Part 3 of 3: The vendor meeting is theatre. These questions burn the script.

The vendor meeting is theatre.

The slides are rehearsed. The references are pre-screened. The solution architect has answered the same questions a hundred times. By the time you’re in the room, the entire conversation has been engineered to lead you to yes.

You have one weapon against this. Asymmetric questions — questions the vendor isn’t trained for, that probe architecture instead of features, and that produce visible discomfort when the architecture isn’t there.

Seven of them. Thirty minutes. The right answers feel like relief. The wrong ones feel like a vendor sweating.

Part 1 framed the stakes — the dark factory is real, and your MES decides whether you reach it. Part 2 gave you the architectural test — stream-based or database-based, listen for the verbs.

This post is the action. Print these. Take them into your next vendor meeting. Ask them in order.

Question 1: “Walk me through the path of a single machine event. From the machine to the decision.”

This is the test from Part 2, asked first because it sorts the room in 90 seconds.

A real answer names the broker (Kafka, Pulsar, NATS), names the protocols (MQTT, OPC-UA), and gives a latency number under 200 milliseconds. A fake answer talks about databases, polling intervals, and dashboards.

If the answer is fake, the next six questions will tell you how fake. But you’ll already know.

Question 2: “Show me your AI running in production on a real customer. Today. Not in a demo.”

Not a deck. Not a slide. A live screen-share of a production deployment.

A serious vendor will be eager. They’ll show you the model, the data it reads, the decisions it makes, the customer’s results. Five minutes, max.

A fake vendor will offer to “set up a follow-up session with our AI specialist.” That follow-up never produces what was promised. The AI exists in the brochure, not in production.

The follow-up offer is the answer.

Question 3: “How long does it take you to integrate a machine you’ve never seen before?”

Acceptable: “Two to three weeks. Our integration framework lets us write a new adapter in days.”

Unacceptable: “That would need to be a scoped engagement with our services team.”

When integration is a project, every machine you add for the next five years is a project. Every project is a quarter. Every quarter is budget. The architecture isn’t built for it, and the contract is structured to bill you for that fact.

Read the answer literally. Scoped engagement means we will charge you to learn your factory.

Question 4: “What was the actual go-live timeline for your last three customers? Not your sales deck — the real one.”

The deck always says 12 months. The reality, with legacy vendors, is 18 to 24.

A modern AI-native MES has the first line live in 8 to 12 weeks. Not because they’re cutting corners — because the architecture deploys faster.

Then ask for direct contact with those last three customers’ implementation leads. The most recent ones. Not the curated reference list.

A vendor who has nothing to hide will share. A vendor whose last three implementations went sideways will hesitate, then redirect to “alternative references.”

The hesitation is the answer.

Question 5: “When my process changes, what does it cost me to change the system?”

Walk through four specific scenarios:

  • Add a new station to an existing line. Configuration, or change order?
  • Add a new field to the work-order screen. Configuration, or change order?
  • Add a new business rule for rework approval. Configuration, or change order?
  • Add a new dashboard for a new role. Configuration, or change order?

If three or four are change orders, the system isn’t yours. You’re renting the right to ask the vendor to modify it. Every adjustment is a PO. Every PO is six weeks. By year three, the cumulative customization spend exceeds the original license.

A modern MES exposes configuration. Stations, fields, rules, dashboards — your team owns them.

Question 6: “Show me a live SAP S/4HANA integration in production.”

This is the question that breaks vendors.

Real-time SAP integration — events flowing both directions, sub-second latency, transactional integrity, graceful degradation — is genuinely hard engineering. Most vendors fake it with middleware that polls SAP on a schedule. It “works” in the sense that data eventually moves. It does not work in the sense of running a real-time factory.

A vendor with real SAP integration will screen-share it. They’ll show OData endpoints, error logs, both systems updating in lockstep.

A vendor who fakes it will pivot to “middleware,” “partner ecosystem,” or “we work with Boomi/MuleSoft/Tibco.”

That pivot is the answer.

Question 7: “Read me the clause about who owns my operational data.”

This is the question I save for last because almost no leader thinks to ask it, and it has cost more factories more leverage than any other contract term.

In nearly every legacy MES contract, somewhere in the back of the MSA, is some version of:

  • The vendor may use your operational data for product improvement, benchmarking, and service quality.
  • Your data may be aggregated for industry reports and machine learning training.
  • Termination triggers export in a vendor-defined format.
  • Models trained on your data remain the vendor’s IP.

What this translates to:

Your factory’s operational reality is the vendor’s training data.

You’re not just buying software. You’re providing — for free, in perpetuity — the ground truth on which the vendor will train the AI features they sell to your competitors next year.

A serious modern vendor will give you:

  • Your data is yours. Period.
  • Models trained on your data, for your use, are your IP.
  • Termination triggers full export in your preferred format, including raw event streams.
  • Industry benchmarking is opt-in, not opt-out.

This is not negotiable. If the vendor will not give you ownership of your own factory’s data, walk out.

What Happens In The Room

I’ve watched this play out three times in the last year. The pattern is identical.

By Question 2, the solution architect has stopped reading from the deck and is improvising. Their pace slows.

By Question 4, they’re asking to “circle back with the technical team” on the implementation lead references. The CFO across the table looks up.

By Question 6, the vendor has used the word middleware twice. The deal is already dead — no one has said it out loud.

By Question 7, the vendor is offering to “have legal review the contract language and revert next week.” That’s the polite way of saying we never expected anyone to ask.

The frame the vendor walked in with — we are the safe, established choice — has collapsed by minute thirty. What replaces it isn’t a different vendor. It’s a different question:

Why was I about to sign this?

That question, asked before the contract, saves the next five years.

The One Question Behind All Seven

Every question above is, at heart, the same question:

Is the architecture flexible, or rigid?

Flexible architectures get cheaper every year. Rigid ones get more expensive every year.

In year one, both look identical on the RFP. The compounding starts in year two and is uncatchable by year four.

This is the entire game.

The Closing Frame

The dark factory is not theoretical. The architecture that gets you there is not a buzzword. The vendors who can deliver it are a small subset of the ones pitching it.

The seven questions in this post are how you tell the difference.

If you take one thing from this series:

You are not buying a tool. You are picking the substrate on which every operational decision in your factory will be made for the next decade. The vendor will frame it as a software purchase. It is not. It is a foundational bet on the architecture of your future operations.

Get it right, and your factory compounds.

Get it wrong, and you’ll spend the next five years explaining to your board why the dark factory across town is shipping at twice your throughput on half your headcount — while your team is still chasing a change order from month 14.

The decision feels like it’s about MES. It’s not. It’s about whether your factory makes it to 2030 as a competitor or as a case study.

Pick carefully. Ask the seven questions. Listen to the verbs.

The next decade is being decided in conference rooms right now. Make sure yours has the right questions in it.

This concludes the three-part series. Part 1: the dark factory and why MES is the spine. Part 2: the architectural test that separates real AI-native from buzz.

Lava writes about manufacturing technology and the operating systems of modern factories. Find him at imlava.in. For consulting on MES architecture decisions, reach out via the website.