AI integration and automation
I automate the tasks that eat up hours every week, get your software talking to each other, and set up AI assistants that know your own documents. No magic promises: one problem at a time, measured before and after.
Muleky, an app built by HEMGE. Fictional team and players.
AI assessment
With a half-day assessment: we list your repetitive tasks, measure how long they take and pick three costed priorities together. You leave with a clear plan, whether or not you work with me afterwards.
Most businesses without an AI plan think it doesn't apply to them. Yet many already use it without realizing, in their email or their accounting software. My job is to show you, in your own trade, what's worth doing now.
78.1%
of businesses with no AI plans say it isn't relevant to them.
Statistics Canada, Q3 2025
66%
of entrepreneurs were already using AI once shown a list of tools (only 39% thought they were).
BDC, 2024
A frequent, repetitive task
Daily or weekly, and easy to time.
Low-sensitivity data
Or data you can anonymize before handing it to a tool.
Someone who reviews
Nothing reaches a customer without a human check.
Visible results in under three months
Small and measurable beats a vague “transformation”.
You walk away with: an inventory of your tasks, three costed priorities (gain, cost, risk, data involved), recommended tools, the Law 25 points to watch and funding programs that may apply.
Depending on your size, part of the assessment may be subsidized (ESSOR, Investissement Québec). The application goes through the organization itself.
Real use cases
The best candidates are repetitive, frequent and follow clear rules. Here's what I set up most often, by department. And sometimes plain automation with no AI does the job: it's simpler and more reliable.
Before: copy the products, build the PDF by hand, send it, then re-enter everything in accounting.
Now: a request from your website becomes a draft quote based on your price list. You review it, send it, and it turns into an invoice in one click.
Web form · database · PDF generation · accounting API
Before: supplier invoices arrive as PDFs or photos, and someone retypes them one by one.
Now: amounts, dates and suppliers are extracted automatically and sent to your accounting software, ready to approve.
Language-model extraction · QuickBooks · automatic reminders
Before: the same questions about hours, lead times and prices, by phone and email, every single day.
Now: an assistant answers from your documents, says clearly that it's automated and hands off to you what it doesn't know.
Website assistant · knowledge base · human hand-off
Before: Excel files nobody has time to analyze, a monthly report pieced together by hand.
Now: an always-current dashboard, and an AI that sums up trends and answers your questions in plain language.
Data cleanup · web dashboard · verified summaries
Before: the newsletter and social posts always come last, for lack of time.
Now: a first draft in English and French in minutes, which you rework so it sounds like you. Proofreading stays human.
Prompt library · brand templates · review
Before: procedures live in two employees' heads; training a new hire takes weeks.
Now: your notes and videos become clear procedures, then an internal assistant that answers new hires' questions. AI helps write job postings; hiring stays a human decision.
Procedures · document assistant · job postings
Process automation
I map your process as it really runs, time it, then build the automation and test it with you. After 30 days, we measure the time saved.
Admin tasks are what Quebec SMBs automate first, with a median productivity gain of 17% and 62% of projects paying for themselves in under three years. (CFIB and Investissement Québec, 2025; automation in general, not just AI)
Real example · PoPadema
From request to QuickBooks invoice
Request
Customer form or email
Catalogue
Current products and prices, no copying
PDF quote
Generated with your branding
You approve
A person signs off before it goes out
Send and track
Email, status, follow-up
QuickBooks invoice
Converted in one click, no re-entry
PoPadema is the app I built for a cleaning-solutions company: creating, sending and tracking purchase orders and quotes, syncing with Zoho CRM, then converting them into QuickBooks invoices. Any data shown in my demos is fictional.
API integrations
An API is a door that lets two pieces of software talk to each other directly. A request filled out on your website can create the customer in your CRM, a draft quote and a follow-up task, without anyone retyping a thing.
Older software, or tools that don't talk? There's often still a way: an automatic export, structured emails or a small in-between database. I propose the simplest solution that holds up, and I document everything so you're not dependent on me.
66% vs. 32%
Growing SMBs are twice as likely as declining ones to have well-integrated tools.
Salesforce, Small & Medium Business Trends Report, 2025 (survey by a software vendor)
QuickBooks
Customers, products, invoices
Zoho CRM
Contacts, leads, follow-ups
Stripe
Payments, subscriptions
Your business, at the centre
Node.js and MongoDB for custom code; Make or n8n when an existing tool is enough.
Google Workspace
Calendar, Gmail, Drive
Microsoft 365
Outlook, Excel, Teams
AWS SES
Transactional email
AI assistant
A good AI assistant answers only from your documents (FAQ, policies, products, procedures), says clearly that it's automated, and hands off to a person when it doesn't know. It's the AI use Canadian businesses plan for most.
+14%
more customer requests resolved per hour with an AI assistant, and +34% among newer employees.
