The pitch usually sounds the same. A clean deck, the word "agentic" six times, and a promise to automate half your business by Friday. A lot of the shops selling AI automation in 2026 did not exist eighteen months ago, and a fair share of them are reselling someone else's platform with a markup and a logo swapped in. You do not need a technical background to tell the builders from the middlemen. You need about ten blunt questions, asked before you sign anything. Here are the ones that separate a real partner from a demo merchant.
Who Is Actually Going to Build This?
The person running your sales call is often not the person who will write a single line of code. Plenty of agencies are a marketing front with the work quietly subcontracted overseas, which is how a project you were promised in three weeks turns into three months of broken telephone. Ask directly: who writes the code, is any of it subcontracted, and who do I message when something breaks at 7am on a Saturday?
Get a name and a calendar link for the person doing the work, not just the closer. If the answer gets vague, that is your answer. A studio that builds its own work will happily put you on a call with the builder before you pay a deposit.
- Is the person selling this the person building it?
- What gets subcontracted, and to whom?
- Who is my direct contact after launch, and how fast do they respond?
Do You Own It, or Are You Just Renting?
This is the question that catches the most people. Some agencies rent you access to a platform they control and hold the code, the prompts, and your data on their own accounts. The system feels like yours right up until you stop paying, at which point the whole thing evaporates and there is nothing to migrate. That is a subscription wearing a custom-build costume.
Insist on plain terms in writing: on full payment you own the source code, the data, the prompts, and the IP, and it runs on your infrastructure and your accounts. Ask what happens the day you part ways and what you walk out with. If the honest answer is "nothing," you are buying a dependency, not an asset.
What Does "Done" Mean, in Numbers?
Most automation projects that disappoint were never given a target to hit. Pin down one measurable outcome before work starts, or you will have no honest way to judge the result. Vague promises of "10x efficiency" are noise.
Say you run a plumbing company and roughly a third of your inbound calls hit voicemail while crews are on jobs. Call-tracking data consistently shows most callers who reach voicemail just dial the next plumber. A real success metric is concrete: previously-missed callers get a text back inside 60 seconds, and you measure how many of them book. Now "done" is a number you can check on a Tuesday, not a feeling. Nail down the metric, the support arrangement after go-live, and how quickly a broken piece gets fixed.
Has Anything They Built Survived a Year?
Demos are easy. Anyone can wire up something that looks brilliant for ten minutes on a screen share. Staying reliable under real traffic, at month eleven, when an integration changes and a customer sends an input nobody planned for, is the hard part and where most vendors quietly fall over. Ask for something they built that has been live in production for twelve months or more, and ask to see it running.
We hold ourselves to the same test. We built Zebra Landscaping a quoting tool that took their estimating from about four hours down to under twenty minutes, and it has to hold up through an Edmonton spring when the quote requests pile up faster than anyone can type. For an Alberta contractor we built a system that cut bid assembly from two or three hours to minutes. Those are systems that run, not slides. Bid Pro's, our government-bid and proposal platform, is launching soon on the same principle.
Get the Whole Number, Including Month Thirteen
The figure on the deck is almost never the figure you pay. In 2026 a custom build commonly lands anywhere from roughly $15,000 to well past $100,000 CAD depending on scope, with monthly support somewhere between about $1,000 and $5,000 (illustrative ranges, not a quote). The trap is the quote that covers the happy path and nothing else.
Here is an illustrative walk-through. You are quoted $18,000 for a custom booking assistant. The QuickBooks and Google Calendar integration you assumed was included is not, add $6,000. Your dispatch team needs a different flow than the front desk, add $4,000. Then a $1,200-a-month retainer runs to $14,400 over the year. The $18,000 on the deck was really about $42,000 in year one. None of that is dishonest if it is disclosed. So make them disclose it: what is not included, what does year two cost, and what triggers a change order?
When the Right Answer Is Don't Build Anything
An honest partner will sometimes talk you out of the project, and you should trust the ones who do. If a task takes two minutes and happens a few times a day, a custom build will not pay for itself for years. Do the arithmetic before you fall in love with the idea. Automation that saves five hours a week at a $50 internal hourly cost is worth about $13,000 a year, so a $15,000 build pays back in a little over a year. Save only one hour a week and the same build takes six years to break even. Don't build it.
A lot of what gets pitched as "custom AI" is already solved by tools you can buy this afternoon. Jobber, built right here in Edmonton, handles scheduling and invoicing for trades. Stripe handles payments, QuickBooks handles the books, and a Calendly-style link handles bookings. Custom work earns its keep when your workflow needs real discovery, messy data cleaned up, systems that do not talk to each other stitched together, or a human approval step in the middle. If an off-the-shelf app covers it, the smart Alberta operator buys the app and spends the saved money on winter marketing before quoting season.