If you have asked three AI agencies in the GTA what they charge, you have probably got three versions of the same answer: it depends on your requirements, let’s book a discovery call.
That answer is not always evasive — scope genuinely does vary. But it is not much use when you are trying to work out whether to take the meeting at all. So here are actual numbers.
The four price bands you will run into
Band 1 — DIY tools: $0–$200/month. Zapier, Make, an off-the-shelf chatbot. These are genuinely useful and you should not dismiss them. What you get is automation: rules you write yourself, running when you tell them to. What you do not get is judgement. A rule cannot look at a 40-day-overdue invoice from a good customer and decide to be gentle about it.
Band 2 — Productised AI services: $300–$2,500/month. A fixed platform configured to your business, usually with a setup fee between $700 and $8,500. This is the band Orbys sits in, and it is where most businesses between 15 and 200 people should be looking.
Band 3 — Custom agency builds: $25,000–$80,000 one time. A bespoke system built for you. It can be excellent. Two things to check before you sign: who maintains it in eighteen months when the model API changes, and whether you own the code.
Band 4 — Enterprise platforms: $100,000+. Ignore this band unless you have a compliance department.
What the setup fee is actually paying for
This is the part that confuses people, because the software itself is not what costs money. Anyone can point a language model at a spreadsheet in an afternoon.
The cost is in making it yours. That means someone sitting in your building for half a day watching how work really moves — which is never how the process document describes it. It means loading in your products, your pricing logic, your suppliers, your quality standards, and the twenty exceptions that live in one person’s head. It means connecting whatever systems currently hold fragments of the truth, and testing the output against your last three months of real situations until it produces something you would have signed yourself.
Skip that work and you get a chatbot that sounds confident and is wrong about your business. That is worse than nothing, because now your team does not trust it.
The arithmetic that decides it
Forget the vendor’s ROI slide. Do this on the back of an envelope:
One. How many hours a week does your team spend on work that involves no judgement — copying between systems, chasing the same invoice, rewriting near-identical quotes? Be honest. In most small businesses it is between 10 and 25.
Two. Multiply by a loaded hourly rate. In the GTA, $30–$45 an hour is realistic once you include employer costs.
Three. Add the deals you lose to slow response. If you quote in three days and a competitor quotes the same afternoon, you are losing jobs you never knew you were in the running for. One $1,000 job a week is $4,000 a month.
Four. Add the cost of money sitting in someone else’s account. Thirty days of receivables is real, and it is expensive if you are financing anything.
If that total is comfortably above what an AI service would cost you monthly, the decision is easy. If it is close, do not do it yet — fix the process first, then automate it. Automating a bad process just makes you wrong faster.
Questions worth asking any vendor
- What happens if the AI sends something wrong to a customer? The correct answer involves a human approval step, at least at the start. If they tell you it never makes mistakes, walk away.
- Where does my data live, and is it used for training? In Canada, and no. PIPEDA makes this more than a preference.
- What is the contract term? Month to month is the honest answer for a service that is supposed to prove itself.
- What happens if I leave? You should get your data, and they should delete their copy.
- Can you show me it running, on numbers like mine? Not a video. Live.
Our numbers, since we are asking others to publish theirs
Orbys is priced by the size of your company, and published:
| Tier | Employees | Monthly | Implementation | Custom builds / yr |
|---|---|---|---|---|
| Starter | 15–25 | $650 | $3,500 | 1 |
| Growth | 26–75 | $1,100 | $5,500 | 2 |
| Established | 76–200 | $1,900 | $8,500 | 4 |
Plus HST. Every tier includes all sixteen catalogue agents, and your AI and voice usage within fair-use limits — we absorbed that rather than passing it through, because an unpredictable line item is worse than a slightly higher predictable one.
The part worth understanding is the last column. There is no charge for having an agent or for running one, in any tier, ever. What we charge for is building a new agent from scratch for a process that only your business has — and your plan already includes one to four of those a year. An extra one is $1,800, once, maintenance included.
That is deliberately the opposite shape of Band 4. An agency will build you one custom agent for $25,000–$80,000 and hand you the maintenance forever. We are betting that including the build is what makes a client stay, which means we only make money if the thing keeps working.
If you have never heard of us, do not start here. Start with a 30-day pilot: $2,500, one process, one custom agent built on your real data, with the success criteria written down before we begin. If it works, the $2,500 comes off your implementation. If it does not, you keep the diagnostic and the agent anyway. The pilot is explained in full here.
We are new in Canada and we are not going to pretend otherwise — which is exactly why we would rather prove it on one process than ask you to believe a case study you cannot verify. If you want to know whether any of this is worth it for your specific business, book thirty minutes, and if the answer is no we will tell you on the call.