Home / Resources / AI in Action

AI in Action

A practical guide for Quebec SMB leaders. Your employees are probably already using AI: four steps to take the lead.

By Bruno Poirier, AI adoption specialist · Version 2.0 · November 2026

Foreword

AI is already in your company

When I meet an SMB leader, the conversation almost always starts the same way: "We haven't started with AI yet, where do we begin?" Very often, that isn't accurate. The company has already started. It just doesn't know it.

Your assistant summarizes her meeting notes with ChatGPT. Your sales rep has his quotes rewritten. Someone in accounting figured out how to make a spreadsheet talk. Nobody asked for permission, nobody got any training, and nobody really knows what data is leaving the company.

So the real question is no longer "should we start?". It's "are you the one leading this shift, or is it happening without you?"

I wrote this guide for leaders who want to take back the controls without launching a big technology project. It starts from what's already happening in your team, then offers four steps: take stock, set rules without blocking, train everyone starting with yourself, and only then transform a process. That's the order that works in the SMBs I support.

I've worked in data and technology since 1999, including several years leading data and AI projects at a large mining company. What I learned is that technology is rarely the problem. What makes the difference is the order in which you do things and the room you leave for people.

Each section ends with a few concrete actions to take within the week.

Bruno Poirier
AI Adoption Specialist, DécodAI

Section 01

Your employees already use AI

Before asking where to start, look at what has already started without you.

Official statistics give the impression that businesses here are moving slowly. Employee surveys tell a different story.

12%

of Canadian businesses report using AI to produce their goods or services.

Statistics Canada, second quarter of 2025 (12.7% in Quebec according to the ISQ)

51%

of Canadian employees say they use generative AI in their work.

KPMG Canada, Generative AI Adoption Index 2025

The two figures don't measure the same thing: the first counts businesses that have built AI into their production, the second counts people who use it day to day. But the gap speaks for itself. A good share of the AI used at work goes through personal accounts that management doesn't see.

It's even more pronounced in SMBs. According to Microsoft and LinkedIn's 2024 Work Trend Index, 80% of AI users in small and medium-sized businesses bring their own tools to work. And in Canada, nearly half of employees who use AI say they've received no training (Future Skills Centre, 2025).

The good news

The interest is already there. You won't have to convince your team that AI can help them: several have already seen it for themselves.

The risk

Nothing is governed. Client data can end up in free tools, and good discoveries stay in one person's head.

Typical case

A 25-employee distributor runs a short anonymous survey. She learns that three people in customer service use the free version of ChatGPT to write their replies to complaints, pasting in the customer's name and order. The replies go out faster, and nobody had told her, for fear of being told to stop. She now has a good idea to roll out and a real risk to fix: that's exactly the starting point this guide proposes.

Key takeaways

  • Your company has probably already started with AI, with no plan and no rules.
  • Your starting point isn't a blank page, it's what your team is already doing.
  • The real challenge: moving from hidden use to directed use.
Section 02

Why now

AI isn't a fad. It addresses pressures you're already dealing with.

You surely have good reasons to wait. The pressures on your business won't.

Workforce

Positions that are hard to fill. AI takes over part of the repetitive tasks so your people can focus on what requires their judgment.

Margins

Fewer costly errors and less wasted time, without cutting quality or service.

Competition

Your competitors are already trying. You can catch up, but the longer you wait, the wider the gap.

Clients

They expect faster, more personalized answers, without you having to hire.

AI strengthens your teams, it doesn't replace them

In an SMB, AI takes on the thankless work: sorting, data entry, first drafts, looking up information. The time freed up goes to judgment, client relationships and decisions. The data confirms this for now: in Canada, about 6% of businesses that have adopted AI report a reduction in staff, a proportion that has remained stable (Statistics Canada, 2026).

The real challenge is to adapt roles and help your people build their skills. That's also why training is at the heart of this guide.

AI doesn't replace your best employee. It removes what keeps them from being your best employee.

DécodAI

Key takeaways

  • AI tackles your real pressures head-on: workforce, margins, competition, client expectations.
  • Its role is to complement human work. Canadian data shows no wave of AI-related layoffs.
Section 03

AI demystified, 2026 edition

Forget the technical categories. Here's what you'll actually come across in your business.

"AI" covers very different tools. For an SMB leader, four families are enough to find your way.

Built-in

The AI already built into your tools

Copilot in Microsoft 365, Gemini in Google Workspace, the AI features added to your accounting software, your CRM or your management system. You may already be paying for them.

Starting point: look at what your current licences offer before buying anything.

