Development and artificial intelligence
Artificial intelligence for SMBs in Quebec: task automation and AI agents
Artificial intelligence starts being useful to a small business the day it takes work off someone’s desk. OKTO Solutions starts from one specific task that costs you hours every week, hands it to an AI agent or an automation, then moves on to the next one. Your data stays yours, and you keep control over what the tool is allowed to do.
Supplier invoices For a Quebec SMB, artificial intelligence pays off mostly on reading and typing work: sorting incoming mail, pulling data off an invoice, drafting a standard reply, summarizing a meeting. The gain is measured in hours handed back each week, provided the chosen task is frequent and clearly described.
- Plenty of needs get settled with no artificial intelligence at all
The real gain
What does AI genuinely give back to a small business today?
Three things, and none of them are spectacular. First, time recovered from reading and typing: the hours spent copying information from one system into another go back to work nobody else can do. Second, consistency. An automation performs the same sequence every single time, so forgotten steps and typing mistakes disappear, and with them the downstream corrections that often cost more than the original task. Third, memory: a team finds in seconds what was sleeping in ten years of email and contracts. What artificial intelligence still does not do well is decide for you, carry a client relationship or repair a vague process. A poorly defined process, once automated, is still a poorly defined process, running faster.
- The gain is counted in hours per week, not in promises
- A frequent, well described task always beats a rare and prestigious one
Real examples
What it looks like in an ordinary work week
Six uses we put in place regularly at Quebec SMBs. These are not demos: they are tasks that were being done by hand the month before.
An email that becomes a ticket
A request lands in the shared mailbox. It gets read, filed by subject, attached to the right client, ranked by real urgency and assigned to the right person. By morning, the box is already sorted.
An invoice that files itself
The supplier invoice is read, the number, date, amount and taxes are pulled out, and the entry is prepared in the accounting system. Anything out of the ordinary is set aside for a human pair of eyes.
A meeting summary
The Teams call ends and the minutes and follow-up list are ready before everyone is back at their desk. Each follow-up carries a name and a date, not just good intentions.
A standard reply to a client
The same question comes in twenty times a week. The draft answer is prepared from your own documents, with the right timelines and the right conditions. A person reads it over and sends it.
Triage of incoming requests
Incoming requests get separated: what is urgent, what is waiting on a part, what belongs to another team, what is spam. Nobody loses thirty minutes a day doing that sorting by hand.
Search across your documents
A question asked in plain language, an answer drawn from your contracts, procedures and email, with a link to the source document so it can be checked. No more missing file on a Friday afternoon.
Automation
Business process automation: where it starts
Automation connects your software so information travels without copy and paste. It is the most profitable and least glamorous half of any business AI project. A report assembled every Monday morning, a sync between the CRM and accounting, a reminder sent three days before a due date: none of that needs a language model, only clear rules and clean plumbing. We always begin by measuring where the time actually goes, keep the task that pays back fastest, and decide from the outset what stays under human control. Artificial intelligence enters the picture only when the content has no fixed shape: a freely written email, an invoice from a supplier nobody has seen before, a question asked in everyday language.
- Pick the right task. We go through the repetitive work with you, the work that generates errors and the work nobody wants. We keep the one that pays back fastest and describe how it runs today, step by step.
- Build a first version. A prototype that handles real cases with your real data. You try it, you tell us where it gets things wrong, we correct it. This is where the call is made: the tool acts alone, or it proposes and waits for approval.
- Put it in service and log it. We connect the solution to your systems, train the people involved, and every run leaves a trace you can consult. An automation nobody can audit has no business running in production.
- Measure, adjust, move to the next one. After a few weeks, we compare the time recovered against what was expected. If the numbers are not there, we say so. If they are, we pick the next task.
Comparison
Which tasks are worth automating?
