Saskatchewan manufacturers have been told for a while that AI will create value. The sharper question, the one a provincial session put on the table, is how you approach it.

On November 24, 2025, Excellence in Manufacturing Consortium ran a virtual session called Charting Your Operational AI Path for Saskatchewan Manufacturers. The agenda put the choice in plain language. For many Saskatchewan manufacturers, "the question isn't if AI can create value," it said. "It's how to approach it." Then it asked the decision that actually changes how you spend money and time: "Should you buy ready-made solutions, or build your own?"

That is a better starting point than "should we use AI?" for a Regina fabricator, a process manufacturer, or a shop that feeds construction and resource work. Buying a tool and building one are both ways to spend money. They fail in different ways, and they fail faster if you have not decided which job the tool is for.

Where the question came from, and the limits of the sources

An association webinar is not a survey. It tells you what EMC thought Saskatchewan manufacturers needed to hear in late 2025, not how many Regina shops have tried a tool, or what they decided. The session was led by a consultant who helps manufacturers build data and AI capabilities, so he has an interest in the "build" side of the choice. Read it as a framing, not as a census.

Other Saskatchewan sources point in the same direction, though. Innovation Saskatchewan has been running hands-on sessions on using specific tools for business work, including a Claude AI lunch and learn in June 2026 and a Launch Pad AI Basics workshop. Those are "how do I use this?" sessions, which usually follow the build-or-buy question rather than replace it. Canadian Manufacturers and Exporters has also offered a practical AI workshop for manufacturing teams, covering use-case identification, data governance and low-cost pilots. That workshop is not Saskatchewan-only, so it is supporting context, not local evidence.

None of these sources will tell you which of your processes should be bought, built or left alone. That part is yours.

Why the choice matters in Saskatchewan right now

BuildForce Canada's 2026 to 2035 outlook for Saskatchewan describes a province that is busy, uneven and short of people. Residential construction employment is expected to rise 23% from 2025 levels by 2035. Non-residential work stays lumpy, tied to the timing of resource and utilities projects, including the BHP Jansen potash mine, the Foran-McIlvenna Bay mine, the Aspen Power Station and the anticipated start of Denison's Wheeler River uranium project. ICI buildings work is supported in the short term by health and education projects, then eases as those jobs pass their peak.

The labour math is tight. The same outlook says the province may need to hire 13,500 construction workers by 2035, including 9,400 people expected to retire, and could still be 3,600 short after local new entrants. BuildForce also notes that if proposed projects now under consideration, including the Bell Canada AI data centre, reach a final investment decision, non-residential labour demand could rise further. That last point is a conditional, not a forecast of a project that has already been built.

For a Regina manufacturer or fabricator, that mix is the backdrop for the build-or-buy question. Work is there. Experienced people are harder to replace. A tool that genuinely reduces paperwork, rework or quote cycle time is worth considering. A tool that creates a second system for someone to feed is not.

When buying is the better path

Buy when the job is common, the output is easy to check, and you do not want to own the software.

Good candidates are the jobs almost every shop already does with templates and email: first drafts of quote cover letters, meeting notes turned into action lists, a summary of a supplier PDF, a checklist for a recurring inspection or closeout package. Off-the-shelf chat tools and the AI features already sitting inside software you pay for can do this work today. You do not need a custom model to draft a follow-up email.

Buying still needs rules. Someone has to review the output before it goes to a customer, a general contractor or a regulator. Someone has to decide what never gets pasted into a tool, including customer drawings you do not have permission to share, and your own labour rates and margins. And someone has to notice when the tool is confidently wrong. More on that habit is in Don't Believe Everything the Robot Says.

If a vendor is selling you a closed system that only works if you move your scheduling, quality records and costing into their platform, slow down. That is not a small buy. That is a systems decision, and it should be priced and scoped like one.

When building is the better path

Build when the work is specific to your shop, the data should stay in your building, and a generic tool keeps missing the point.

A custom workflow can make sense if you have a repeatable process that no packaged product handles well: a particular estimating method, a quality hold point unique to your product, a way you hand work from the office to the floor. Building, in this sense, does not have to mean hiring a data-science team. It can mean a small, documented workflow on top of tools you already have, with your own examples and your own checks.

Building has costs that do not show up in a software quote. Someone has to maintain it. Someone has to retrain it when your process changes. Someone has to own it when the person who set it up is on holidays. If you cannot name those people, you are not ready to build. Buy something simpler, or wait.

The EMC session promised a framework for this decision and for spotting where AI can produce a real return. It did not, and could not, tell you what your return will be. That depends on a baseline you measure yourself, on one task, before anything is installed. The same caution shows up in How Long Does AI Implementation Actually Take?: the calendar is driven by the work you chose and the data you already have, not by the tool's brochure.

What you should not buy or build

Some work should stay human even if a vendor says their system can do it.

Price, scope, warranty language and delivery promises belong to a person who can be held to them. A Saskatchewan vendor, Prairie AI, puts that boundary bluntly in its note that AI can help prepare an estimate packet, but the estimator still owns the judgment. That is a vendor's advice, not a survey. It still holds up. "Included," "should be straightforward," and "we can start Monday" are easy for a tool to write and expensive for you to unsay.

Safety-critical calls, customer commitments and anything that changes the contract should not be delegated to a system, bought or built. The tool can draft a background brief. A person signs the result.

The useful question isn't "can we get an AI tool?" It's "which job is this for, and who still signs the output?"

If you cannot answer those two questions in one sentence, you are not choosing between build and buy yet. You are still choosing a job.

Three tests before you spend

Use these on any proposal, including one that comes from inside your own shop.

Is the task named and measured? "Be more efficient" is not a task. "Cut the time to assemble a standard quote package from four hours to two, without changing the review step" is a task. If you do not know how long it takes today, you will not know whether the tool helped.

Is a mistake cheap and visible? A bad first draft of a meeting summary is cheap. A bad quantity, a bad ship date or a bad sign-off is not. Buy or build only where the next human check will catch an error before it leaves the building.

Do you already have the inputs? A tool cannot fix a quoting process that lives in three spreadsheets and someone's inbox. The same is true of quality records and shipping documents. If the source material is a mess, clean up the folder structure and the naming first. National construction figures, summarized in Construction Sits Near the Bottom of Canada's AI Adoption Numbers, keep circling the same point: fragmented information, not a lack of tools, is often the thing in the way.

A Regina shop that sells into construction should also be honest about volume. If you quote a handful of large jobs a year, a custom estimating build may cost more than it saves. If you quote the same kind of work every week, a modest bought tool, used with a tight review, is usually the better first move.

What to do with this

Pick one recurring job in the office, not on the floor: a quote package, a production update, a closeout checklist or a quality report. Write down who does it, how long it takes, what "done well" looks like, and what must never be sent to a third-party tool.

Then make the build-or-buy call in writing, in a short paragraph. If you buy, name the tool you will try, the review step, and a date to keep or drop it. If you build, name the owner, the first version's scope, and what you will not try to automate. If you cannot make either case, leave the task alone for now.

Give it four weeks, then compare the result to the baseline you wrote down. The answer will be more useful than any product demo, and it will tell you whether the next task should be bought, built or left with the people who already know how to do it.

Explore the assessment →