Chat

Conversational assistants

ChatGPT, Claude, Gemini, Copilot. You write to them in plain language and they draft, summarize, translate, analyze. Each one comes in a free version and an enterprise version.

What matters: enterprise versions generally offer better guarantees on how your data is used. Free versions, not always. Read the terms.

Auto

Automation and machine learning

Systems that learn from your past data to forecast, classify or spot what's out of the ordinary, and automations that move information between your software without copy-pasting.

Examples: forecasting demand, detecting an abnormal invoice, transferring an order from email to the management system.

Agent

Agents

Assistants that can chain several steps: find information, fill in a document, send a draft for approval. They're arriving in mainstream tools, but they still need close supervision.

Keep an eye on: this is what's coming. Better to master the first three families first.

Pitfall

A conversational assistant can be wrong with great confidence. It produces what looks right, not necessarily what is right. A human always validates before acting.

What AI is, and what it isn't

The best way to calm both fantasies and fears is to put things in black and white.

What AI isWhat AI isn't
A tool that helps with writing, analysis and decisions.A replacement for your employees or your managers.
A system that spots trends, anomalies and repetitions.A tool that understands your context like an experienced employee.
An accelerator for repetitive tasks and first drafts.A colleague you can leave to work without supervision.
A way to reduce errors when someone checks the result.A guarantee of accuracy: it can be wrong, and make things up.
A technology you integrate step by step, at your own pace.An instant fix for an organizational problem.
A tool that's only as good as the data you give it.A tool that works well with messy data.

Key takeaways

  • Start with the AI you're already paying for, before buying a new tool.
  • Free version or enterprise version: the difference is what happens to your data.
  • AI predicts, it doesn't know. Someone always validates.
Section 04 · Step 1

Take stock of what's already happening

Before choosing a project, know who uses what, for which tasks and with which data.

You can't lead what you can't see. The first step takes a week: discover the AI already circulating in your company. Tone makes all the difference.

If your team senses a witch hunt, they'll go quiet and keep doing it in secret. Instead, present the process as a search for good ideas: you want to know what works so you can share it, and what's risky so you can secure it.

Two simple ways to do it

The anonymous survey

Five questions in an online form, ten minutes to answer. Anonymity lets people say what they really do.

The team meeting

Half an hour around the table: "Who already uses AI, and for what?" Start by describing your own use, it sets the tone.

The five questions to ask

  1. Do you use an AI tool for your work, even occasionally? Which one, and with which account (personal or company)?
  2. For which tasks? Give one or two concrete examples.
  3. How much time do you estimate you save per week?
  4. What kind of information do you put into it: general text, internal documents, client or employee data?
  5. What would help you use it better: training, a company-provided tool, clear rules?

With the answers, you'll know which uses deserve to be shared, which risks to fix right away (client data in a free tool, for example) and which curious people can serve as relays in step 3. This inventory can also be done with an outside perspective, which often helps employees talk more freely about how they use AI.

To do this week

  1. Announce the process to your team in one sentence, stressing that no one will be blamed.
  2. Send the survey or schedule the meeting.
  3. Sort the answers into two columns: to roll out, to secure.
Section 05 · Step 2

Set rules without blocking

Banning doesn't work. Offering an approved path does.

The natural reflex, after the inventory, is to ban. It's the worst option: use continues, but in secret, and you lose the visibility you just gained.

Simple rules, on one page

No need for a thirty-page policy. A sheet everyone can read in two minutes is enough, as long as it clearly answers four questions:

  • Which tools are approved? Name them. Anything not on the list isn't used for work.
  • Which data never goes in? For example: personal information about clients or employees, confidential financial data, documents provided by a client under confidentiality.
  • Who validates? A human rereads and takes responsibility for everything that comes out of AI before it goes to a client or is used for a decision.
  • Who do people ask? A designated person, so that a doubt doesn't get settled in secret.

Choose an approved tool

Rather than letting everyone use their own free version, offer an official tool, ideally an enterprise version that protects your data. If you already have Microsoft 365 or Google Workspace, check what's included in your licences: it's often the shortest path.

Law 25: what really concerns you

There is not yet an AI-specific law for businesses in Quebec or Canada. But Quebec's private-sector privacy act, modernized by Law 25, already applies to how you use AI:

Automated decisions

If a decision about a person is based exclusively on automated processing, you must inform them and allow them to present their point of view (s. 12.1). When a human genuinely takes part in the decision, this specific obligation doesn't apply.

Data outside Quebec

Disclosing personal information outside Quebec first requires a privacy impact assessment (s. 17). Pasting client data into a tool hosted elsewhere is exactly that case.