Brynjolfsson, Li and Raymond, Quarterly Journal of Economics, 2025
You're responsible for what it says. In 2024, Air Canada had to compensate a customer over wrong information given by its chatbot. Hence my safeguards.
Moffatt v. Air Canada, 2024 BCCRT 149
In Free Board Society, the online board game platform I built, each creator configures their own AI agent: its name, instructions, welcome message and what it's allowed to do. For your business, the logic is the same: you stay in control of what the assistant says and does.
Data interpretation
AI can clean up your data, write formulas, summarize a monthly report and spot trends. I build the dashboard that brings your numbers together, then connect an AI that explains them in plain language.
One rule never changes: always check the source numbers. AI can get a calculation or an interpretation wrong, and sound very sure of itself.
36.6%
of Canadian businesses that use AI use it to analyze data. It's the number one use.
Statistics Canada, Q2 2026
Prompt training
A three-hour or full-day workshop on your real documents is often enough for a team to handle everyday uses on its own. Each participant leaves with a prompt library tailored to their role.
< 1 in 10
SMBs has access to formal AI training; nearly half of employees who use AI have never been trained.
Diversity Institute and Future Skills Centre, 2026
72%
of micro-businesses roll out AI with their current staff, without hiring a specialist.
Diversity Institute, 2026
Training may sometimes be subsidized by Services Québec (Workforce Training Measure). Rates should be confirmed with a business advisor.
Prompt example
Vague prompt
Write a follow-up email.
Result: generic text that everyone recognizes instantly.
Precise prompt
You are the office manager of a small plumbing company in Lac-Saint-Jean. Write a follow-up email for a quote sent 10 days ago. Warm, professional tone, 80 words maximum. End with one simple question. Do not invent any price or date.
Result: text you can almost send as is, after a quick review.
Data and Law 25
You can use AI and still comply with Law 25, as long as you pick the right plans and set written rules. The problem isn't AI: it's the free version used with no guidelines.
I'm not a lawyer: I provide practical groundwork and work with your legal advisor when needed.
No names, emails, numbers or identifiable files in a free or personal version of an AI tool.
Anonymize, or replace names with codes, before asking for an analysis.
Business plans or the API for customer data, with a short privacy impact assessment.
A human in the loop for any decision that affects an individual (hiring, credit, refusals).
One or two pages, presented to the team in 30 minutes:
OpenAI help pages and Anthropic's terms (August 28, 2025 update), checked in October 2026. Terms change: check before you choose.
53%
of Quebec workers have no clear guidelines from their employer on AI.
NETendances 2025, Université Laval
63%
of Quebec internet users worry about the privacy of their data in generative AI.
NETendances 2025, Université Laval
Since September 22, 2024, all of Law 25 is in force. Before sending personal information outside Quebec, which most cloud AI tools do, you must assess the risks and be able to justify your choice in writing.
Government of Quebec; Raymond Chabot Grant Thornton, 2024
Realistic gains
Serious studies measure real but modest gains: roughly one to two hours a week per person. Over a year and a whole team, that adds up. Be wary of anyone promising “ten times the productivity”.
2.2 h
per 40-hour week, on average, for workers who use generative AI.
Federal Reserve Bank of St. Louis, 2025
26 min
a day on writing, email and meetings, in a trial with 20,000 civil servants.
Government of the United Kingdom, 2025
17%
median productivity gain in Quebec SMB automation projects.
CFIB and Investissement Québec, 2025
2%
of organizations see a return on generative AI. The difference: a specific problem and a measurement.
KPMG Canada, 2025
How it works
One process at a time, in five steps. “Complete transformation” projects are the ones that go off the rails most often.
We pick the process that costs you the most time.
We note the time and frequency, to compare later.
A first working version within days, on your real data.
Testing, security, backups, documentation and user training.
We compare. If the gain is real, we move to the next process.
Tools and technologies
Common questions
The short answers. For costs, funding and tools in detail, see the full FAQ.
Full AI FAQ for SMBsIt depends on the process and the software involved. Connecting two applications is a few days of development; a custom tool, like a quote generator or an assistant, is a bigger project. That's why I start with an assessment: you know the expected gain and the cost before you decide.
Definitely not. I automate one process at a time, we fine-tune it, then move to the next. And a person keeps the sign-off at critical steps, like sending a quote or a payment: automation should save you time, not take away your control.
No. Moving a website request into your CRM or sending an appointment reminder is classic automation, simpler and more reliable. I add a language model only where it brings something: reading a free-form email, extracting invoice data, writing a first draft.
Yes, and confidently. That's why prices come from your price list, not from the AI, and why a person reviews any document containing numbers, commitments or legal content before it reaches a customer.
Yes, depending on your size. ESSOR (Investissement Québec) can cover up to 50% of a digital assessment that includes AI for businesses with at least $2.5M in revenue, and Services Québec can subsidize part of the training. I can point you in the right direction; the application goes through the organization.
Related services