Two pitches, and only one of them is ours
Walk into almost any plant and ask about AI, and the pitch that shows up first is the shop floor: a camera system that catches a defect an inspector missed, a model that flags a bearing about to fail before it takes down a line, a digital twin of the whole operation. Those are real disciplines, and for the right plant they are worth the investment. They are also capital projects — sensors to install, historian data to build, an integration effort that touches operational technology most IT teams do not own, and a payback horizon measured in years rather than months.
While that conversation happens, the office keeps running the way it always has. An RFQ sits in an inbox because the estimator who can price it is buried in three others. A spec change arrives as a PDF and gets re-keyed into the ERP by hand, and the mistake shows up on the shop floor three weeks later as a wrong revision on a traveler. Nobody scoped a project for any of that, because it never got a demo at a trade show.
Calfy's job in this engagement is to put both lists in front of you honestly, side by side, without pretending they compete for the same budget or the same timeline. That means stating our own position plainly: Calfy builds the office-side systems — quoting, order intake, supplier communication, reporting, documentation. We do not sell computer vision, predictive maintenance, or the sensor infrastructure a shop-floor project needs. Where the audit turns up a project that belongs on the floor, we say so and point you toward the kind of vendor that project actually needs, rather than stretching an office system to cover ground it was never built for.
What the engagement produces
The engagement produces a document three different people can act on without a follow-up meeting: your operations lead, whoever controls the capital budget, and whoever is going to actually build whatever comes next. First, a single ranked list that includes both office opportunities and shop-floor capital projects, scored on the same page against effort, cost, and what it is actually worth — not two separate conversations that never get reconciled. Second, a clear call on which items Calfy can scope and which need a systems integrator, a machine-vision vendor, or a controls engineer instead. Third, a sequence: what gets built first, second, and only once the earlier stages have paid for themselves.
None of that requires you to have already picked a vendor for the shop-floor half before the engagement starts. Plenty of clients come to us with a board-level mandate to do something with AI and no clear sense of which half of the plant to start with. Sorting that out is the first working session, not a prerequisite for booking one — the same AI strategy work we run for businesses well outside manufacturing, adapted here to a plant with two genuinely different kinds of opportunity on the table.
The data reality behind the roadmap
Before we rank anything, we look at where the truth about your operation actually lives, because it is rarely in one place. The ERP holds jobs, parts, and costs — but usually not the real story of why a quote took four days or which supplier is chronically late. The real story lives in spreadsheets: a planner's private tracking sheet, a quoting engineer's own margin calculator, a shift-report tally nobody else has seen. And the engineering truth — the actual current revision of a drawing, a tolerance a customer agreed to informally over email three years ago — often lives on a shared drive organized by whoever happened to be doing the filing at the time.
None of that is a criticism. It is how manufacturing operations accumulate over twenty or thirty years of runs, system changes, and staff turnover. But it changes what gets sequenced first. A candidate that looks straightforward on paper — automate the quote-to-order handoff — can turn out to depend on data spread across the ERP, a spreadsheet, and a shared drive that do not agree with each other. That candidate either needs a data-cleanup step first or moves down the list behind something simpler. We would rather say that in the strategy session than let an agent build confidently on data nobody actually trusts.
Retiring workforce knowledge is a deadline, not a nice-to-have
Most plants we talk to have at least one person whose head holds more institutional knowledge than any system does — the estimator who can eyeball a drawing and know the price without opening the ERP, the quality lead who remembers exactly which supplier's material caused a problem years back, the planner who knows which customer's "urgent" actually means urgent. That knowledge was never written down because it never had to be; it lived in a person who showed up every day.
It becomes a strategic problem the moment that person is within a few years of retiring, and manufacturing is dealing with more of that right now than most industries. This is not a someday risk that can sit at the bottom of a list indefinitely — it has an actual date attached, whatever that person's own retirement timeline turns out to be, and once they leave, capturing what they knew gets a great deal harder and more expensive to do well. Part of the audit is asking, plainly, who is closest to walking out the door and what they know that exists nowhere else. Where the answer is uncomfortable, that project moves up the sequence — not because it is the most exciting item on the list, but because it is the one with a deadline nobody set on purpose.