Purpose

Personal information collected for a specific purpose can't be used for anything else without consent, except where the law provides otherwise (s. 12).

Other issues

Intellectual property: read the tools' terms of use. Fairness: watch for bias, especially in hiring. Labour standards: AI isn't used to monitor your employees without their knowledge.

Source: Commission d'accès à l'information du Québec. This guide is not legal advice: if in doubt, consult a professional.

To do this week

  1. Write your rules sheet by answering the four questions above.
  2. Check which AI features are included in your current licences.
  3. Fix the risks spotted in the inventory first, especially client data in free tools.
Section 06 · Step 3

Equip and train the whole team, starting with yourself

This is the step most companies skip. It's also the one that makes the biggest difference.

AI adoption has long been presented as a project to choose. In reality, it's first a skill to develop, in everyone.

Start with yourself

A leader who uses AI in their own work gives others permission. And it shows: in BCG's AI at Work 2025 survey, the share of employees who view generative AI positively rises from 15% to 55% when leadership clearly supports it. Use it to prepare your meetings, summarize a report, draft a delicate reply. Then talk about it with your team.

Train everyone, not just a project team

The same BCG survey shows that regular use is clearly higher among employees who received at least five hours of training, with in-person training and coaching. Five hours isn't a university program. It's a half-day of training and a few follow-up sessions.

Needs will surface on their own

In Quebec, difficulty pinning down specific business needs is the obstacle cited most often: that's the case for 73% of companies according to a study by the Conseil du patronat du Québec published in 2024. Leadership is often asked to imagine use cases for a tool it barely knows. Flip the logic: once the team knows what the tool can do, ideas come from the people who do the work.

Set up simple routines

  • Five minutes at a team meeting to share a good use from the week.
  • A curious person, spotted during the inventory, who acts as a relay and answers questions.
  • Protected time to learn during work hours, not just in the evening.

To equip each person in their day-to-day use, DécodAI's Guide to Context-Driven AI offers a method and ready-to-use prompts.

To do this week

  1. Use the approved tool for three of your own tasks, then tell your team how it went.
  2. Schedule a first training session for everyone, at least half a day long.
  3. Name your relay and add use-sharing to your meeting agendas.
Section 07 · Step 4

Transform a process

Only now, choose a project. It will be fed by everything the first three steps brought to the surface.

Adding AI to an inefficient way of working gives you a slightly faster inefficient way of working. The real gain comes when you rethink the task.

It's the clearest finding of the major surveys. In McKinsey's State of AI 2025 report, the organizations that get the most value from AI are nearly three times more likely to have fundamentally redesigned how they work. This data comes mainly from large organizations, but the idea applies even better in SMBs, where a process can be redrawn around a table in a morning.

Plugging in AI

Each customer service employee asks an assistant to write their replies, one at a time, as before.

Rethinking the task

Requests are sorted and prioritized as soon as they arrive, a standard reply is prepared for simple cases, and the team spends its time on the difficult ones.

The 5 criteria of a good first project

Run every idea the team brings up through this filter. If it doesn't tick all five boxes, keep it for later.

1

Impact

Does the problem have a measurable effect on costs, quality or the experience of clients or employees?

2

Frequency

Does the task come up every day or every week? The more often it recurs, the sooner the gain arrives.

3

Complexity

Can it be rethought, tested and rolled out in a few weeks?

4

Risk

Is the risk acceptable? Avoid critical or heavily regulated processes to start.

5

Available data

Do you already have a minimum of usable data, even if imperfect? AI needs something to stand on, not a perfect database.

Processes well suited to a first project

Sort

Incoming requests

Emails, service requests, quote requests: classified, prioritized and routed to the right person as soon as they arrive.

What changes: you no longer read everything to find what's urgent, the urgent comes to you.

Docs

Repetitive documents

Meeting minutes, inspection reports, quotes, project follow-ups: pre-filled from your notes or summarized in a few minutes.

What changes: the person reviews and adjusts instead of starting from a blank page.

Alert

Anomalies

Duplicate invoices, price discrepancies, data that's out of the ordinary: flagged before the error gets expensive.

What changes: you check the exceptions instead of checking everything.

Second wave. Sales or purchasing forecasts and staff scheduling give good results, but they require clean data and enough history. Count on several months rather than a few weeks, and keep them for after a first success.

How to measure

Forget calculating return to the dollar in the first month. Instead, track what really moves: the number of people using the tool each week, the time recovered on the chosen task, the errors avoided and what the team thinks of it.

