Home

Line 01 · ROI assurance for automation and AI

Zunft Proof

Before the CFO signs, find out whether the return is real.

Most automation business cases are built to be approved. They count the labor saved and leave out the ramp-up, the downtime, the scrap, the changeovers and the cost of people learning a new line. AI fails the same way: the pilot works, the real workflow doesn't, and the people who knew the exceptions have gone.

Zunft Proof re-underwrites the capital request before it is approved. We rebuild the return on our own math, name and size what threatens it, and set out the moves that raise the odds of hitting it. We never sell, install or manage what we assess, and we take no referral fees, so our call is worth trusting.

The plumb line: the mason's test that a wall is true before anything is built on it.

For

The CFO, CEO or board with an automation or AI capital request on the table, from a single robot line to an enterprise AI suite.

Why the business case breaks

The machine usually works. The system around it doesn't.

What the request countsLabor savedThe vendor's rateUptime at the rating on the plateBuilt to be approved.The robot is the cheap part.An installed robot system typically costs threeto four times the robot itself (NIST).What it leaves out, and Zunft Proof puts backRamp-up and debuggingDowntime in the first yearScrap while the line settlesChangeoversPeople learning the new lineKnow-how that left with the people cutEnd of life: late failures and the cost to tear it outA cheaper technology arriving before payback
Illustrative. Zunft Proof tests every item on both sides against what lines like yours actually did in their first year.

AI is automation. Every way a robot line underdelivers has an AI twin.

Where it breaksOn the robot lineIn the AI rollout
Demo against realityRuns at the integrator's site, struggles on your floorWorks in the pilot, never fits the real workflow
Hidden ramp-upDebugging, changeovers and scrap left out of the ROIData clean-up, integration and workflow redesign left out
Lost know-howExperienced operators cut; the line loses what they knewThe people who knew the exceptions leave; the agent can't handle them
TrustOperators work around the machinePeople ignore or rewrite the AI's output
DriftPLC logic stays stable for yearsModels drift as conditions change, so formed people matter more

Before the numbers

Is the plant ready to automate at all?

Automation magnifies what it finds. On a disciplined floor it multiplies the efficiency; on an unready one it multiplies the mess. So we check four things before we touch the return.

Product

Designed to be automated: standard variants and workable tolerances, not thousands of one-offs.

Process

One best way, written down, running the same on every shift and in every plant.

People

Engineers, controls and maintenance people to own it, and frontline leaders formed to run it.

Systems

A plant network, data down to the part, and a supplier who can dial in when the line stops.

Safety and quality come first. Protecting the worker and the customer is not a business case, and we never ask you to justify it. The business case starts beyond them: productivity, then the flexibility to make what comes next.

From the floor

GM · Saginaw and beyond, from 1989

Same system. Ten transfers. Half failed.

GM captured a veteran engineer's ability to diagnose a machine from its vibration in an expert system called CHARLEY. In repeated tests it identified seventy-two kinds of machine fault with 95 percent accuracy, as good as the best human analyst.

Venki's doctoral research followed CHARLEY as GM moved it into other plants. Of the ten transfers he studied, four succeeded, one partly, and five failed. The rules were identical everywhere. What differed was the receiving plant: whether it had trained people on real machines, and a foreman and a coordinator who would act on what the system said.

The technology was never the bottleneck. Whether anyone had been formed to act on it was. That is the first thing Zunft Proof checks in an AI case today.

Told in full in Venki's book Built to Extract.

From the floor

Mercedes-Benz · Rastatt

The machine found the defects. Then nobody else could.

On the A-Class paint line, a vision system scans every body, finds paint defects invisible to the eye and marks the exact spot for the polishers downstream. It is a magnificent piece of technology.

Over months, the polishers who used to find defects themselves, running their fingers across a fender, stopped looking. They waited for the screen. When the system met something it had not been trained on, a new defect pattern or an unusual paint batch, there was no one left with the hands or the eyes to catch it.

Automation and AI can quietly unform the people around them. Zunft Proof prices that risk into the case, and names the moves that keep people sharp.

Told in full in Venki's book Already Paid For.

What you receive

In writing, and yours to keep

01

The case, re-underwritten

Every assumption tested: ramp-up, downtime, scrap, changeovers and the people the plan takes for granted.

02

The return, rebuilt

What the investment will actually return on our math, counting the growth it can unlock as well as the cost it removes, with each risk named and sized in dollars.

03

Five to eight countermeasures

Specific moves in selection, formation, frontline leadership and tracking that raise the odds.

04

A clear call

Go, no-go, or go with conditions, and the conditions written down. Sometimes the right call is a smaller, assisted system that keeps people where judgment is needed and costs less.

05

The twelve-month audit

Included. Twelve months on, the actual return reconciled against the promise, so the next request starts from evidence.

How it runs

Four steps. Each one says yes before the next begins.

01SuitabilityA live capital request,not yet approved.fits02Base dataThe request, the vendorproposal, line data.NDA signed03Site visitThe line walked; operators,maintainers, finance.data enough04ReportThe return rebuilt andthe call, in person.

If the screen says it does not fit, we tell you why and charge nothing. No data moves before an NDA. Every step leaves you with something in writing, so you can stop after any of them.

Terms

Pricing to follow · about two weeks from data to read-out

A fixed fee: half up front, half on completion. Travel billed at cost, from submitted expenses. We do not promise a return. We improve the odds and show our work.

Talk about a capital request