Most businesses already have tasks that AI could do well. Someone reads every incoming email to decide who should handle it. Staff search a shared drive for the right policy. Invoices arrive as PDFs and get typed into a spreadsheet by hand. Our AI integration services add a language model to the tools you already use, so those tasks take less time.
We connect a model from a provider such as OpenAI, Anthropic or Google to your existing website, app or workflow, and we write the code that controls what it sees and what it does.
Tasks that AI does well
Language models are good at reading and writing text. These are the jobs where they are most useful for a business:
- Answering questions from your own documents. The model first searches your manuals, policies or help articles, then writes an answer from what it found, with a link to the source. This method is called retrieval. It keeps answers based on your own content.
- Drafting and summarizing. First drafts of replies, quotes or reports, and short summaries of long emails, calls or case files. A person reviews the draft before it goes out.
- Sorting and tagging. Reading incoming messages, forms or support tickets and labelling them by topic, urgency or the team that should handle them.
- Extracting data from documents. Pulling names, dates, amounts and line items out of PDFs, scanned forms and emails, and saving them to your database or accounting tool.
- Search that understands meaning. A search box that finds “how do I cancel” when the page says “ending your subscription”.
If your task needs the AI to take actions in other systems, such as booking an appointment or issuing a refund, see our page on AI agent development.
Start with a small test
AI works very well on some tasks and poorly on others. It is hard to tell which before you try it with your own data. So every project starts small:
- We pick one task and agree what a good result looks like.
- You share a set of real examples, such as past emails, documents or questions, with personal details removed where possible.
- We build a simple test version and run it on those examples.
- We show you the results side by side with the correct answers, plus the cost per use.
- You decide whether to build it into your product, change the approach or stop.
We price the test on its own. If the results are not good enough, you stop there and keep what we learned.
Choosing a model provider
The main providers are OpenAI (the company behind ChatGPT), Anthropic (Claude) and Google (Gemini). There are also open models that you can host on your own servers. We compare them on your task, looking at:
- How accurate the answers are on your examples.
- The cost per use.
- The provider’s data terms, including whether your data is kept or used for training.
- Where the data is processed.
We write the integration so the model can be swapped later. Prices and quality change often, and you should not be locked into one provider.
What it costs to run
Providers charge by the token, which is a small piece of a word. You pay for the text you send to the model and the text it writes back. A short answer costs very little. Reading a 50-page PDF on every request costs much more.
We keep running costs down by sending only the text the model needs, using smaller and cheaper models for simple jobs such as tagging, and saving answers that repeat. We also set a monthly spending cap with the provider, so a bug cannot run up a large bill.
Privacy under Canadian law
AI tools do not sit outside privacy law. If a model handles personal information, PIPEDA applies, or a similar provincial law in Quebec, Alberta or British Columbia. The Office of the Privacy Commissioner of Canada says that a business that sends personal information to another company for processing stays accountable for it. The business should also tell customers when their data may be processed outside Canada. Canada’s privacy regulators have also published principles for generative AI.
In practice, we:
- Send the model only the data the task needs.
- Remove names and contact details before sending text, where the task allows it.
- Choose provider plans that do not train on your data.
- Log what was sent, so you can answer a customer who asks.
- Help you update your privacy policy to mention the AI provider.
We are not lawyers. For sensitive data, such as health or financial records, check the plan with your own legal advisor.
Testing answers for accuracy
A demo that works on five questions can fail on the fiftieth. Before launch, we build a test set from your real examples and check the model’s output against it. We run the same test again whenever we change the instructions or the provider releases a new model version. After launch, we add a simple thumbs up or down on answers, so your team can flag bad results and we can fix the cause.
How to get started
A full build is scoped as a custom web app project. Prices are on our pricing page. If your data sits in several systems, API development may come first.
Contact us with the task you have in mind and a rough idea of the data involved. We will tell you whether AI is a good fit and what a small test would cover.
