floor could stand behind it. The first time a straight question gets a shrug, the promoter stops trusting the system.

This week: how to make sure someone on your floor can defend the number - before the promoter ever asks.

WHO CAN DEFEND THE OUTPUT DECIDES WHETHER IT SURVIVES

Same AI output. Two floors. Only one survives the promoter's question.

THE BLACK BOX

"The system said so"
Nobody knows which inputs drove it
Staff hide that AI was involved
Or oversell it as "objective"
Dies the first time it's questioned

 

THE NUMBER OWNER

One named person per output
Can say why in one sentence
Knows the two inputs it depends on
Knows what to check if it looks wrong
Survives the promoter's question

The real deliverable of a pilot isn't the model. It's one person who can explain the number without calling the vendor.

Behaviour patterns (hide the AI's role / oversell objectivity / can't explain the logic) per HBR Management Tip, 27 Jul 2026. Illustrative framing, not a named client outcome.

ONE FIELD NOTE

The pilot ran clean for a month. It died in one sentence: "the system said so."

The AI layer was doing its job - reorder quantities, a production plan, a supplier flag. Then the promoter, on a floor walk, pointed at a number and asked the operator why it was that and not something else.

The operator didn't know. Not because they were careless - because the pilot handed them an output and never handed them the reasoning. Kaizen Institute made the same point this week about why big changes stall at scale: sustaining an improvement is a matter of practitioner capability - the people running it actually understanding it - not the tool itself.

A system integrator ships a black box and calls it done. That's exactly the thing that dies at scale.

A model that works proves the AI ran once. A person who can explain it proves the plant will keep using it.

Illustrative composite of an Indian manufacturing MSME. Not a named client. "Sustaining change rests on practitioner capability" per Kaizen Institute, 27 Jul 2026.

ONE TOOL TO TRY

Google NotebookLM - make your own documents answer "why."

Upload the documents you already have - an SOP, a maintenance manual, a quality standard, a supplier contract - and ask questions in plain language. Every answer is grounded in your document and points back to the exact line it came from. It's the cheapest way to make a document explain itself, which is this week's whole point.

Setup, once:
1. Go to notebooklm.google.com
2. Sign in with a Google account
3. Click "New notebook" and upload your PDF
4. Ask it "why" like you'd ask a senior operator - e.g. "why does this SOP require a second inspection at station 4?"

The line doing the work is it points to the exact line the answer came from. An answer you can trace is an answer your operator can defend. An answer with no source is just another black box.

Free at the time of writing - worth a 10-minute test before you trust it with anything critical. Confirm the free tier is still live in India before you promote it to the team.

YOUR ONE MOVE THIS WEEK

Name a "number owner" before your next AI output goes live.

One person per report who has to answer "why this number?" without calling the vendor. Not you. Not the vendor. Someone on the floor who has to say the sentence out loud. Reply "workshop" if you want this worked through live at the next session.

Reply "workshop" →

THIS WEEK'S NUMBER

3.

HBR named the three things people do when they're handed an AI output they can't explain: they hide that AI was involved, they oversell it as "objective," or they simply can't explain the logic. On a shop floor, all three are fatal - each one is a different way of failing the promoter's "why?"

My read: none of these is a training problem you fix after the fact. They're a handover problem. If the pilot handed over the output but not the reasoning, you built all three behaviours in on day one.

Source: Harvard Business Review, "The Management Tip of the Day - Help Employees Explain AI Decisions," 27 Jul 2026. External framing, not a Scalatris study.

THIS WEEK'S QUESTION

On 22 July, Parliament confirmed India is backing 20 home-grown AI models - including small, domain-specific ones aimed at industrial use (IndiaAI Mission, Lok Sabha, 22 Jul 2026). Which puts a question on every promoter's desk: when you adopt AI, whose AI is it - and where does your plant's data go?

Hit reply and tell me: has a data-ownership clause ever stopped an AI conversation cold in your business? I read every reply, and the best ones shape a future issue.

COMING NEXT WEEK

The one held over from Issue #9: what to do when your visual management wall turns red and the team starts hiding the real numbers instead of posting them - the honesty problem that shows up the moment a number has an owner.

P.S. The next AI in Manufacturing masterclass cohort is being planned - dates being finalised. Reply with SEAT if you want first notice when registration opens.

And the one-line version of this whole issue: don't ship your team a number. Ship them a number they can defend. One of those survives a floor walk. The other one doesn't.

Until next week,
Sachin
Founder, Scalatris LLP - scalatris.com

Name your first number owner before the next shift:

Reply "workshop" →

REFER A COLLEAGUE

Know a plant manager whose team can't explain what the AI told them to do? Forward this issue and send them your link.

1 referral: AI Prompt Card, 15 ready-to-use prompts for manufacturing ops, QC, and cost analysis (PDF)
3 referrals: AI Prompt Toolkit, 30 battle-tested prompts for operators, quality managers, and MSME founders
10 referrals: private reward, confirm live Beehiiv setting before sending

Reply

Avatar

or to participate

Recommended for you

View all
caret-right