Sector Profile · / 04

Manufacturing

Connect the plant floor to ERP and unlock real operational visibility.

Operational Pain Points

Production downtime, quality variance, and disconnected MES, ERP, and quality systems.

The plant knows things the ERP finds out about days later, and the ERP knows things the plant never hears at all. In that gap live the classic mid-market manufacturing losses: unplanned downtime that surprises scheduling, quality variance discovered at final inspection instead of at the machine, and planners running the real schedule in a spreadsheet because the system's version can't be trusted.

  • Instrument the constraint first — one bottleneck line with honest OEE data changes the conversation.
  • Close the loop between MES, quality, and ERP so scrap and downtime hit planning in minutes, not days.
  • Replace the spreadsheet schedule by making the system schedule trustworthy, not by decree.
  • Give supervisors shift-level dashboards that answer "what hurt us today" without a data analyst.

Legacy System Issues

On-prem ERP, custom PLC programs, and brittle integrations layered over decades.

Manufacturing legacy is physical: the ERP is on-prem and heavily customized, the PLC programs were written by a contractor who retired, and the integration layer is a folder of CSV exports on a schedule. It all works — until the one person who understands it takes vacation. Modernization here means respecting that production cannot stop while the estate is untangled.

  • Build the integration map including the human ones — the analyst who re-keys is an interface too.
  • Put an industrial data layer between the plant floor and business systems so neither holds the other hostage.
  • De-risk the ERP question with a customization audit — most "requirements" turn out to be habits.
  • Version-control and document PLC and HMI programs before the knowledge walks out the door.

AI Opportunities

Predictive maintenance, computer vision QA, demand forecasting, and yield optimization.

Manufacturing AI works when it starts from a physical truth: the data is sensor data, the payoff is uptime and yield, and the operators must trust it or it dies. Predictive maintenance on critical assets, vision-based inspection at speeds humans can't sustain, and demand forecasts that actually reach the production schedule are proven, measurable plays — once the plant-to-ERP data spine exists to feed them.

  • Start predictive maintenance on the asset whose failure hurts most — credibility compounds from there.
  • Deploy vision QA at the inspection point with the worst escape rate, measured against a human baseline.
  • Connect demand forecasting to the schedule people actually use, or it's just a chart.
  • Design every model with operator override and feedback — trust is the deployment platform.

Regulatory Complexity

Traceability, ISO/QMS, and increasingly aggressive cyber-physical safety rules.

Between customer audits, ISO recertification, and rising cyber requirements on anything connected to a machine, mid-market manufacturers face compliance pressure without the compliance staff of larger operators. Traceability demands from OEM customers now effectively mandate the same data spine that good operations wanted anyway.

  • Make lot and serial traceability a data-architecture feature, not a paperwork exercise.
  • Segment plant networks so connecting machines for data doesn't connect them for attackers.
  • Automate the evidence trail for QMS audits out of systems of record.

Signals

You know it's time when…

  1. The real production schedule lives in a spreadsheet owned by one planner.
  2. Downtime is explained in Monday meetings using numbers everyone quietly disputes.
  3. Quality escapes are caught by the customer more often than by the line.
  4. The ERP upgrade is unthinkable because of two decades of customizations.
  5. One contractor holds the only current knowledge of your PLC programs.
  6. An OEM customer just made digital traceability a condition of the next contract.

Engagement

How the climb typically unfolds

Weeks 1–3

Walk the floor

Gemba first, systems second: map material and data flow from receiving dock to shipping dock, instrument the constraint, and quantify the downtime and quality-variance cost in numbers finance signs off on.

Weeks 4–10

Connect

Stand up the industrial data layer between floor and ERP, ship supervisor dashboards, and retire the highest-risk manual data bridges.

Months 3–9

Optimize

Deploy the first AI plays — predictive maintenance and vision QA where the economics are strongest — while the ERP customization audit sets up the modernize-or-replace decision with real evidence.

Months 9–12

Hand off

Transfer a documented, instrumented estate and a prioritized roadmap to the permanent leader, with plant leadership already running the operating rhythm.

Field Notes

Mid-market manufacturers are told they need a “smart factory,” usually by someone selling sensors. What they actually need is simpler and harder: for the plant floor and the business systems to stop being strangers. Everything else — the downtime surprises, the quality escapes, the spreadsheet schedule, the AI pilots that never leave the demo — is downstream of that one disconnect.

Two worlds, one company

Every manufacturer we walk into runs as two loosely-coupled worlds. The OT world — PLCs, HMIs, machines, the people who keep them running — operates in seconds and shifts. The IT world — ERP, quality systems, finance — operates in transactions and month-ends. Between them sits a human integration layer: supervisors re-keying production counts, quality techs transcribing measurements, a planner maintaining the real schedule in Excel because the system’s schedule stopped being believable years ago.

The fix is not a monolithic platform. It’s a thin, honest industrial data layer that lets each world see the other in near-real-time, deployed line by line, starting at the constraint. When downtime on the bottleneck line hits the planner’s screen in minutes instead of at tomorrow’s production meeting, behavior changes — and every subsequent investment gets easier to justify because its impact is finally measurable.

AI that survives the floor

Manufacturing has the highest-yield AI plays in the mid-market — predictive maintenance, vision QA, demand-to-schedule forecasting — and the highest kill rate for pilots. The difference is almost never the model; it’s whether the data spine exists and whether operators trust the tool. We sequence AI after instrumentation, target the asset or inspection point where pain is felt every day, and design override-and-feedback into everything. Trust compounds exactly like OEE does.

The Sherpa difference

This work needs a leader who is comfortable in steel-toes at 6 a.m. and in the board meeting at 2 p.m. — someone who has run plants’ technology before and knows which vendor promises evaporate on contact with a running line. That’s who takes the engagement: embedded a few days a week, measured on downtime, yield, and schedule trust, gone in about a year, leaving a plant that sees itself clearly.

FAQ

Questions manufacturing leaders ask us

Do we need a full MES before any of this works?

No. Full MES implementations are multi-year commitments that mid-market plants often don't need to start. A lightweight industrial data layer over existing PLCs and sensors delivers the visibility that matters in months, and makes any later MES decision better-informed and lower-risk.

Our ERP is heavily customized and ancient. Replace it?

Audit it first. In most plants, the majority of customizations encode habits, not requirements, and a large share of the pain attributed to the ERP actually lives in the integrations around it. We de-risk the decision with evidence; sometimes the answer is replace, more often it's simplify and wrap.

Will operators actually use AI tools on the floor?

They will if the tools respect them. Every model we deploy has operator override, visible reasoning, and a feedback path — and we start where the pain is felt daily, like the asset that keeps failing on night shift. Operator trust is the real deployment platform; without it the fanciest model is shelfware.

What size manufacturer is this for?

Typically $50M–$500M revenue operations — big enough that downtime and quality variance are seven-figure problems, not big enough to carry a full-time CTO plus a data engineering team. Fractional leadership fits exactly that gap.

Talk to a Manufacturing Sherpa

Thirty minutes with a fractional executive who has led manufacturing transformation before. No deck, no pitch — just an honest read on your situation.