Which process should you automate first?
The one that runs often, is written down, and can be checked in under a minute. Volume gives you evidence, a written rule gives the system something to follow, and fast checking keeps a mistake cheap. Everything else is a second project.
Statistics Canada puts business use of AI at 19.2 percent in the second quarter of 2026, tripled from 6.1 percent two years earlier. Among the firms using it, the leading applications were data analytics at 36.6 percent and text analytics at 34.5 percent. Reading and sorting. The unglamorous end of the building.
What makes a process a good first candidate?
Four conditions decide it, and a process that fails two of them is a research project carrying a deployment budget.
- It runs weekly at least, daily if you can get it. Ten examples a year will teach a system nothing.
- Someone can write the rule in ten steps without booking a meeting to settle disagreements about what the rule is.
- A person can check one output in under a minute and know whether it is right.
- A mistake is reversible before it reaches a client, a regulator or a bank account.
Count them honestly. Four is a build. Three is workable. Two is a pilot, and one is a strategy deck with a vendor logo on it.
Why the process everyone names first is the wrong one
The process a leadership team names first is almost always the one that hurts most. The annual bid. The onboarding that eats three weeks every time. The account nobody except one person can explain.
Those processes hurt because nobody ever wrote them down. The rule lives in the head of whoever handles the exceptions, and it changes depending on the client. Automating one means documenting it first, under a deadline, while the people who own it defend their judgment across the table. That is an organizational argument with a software invoice attached to it.
A first system should make a boring process faster. It should not be asked to settle a contested one.
| Works as a first system | Wait |
|---|---|
| Supplier invoice coding against a fixed chart of accounts | Annual bid strategy |
| Intake and triage of permits, claims or service requests | Hiring and promotion decisions |
| Quotes assembled from a published price book | Key account relationships |
| Pulling dates and obligations out of signed contracts | Any process whose rule lives in one person’s head |
Measure three things before you touch it
Take three numbers before anything changes. How long one unit takes now. How often it comes out wrong now. What it costs per unit now.
Skip that and you will own a system nobody can prove worked, and the month six review becomes a conversation about how people feel. The first two numbers come out of the readiness assessment. The third is the basis of cost per decision, which is the only pricing view that survives contact with production volume.
Start with the boring one
Pick the process your best people describe as tedious, and run it for ninety days with the numbers kept. You will learn more about your own data quality, your own exception rate and your own review habits in that quarter than any assessment will tell you. Then take that to the process that actually hurts, which is the sequence that keeps a first project from turning into a pilot that never ships.
Data analytics and text analytics. Reading and sorting. That is where the 19.2 percent of Canadian businesses using AI actually landed, and it is where a first ORKA AI build starts too.
Sources
- Statistics Canada, Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026. Reference period 1 April to 6 May 2026.
Software is only the surface. Infrastructure is the rest.
