Data · what can ai do for my business
What can AI actually do for my business? Seven jobs, priced
AI removes repeated work, not decisions. Seven jobs it does well in an Ontario business, what each saves, and the $2,500 to $12,000 that a build costs.
Lasse Pettersen
AI can do any job in your business that is repeated, written down, and checked rather than argued about. That is the whole answer. Everything below is detail.
It cannot make a decision your business has never made, it cannot be accountable to a customer, and it cannot automate a process that lives only in somebody’s head. Those three limits eliminate more candidate projects than cost does.
Statistics Canada asked Canadian businesses that were not adopting AI why not. Cost came fourth, at 10.6%. Security came third, at 13.4%. The largest answer, from 40.0% of them, was that AI is not relevant to what they do. This page exists because that answer is usually wrong, and wrong in a specific way: the owner is picturing a chatbot on the website and correctly concluding it would not help.
The seven jobs that actually come up
Ranked by how often they turn out to be the right first build in an Ontario business of ten to fifty people.
1. Somebody retyping the same data into two systems
The most common and the most reliably profitable. An order arrives by email, gets typed into the quoting system, then typed again into accounting. Ten to twenty five hours a week in a business that has grown past its original software.
What it saves: the hours, plus the error rate, which is the part nobody costs. What it costs: $2,500 to $12,000 for workflow automation, driven by how many systems have to talk. When it does not work: when the two systems disagree about what a customer record is. Fix that first or you are automating a mess at speed.
2. Quotes that take three days because one person writes them all
A single estimator is the bottleneck in an enormous number of Ontario contracting and supply businesses. Drafting a quote from a template, priced from a rate table, checked by a human before it goes, is a job a machine does in ninety seconds.
What it saves: turnaround, which converts. A quote that arrives the same day wins work a quote that arrives Thursday does not. When it does not work: when every job is genuinely bespoke and the price comes from experience rather than a table. Then you are automating the wrong half; automate the document assembly and leave the number to the estimator.
3. Calls that go unanswered after five
An intake agent answers, collects the details, checks a calendar and books, at any hour. Covered in full in what an AI intake agent does and does not do, including the consent rule that applies because the agent transcribes.
When it does not work: under about two calls a day, or where every call needs judgement about price.
4. The same forty questions, every week
Internal or external. Where does this part number cross to, what is our warranty position, which form does a new subcontractor fill in, what is the lead time on that. An assistant grounded in your own documents answers from what you actually wrote rather than from the internet.
The distinction that matters here is between a chatbot and an agent, because they are priced and scoped differently. That is set out in the difference between an AI chatbot and an AI agent.
5. Reading documents that arrive as paper or PDF
Invoices, purchase orders, packing slips, certificates, timesheets photographed on a phone. Extracting the fields and writing them into a system is now reliable enough to be boring, which is the correct standard for this kind of work.
When it does not work: when the documents are genuinely unstructured and the extraction has to be right every time with no human check. Build the check in and the economics still work.
6. Month end, rebuilt by hand every month
Somebody exports three spreadsheets, joins them, and formats a report. Usually a day and a half. Usually the same day and a half for years.
The honest note: this is real and it is frequently not worth doing first, because a day and a half a month is about eighteen days a year and the build competes with candidates saving ten hours a week.
7. Tracking dates that cost you a contract when they lapse
Subcontractor insurance, worker qualifications, equipment certificates, WSIB clearance. Two hours a week of tracking, and the value is not the two hours.
Why this one is different: it is insurance rather than labour saving. A contractor working inside the Sarnia petrochemical complex, a supplier to the mines around Greater Sudbury or a firm in the Pickering energy supply chain has qualification obligations set by a customer far larger than itself. One crew standing down at a gate costs more than the build.
What the national numbers say about who is doing this
Two Statistics Canada figures, and they point in different directions.
Business use of AI to produce goods or deliver services reached 19.2% in the second quarter of 2026, up from 6.1% two years earlier. Worker use of generative AI reached 35.9% in March 2026, measured through the Labour Force Survey. Worker use is running at roughly twice business use, which is covered in the gap between worker and business AI use. If that describes your business, the first thing to buy is not a build. It is a written boundary, which is what AI policy and staff training covers at $1,500 to $4,000 including a half day of training.
The occupational spread inside that worker figure is the useful part:
| Occupation group | Used generative AI, March 2026 |
|---|---|
| Management | 75.1% |
| Natural and applied sciences | 67.5% |
| Natural resources and agriculture | 17.0% |
| Trades, transport and equipment operators | 14.7% |
By industry it runs from 65.6% in professional, scientific and technical services down to 16.3% in accommodation and food services.
Read that table the way an owner should. It does not say the tools do not work in the trades. It says the tools that reached the trades so far are the ones that write text, and the office is where the text is. The work available in a trades business is in the back office, on the phone and in the documents, which is exactly the list above.
What AI will not do for your business
Worth stating plainly, because most pages selling this will not.
It will not decide something you have never decided
An automation encodes a rule. If your business has no rule, the build will invent one, and it will be whatever the person writing the specification assumed on a Tuesday. Writing the process down is the first project, and it typically removes about a third of the process before any software is involved. That argument is made at length in which jobs to automate first.
It will not reduce your headcount by itself
It removes tasks. What happens to the freed hours is a management decision, and in most small businesses the honest outcome is that the same people do more of the work that was being neglected, rather than that anybody leaves.
It will not make marketing work that was not working before
This is the one to be careful about, because it is what most AI marketing tools are sold on. Statistics Canada put AI marketing automation at 19.8% of AI-using businesses in the second quarter of 2026, down from 23.1% a year earlier. It is the only major application on that list that fell, which is what happens when people buy a machine that produces posts and discover that the posts were never the constraint. The repeating parts of the job are worth automating and the reasoning is a page of its own: marketing automation, and what it can and cannot fix. If the real constraint is hours rather than ideas, the order to do it in is in how to market a business when you have no time.
It will not fix an integration problem you have not admitted to
If your CRM and your accounting system were never designed to talk, an automation can carry data between them and it can also paper over a decision you should be making about software. Both outcomes look the same for about a year.
Where to start
Measure one task for one week. Not a study, a notebook. Minutes per occurrence, times occurrences per week, times fifty, times your loaded hourly cost. Compare against the published build ranges in what AI consulting costs in Ontario.
If you would rather have that done properly across the whole business, that is the $999 assessment: an inventory of every repeated task, the three worth doing first with hours and dollars against each, real subscription costs, a build estimate for each, and the ones I recommend you do not do. Ten business days, and the fee comes off any build you go ahead with.