Choosing an AI Partner for Your SMB
An SMB owner who wants to "do AI" typically receives three types of offers today. A software product with an AI feature activated by subscription. An agency pitching a pilot project. An IT provider offering to connect AI to the systems the business already uses.
All three can work, but they don’t answer the same question. Here are the criteria for choosing an AI partner, and the questions to ask before you sign anything.
Quick answer: A good AI partner starts with your process, not their tool. They tell you in writing where your data goes and whether it’s used to train models, they know your current systems, they spell out what you own, they price out the operating cost on top of the project cost, they’re willing to start small, and they’re upfront about what AI gets wrong.
1. First, Know What Problem You’re Buying
Most AI projects that fail in an SMB fail before the technology. They fail because nobody wrote down what needed to change.
A good request looks like this: "Our clerks spend about two hours a day re-entering purchase orders received by email into the management system, and we want to cut that down." A bad request looks like this: "We want an AI assistant for the team."
The first one is measurable. You know how many hours you’re recovering, you know when it’s done, you know whether it’s working. The second one isn’t measurable, so it never ends. A competent partner will bring you back to the first formulation. If they accept the second without asking questions, that’s a red flag.
Three questions to ask yourself before calling anyone:
- Which specific process is eating up time or generating errors?
- How many times a week does that process run? A process that runs three times a month doesn’t justify custom development.
- Who in the company will own the outcome once the tool is delivered? Without an internal owner, the tool gets abandoned within months.
2. The Five Use Cases That Pay Off in an SMB
Profitable SMB projects look a lot alike. They almost always involve text, documents, or repetitive tasks.
Extracting information from documents. Supplier invoices, purchase orders, delivery slips, forms received as images, quotes. The tool reads the document, pulls out the fields, and a human validates. This is the most mature use case and the easiest to measure.
Drafting a first pass. Routine email replies, product descriptions, meeting notes, job postings. The gain comes from quickly reviewing a draft, not from publishing automatically.
Searching your own documents. An employee asks a question in plain language and gets an answer drawn from your procedures, contracts, or technical documentation, complete with the source reference.
Sorting and routing. Classifying incoming requests, routing them to the right person, flagging urgency, generating a summary. The gain shows up in response time.
Supporting sales and service. Preparing a client file before a call, summarizing history, producing a follow-up. The gain is in prep time.
What almost never pays off on a first project: training a model from scratch, building an autonomous agent that makes decisions without oversight, or replacing an entire management system.
3. Seven Criteria for Choosing an AI Partner
They start with the process, not the tool. A useful first meeting focuses on how you operate: how many files per day, what exceptions come up, who approves things, where the hours disappear. A meeting where they show you a platform before understanding your process is selling a product, not a solution.
They tell you where your data goes, and they put it in writing. This is the most important criterion, and the one fewest vendors bring up on their own. Four questions to ask: where are the models hosted and in which country are the servers; does the content you send get used to train the vendor’s models; how long is the content retained and by whom; what contractual clauses cover personal information protection. An acceptable answer lives in the contract, not on a marketing page.
They know your current systems. AI only has value in an SMB when it’s connected to what already exists: email, office suite, accounting system, CRM, file server. A partner who doesn’t know how your data flows delivers a standalone tool, your employees end up copy-pasting to use it, and the tool gets abandoned.
They tell you what you own. Three things need to be clear in writing: the custom code produced, your data and the outputs generated, and the automation itself. If it lives inside the vendor’s proprietary platform, you won’t be able to take it with you. That isn’t necessarily a dealbreaker, but you need to know it before you sign.
They price out the operating cost, not just the project. An AI project has two budgets: the build, which is one-time, and the operation, which is monthly. Operations include model usage, hosting, monitoring, and patches when an upstream system changes. Ask for a monthly estimate based on volume, with the volume assumption written down, and ask what happens if volume doubles.
They’re willing to start small. A first project should deliver something usable within a few weeks, covering a single process, with a human validating results. The goal isn’t performance, it’s learning three things: quality on your real data, adoption by your employees, and the actual operating cost.
They talk about what AI gets wrong. Models sometimes produce wrong answers with confidence, they’re sensitive to the quality of input data, and they change behavior when the vendor updates the model. If nobody talks to you about the limitations, nobody has thought through what happens when the tool is wrong.
4. The Law 25 Point Almost Everyone Forgets
Every Quebec SMB is covered by the Act Respecting the Protection of Personal Information in the Private Sector, as amended by Law 25. Two provisions directly affect an AI project.
Section 17 requires, before sharing personal information outside Quebec or entrusting its processing or storage to someone outside Quebec, a privacy impact assessment and a written agreement. An AI tool connected to your client files processes personal information, so the obligation applies.
Section 12.1 applies to any company that uses personal information to make a decision based exclusively on automated processing. It must then inform the person that the decision was automated, provide on request the information used and the main factors that led to the decision, and give them an opportunity to submit observations to someone who can review it.
The design consequence: choose a tool whose main factors you can explain, and keep a human in the loop on decisions that affect people.
5. Ten Questions to Compare Two Proposals
Ask each vendor the same questions and write the answers side by side.
- Which specific process is targeted, and what indicator will measure the outcome?
- What data will be processed, and does it include personal information?
- Where are the models and data hosted, and in which country?
- Is the content sent used to train models, and where is that stated in the contract?
- Which existing systems will be connected, and through what mechanism?
- Who owns the code, the data, and the automation?
- What is the build cost, and what is the monthly operating cost at a given volume?
- What is the first usable deliverable, and when?
- What happens when the tool is wrong, and who validates?
- Who provides support after delivery, and on what timelines?
The vendor who answers all ten in writing isn’t necessarily the cheapest. They’re the one whose risk you can actually assess.
6. How We Approach These Projects
OKTO Solutions is a managed IT services provider based in Trois-Rivieres, at 994 boulevard du Saint-Maurice, working with SMBs across Quebec. Our AI offering starts from the same place as the rest of our work: the systems you use every day.
In practice, that means custom application development, business process automation with application integrations, and AI tool integration for businesses. Every project starts from a concrete business need and delivers a usable tool, hosted securely. We also manage the surrounding infrastructure: access controls, backup, endpoint protection, and the documentation you need for your compliance obligations. The details are on our IT services page.
Frequently Asked Questions
Do you need perfect data to start an AI project?
No, but you need to know where it lives. A documented process and organized files are worth more than a large volume of disorganized data. Inventorying your sources is often the most useful first step.
Will AI replace employees in an SMB?
Profitable SMB use cases eliminate tasks, not positions. Extracting data from documents and drafting first passes frees up hours spent on tasks nobody finds interesting.
Can you use AI without sending your data outside Quebec?
It’s possible, depending on the use case and the type of model chosen. This is an architecture decision to make at the start of the project, not a checkbox at the end, and it determines the scope of the assessment required under Section 17.
A Measurable First AI Project, in Mauricie
The right first project is small, connected to a system you already use, and measured by a simple indicator. That’s how we work with SMBs in Trois-Rivieres and Mauricie, keeping infrastructure and compliance in the same file. Our managed IT services cover automation, integration, and the security that goes with it. To choose your AI partner and frame your needs, reach out through our contact form or call 450-231-3836.
Reading about AI is one thing. Connecting it to your own data is another: artificial intelligence in business, custom AI application development and our IT services in Quebec City.