Sequencing: the office work funds the bigger projects
Ranked opportunities do not get built in order of ambition, and on a mixed list like this one, the capital projects almost never go first — not because they are not worth doing, but because a plant with no track record running an AI project has nothing to fund or de-risk one with. The office side gets built first for a practical reason: it ships faster, it is cheaper to be wrong about, and the time it gives back — hours returned to an estimator's week, a planner no longer chasing the same three spreadsheets — builds both the budget case and the internal confidence for whatever needs capital approval next.
That sequencing produces something a spreadsheet business case cannot: a plant that has already lived through one AI project landing well, with people who trust the next one because they watched the first one work. By the time the vision-inspection or predictive-maintenance conversation reaches a real budget decision, it is being made by people who have seen what a system built around how the plant actually runs looks like, instead of people evaluating it purely on a vendor's demo.
Walkthrough: sorting the wishlist before it becomes a budget request
A contract manufacturer comes to us with a board mandate to get ahead on AI and a wishlist built from three different trade-show pitches: a vision-inspection system for a chronic defect on one line, predictive maintenance on an aging compressor, and something to help with the RFQ backlog. The first working session does not start by picking one. It starts by mapping what each item would actually require — the vision system needs cameras and a lighting setup that does not exist yet, the compressor has no sensors reporting condition data today, and the RFQ backlog is entirely inboxes, PDF drawings, and one estimator who has not had a quiet week in months.
The roadmap that comes out of that session is honest about the gap between those three. The RFQ triage work is scoped as the first build, with clear pricing and a live date measured in weeks. The vision and predictive-maintenance items stay on the roadmap too, but flagged as capital projects that need a machine-vision integrator and a sensor retrofit respectively — work outside what Calfy builds. We hand the plant a one-page brief on what each of those would need from a different vendor, so the board mandate gets executed against reality instead of the pitch that started it.
Walkthrough: putting a retirement date on a knowledge-capture project
A job shop's quoting is effectively run by one estimator who has been there over twenty years and mentions, almost in passing, that he is planning to retire within the next two years. Everyone in the room treats it as a scheduling detail. The audit treats it as the most urgent item on the list, because his judgment — which tolerances are actually achievable at which price, which customers pad their specs and which do not — exists nowhere except in his head.
The scoped project is not a chatbot that tries to replace him. It is a knowledge-capture and retrieval system built alongside him while he is still there: structured interviews turned into a searchable reference tied to real part numbers and past quotes, cross-checked against the history of what he approved and adjusted over the years. The roadmap sequences this ahead of two more obviously exciting projects, because the cost of building it after he leaves is not just higher — some of what he knows will not be recoverable at any price.
What this becomes: the roadmap turns into a build, or a referral
For the office half of the roadmap, this engagement leads directly into a build almost every time — often a custom AI agent for the RFQ or order-intake workflow that genuinely needs judgment, sometimes a more straightforward workflow automation for reporting or document routing that follows a fixed set of steps, and often a knowledge retrieval system for the drawings, specs, and institutional knowledge scattered across file servers and people's heads. Part of an honest strategy engagement is telling you which of those three fits each item, rather than defaulting to whichever sounds most impressive.
For the shop-floor half, the roadmap hands you a scoped, honest brief to take to the right kind of vendor — what the project needs, roughly what it depends on, and why it sits outside what an AI agent built for office workflows can cover. Some clients take that brief and run the capital project entirely separately, on its own timeline, with its own budget owner, and that is a good outcome. The strategy work is not trying to keep the whole roadmap inside one vendor relationship; it is trying to make sure nothing on the list gets built by the wrong kind of vendor, including us. The wider view of how the office side fits a plant's day-to-day operation is on our manufacturing industry page, for teams weighing this against everything else on the list.