Six tasks we run into almost everywhere, what automation does to them, and the real effort involved. That effort depends mostly on the number of formats to handle and on the quality of the interfaces your software offers.
| Repetitive task | What automation changes | Effort to set up |
|---|---|---|
| Supplier invoice entry | Data is extracted and the entry is prepared in the accounting system. Unusual cases are set aside for review. | Medium. Varies with the number of suppliers and how clean the formats received are. |
| Sorting a shared mailbox | Each message becomes a filed request, attached to the right file and assigned. No more email forgotten at the bottom of the list. | Low to medium. A few weeks of tuning so the filing matches your vocabulary. |
| Answering questions that keep coming back | A draft is prepared from your official documents, with the right timelines. A person reads it over and sends it. | Low, if your documentation exists. High if it has to be written first. |
| A report built by hand every week | The report is generated from the sources and sent on its own, always at the same moment, always in the same format. | Low. No artificial intelligence required, just plumbing. |
| Meeting minutes | Summary and follow-up list produced in minutes, with an owner and a date on every point. | Very low. It is often the first trial we suggest. |
| Retyping between two systems | Information is entered once and travels in the right direction. The person who bridged the gap gets their mornings back. | Medium to high. It all depends on the interfaces the two products offer. |
Privacy
Does your data leave the building?
That is the first question to ask, and it comes before choosing a tool. When an employee pastes a client file into an artificial intelligence service, that information leaves the company and often travels outside Quebec. If it contains personal information, Quebec’s Law 25 treats the gesture as a communication to a third party: it calls for a privacy impact assessment before any communication outside Quebec, a written agreement with the provider, an entry in your register and a mention in your privacy policy. None of that prevents you from using AI. It simply forces the decision to be made in advance, rather than answered after a complaint or an access request.
- We write down which categories of information are allowed to leave, and which stay inside
- We choose services whose terms forbid training on your data
- We favour processing in a Canadian region when the provider offers it
- We strip identifying details where possible before sending, rather than shipping the whole file
- We record the use in the register and tie it to the person responsible for personal information
- We keep a trace of every run, so an access request can actually be answered
The obligations are laid out in detail on our page about Law 25 compliance for Quebec small and mid-sized businesses.
What nobody measures
Shadow AI: your employees are already using it
In nearly every company we look at, people are already using artificial intelligence tools without management knowing. A free account opened with the work email address. A quote pasted into an assistant for a rewrite. A competitor’s contract dropped into a tool to get a summary before a meeting. Nobody is acting in bad faith: these people are trying to do their job better with whatever they can find. The problem sits elsewhere. Information leaves the company with no agreement, no trace, and no way for anyone to say where it ended up. And the day that person resigns, their personal account walks out the door with them, along with the history of everything they put into it.
The picture takes a few days and serves to set the rules, not to punish anyone. In most cases, we end up offering an approved tool that does the same job better, and the unofficial use stops on its own.
- Sign-ups made on outside services with a company email address
- Browser extensions installed on workstations, often the real blind spot
How we measure shadow AI
- Connections to the major AI services, seen from the network and from the managed workstations
- Third-party applications authorized in your Microsoft 365 environment, and what they can reach
Guardrails
An AI usage policy, on one page people will actually read
An artificial intelligence usage policy serves one purpose: letting an employee know, in ten seconds, whether they are allowed to paste what is in front of them into the tool they just opened. Twenty-page documents written by a law firm never answer that question. We write ours with management, keep it to a single page, and phrase it in examples rather than principles. It names the approved tools, says what never leaves the company, explains what to do when a situation is not covered, and states that the output of an AI tool remains the responsibility of the person using it. Then we train the teams for an hour, and revisit the page twice a year because the tools move fast.
What the policy covers
- The list of approved tools, and how to get one added
- Categories of information that are off limits: employee files, health data, payment information, contracts under a confidentiality agreement
- The duty to read and stand behind whatever a tool produces before it goes to a client
- The cases where a human check is mandatory, such as a payment or a contractual commitment
- The rule on accounts: company tools, never a personal account for company work
- Who answers questions when a situation is not covered
Keep reading
Artificial intelligence touches everything else in your environment. These pages round out this one.
Questions and answers
Common questions about artificial intelligence at work
How can artificial intelligence actually help my small business?
What is the difference between automation and artificial intelligence?
Will our data be used to train a public model?
Does Law 25 apply when we send information to an AI service?
What is shadow AI and why is it a problem?
How do we find out which AI tools are used in our company?
Do you have to be a large company to get started?
What happens when the tool gets it wrong?
Name the task that costs you the most time
Tell us what is eating your weeks. We will say whether artificial intelligence is the right answer, whether a simple automation would do, or whether the process needs clarifying first. The first meeting commits you to nothing.