Financial assistance to explore

ESSOR, component 1: digital diagnosis (Investissement Québec)
Reimburses up to 50% of an outside firm's fees for a digital diagnosis and plan, including AI integration, up to $20,000. For companies with 250 employees or fewer and revenue of at least $2.5 million. Program in effect until March 31, 2027.
Productivité-Compétences (Commission des partenaires du marché du travail)
A $55 million program for 2026-2028, including $20 million for AI and innovation training in SMBs. It goes through collective sponsors: employer associations, sector workforce committees, business groups. Ask your association.
Services Québec
Your regional office can provide financial support for training your employees. Talk to a business advisor.
Eligibility rules and calls for projects change often. Check with the organization before committing to an expense: most programs refuse expenses incurred before the application.
Section 08

Six ways to get your AI shift wrong

The mistakes we see most often in SMBs, and how to avoid them.

AI attracts its share of empty promises and pushy vendors. Most failures, however, come from within.

1 · Banning it and thinking the problem is solved

Use doesn't stop, it becomes invisible. You lose control over the data and good ideas stay hidden. Set rules rather than ban.

2 · Starting from the technology

"We want to do AI" isn't a project. Buying a tool because it's trendy leads straight to the drawer. Start from a real task.

3 · Launching a pilot in a vacuum

A targeted project while the rest of the team is neither trained nor guided is an island. It stays fragile and doesn't spread. Train everyone first.

4 · Thinking too big

The big program meant to transform everything at once gets bogged down. One process at a time, then the next.

5 · Neglecting the data

No reliable data, no results. If your data is a mess, that's where the real project begins.

6 · Believing shock figures and miracle solutions

You may have read that "95% of AI projects fail." That figure comes from a 2025 MIT report, widely cited but disputed: small sample, measured after six months, mostly large companies. It says nothing about your SMB. Also be wary of anyone promising results without effort, without training and without data.

Key takeaways

  • Banning hides the problem, setting rules solves it.
  • A project succeeds when the team around it is trained and equipped.
  • A promise with no effort and no data is a red flag.
Section 09

Cheat sheet

The whole guide on one page. Keep it handy.

1

Take stock

Find out who uses AI, for what and with which data. Without looking for culprits.

2

Set rules without blocking

One page of rules, an approved tool, Law 25 in mind.

3

Train everyone, starting with you

Lead by example, offer real training, let ideas surface.

4

Transform a process

Chosen with the 5 criteria, rethought rather than automated as is, measured simply.

Three reflexes to keep

  • Start from what people already do, not from a technology.
  • A human always validates what comes out of AI.
  • Measure use and time recovered before talking about return.

30days

What's realistic in a month:

  • an inventory of existing uses;
  • a page of rules and an approved tool;
  • a first training session for the whole team.

Not a transformed process. That's the following month, and you'll do it on solid ground.

References

Sources

Consulted in October 2026. Sources marked (FR) are in French.

  1. Statistics Canada, Analysis on artificial intelligence use by businesses in Canada, second quarter of 2025, and Economic and Social Reports, 2026. www150.statcan.gc.ca
  2. Institut de la statistique du Québec, Adoption et utilisation de l'intelligence artificielle par les entreprises au Québec en 2024 et en 2025 (FR), November 2025. statistique.quebec.ca
  3. KPMG Canada, Beyond AI adoption: Turning Canada's AI momentum into measurable returns, March 2026 (Generative AI Adoption Index 2025). kpmg.com
  4. Microsoft and LinkedIn, Work Trend Index 2024. microsoft.com/worklab
  5. Future Skills Centre, Bridging the AI Skills Gap in Small and Medium-Sized Organizations in Canada, September 2025. fsc-ccf.ca
  6. BCG, AI at Work 2025: Momentum Builds, but Gaps Remain, June 2025. bcg.com
  7. McKinsey, The state of AI in 2025: Agents, innovation, and transformation. mckinsey.com
  8. Conseil du patronat du Québec, Adoption de l'intelligence artificielle : les entreprises du Québec ne savent pas par où commencer (FR), December 2024. cpq.qc.ca
  9. MIT NANDA, The GenAI Divide: State of AI in Business 2025, August 2025.
  10. Commission d'accès à l'information du Québec, Principaux changements apportés par la Loi 25 (FR). cai.gouv.qc.ca
  11. Investissement Québec, ESSOR, volet 1 : études de faisabilité et diagnostic numérique (FR). investquebec.com
  12. Gouvernement du Québec, Productivité-Compétences : un nouveau programme pour stimuler la productivité de nos PME (FR), February 2026. quebec.ca

It starts with a conversation.

I come to you, at no cost, to understand your situation before proposing anything.