QuarterlyMay 1, 20268 min read

HeitechSoft Quarterly: Spring 2026

February to April 2026

This quarter, we turned AI conversations into systems that clients can test, and in each one AI does the reading, research and drafting while a person keeps the decision. Our lead story is a language-training company whose sales system was built and tested in five weeks, with a second phase signed on April 20. Our lab notes cover an AI tool that reads mill test reports and hands uncertain results to a person, a tender-matching proof of concept built in days, our own AI prospecting, and when AI is not the answer. Our wins include a finalist spot for the Guelph Chamber of Commerce’s Innovation Award, a hands-on AI workshop for a network of realtors, and an AI hiring tool we built for ourselves that now screens for others. This issue covers five industry developments, including BDC financing for smaller firms adopting AI and Microsoft’s Copilot Cowork.

Introduction

This quarter, conversations about AI turned into systems our clients can click through and test, most of them built in weeks. In each one, AI does the reading, research or drafting, and a person keeps the decision.

The biggest surprise came from our own business. In February the AI we use to research and write to prospects booked more meetings than we could take, so we paused it for about ten days. The lesson for any owner: the limit is rarely the AI. It is how many conversations your team can handle, so plan for that before you turn up the volume.

— Tom Heiber, CEO

From discovery to a tested sales system in five weeks

A language-training company ran its sales and onboarding across several disconnected tools. We presented our proposal as an interactive web page built with an AI app builder, and the founder signed it on March 11. Asked about privacy and what happens if they leave, we answered in writing: client data stays in a controlled environment, not outside AI platforms, and the company owns the system.

Five weeks after kickoff, the first phase was built and tested, with one record for every lead and a daily call list. AI suggests the next step for each lead. An "Enrich with AI" button researches a lead and shows each suggested change beside the current value, for a person to accept or reject. AI drafts follow-up emails, and a person sends them. The client's sales lead said the recommendations "often make me feel like, oh, I should have asked that." The switch from the old CRM is planned for May, and the client has signed a second phase.

What it means for you: a system shaped to your own sales process, and owned by you, now takes weeks, not a year. AI does the research and the first drafts, and your people keep every decision.

Lab notes

Reading mill test reports, with a person on the uncertain ones

For a pressure-vessel compliance start-up, we shipped 11 releases in nine weeks with an AI app builder. AI reads each mill test report and pulls out heat numbers, material grades and mechanical properties with a confidence score, and anything it can't read goes to a person. When automatic detection of document types proved inaccurate, we replaced it with a person's choice. A fabricator weighing a pilot told the founder that if the AI can really validate these reports, "this alone could save my guys hours of time."

A tender-matching proof of concept four days after the first site visit

A regional construction association's staff match tender notices to members by hand, and some take five minutes while others take an hour. Four days after a two-hour session on site, we sent a proof of concept of AI tender matching, and on February 25 the association signed an engagement. When access to its tender platform's data was held up, we redesigned the pilot to run on its own. The lesson: don't let one supplier's data access hold your plan hostage.

Our AI prospecting booked more meetings than we could take

We use our own AI to find clients. For each local business, it researches the company and the AI uses that have paid off in its industry, then drafts a personal first email. In our tests, OpenAI's models wrote the best subject lines and Anthropic's the best email bodies. It brought us the construction association above, and an M&A advisory now scoping an AI buyer-matching engine with us. Being local still matters: one foundation executive told us she said yes because we are in Cambridge.

Not everything needs AI, and what does needs checking

Part of our job is saying when AI isn't the answer. A franchisor referred by the Greater KW Chamber of Commerce asked us for two AI solutions. AI fits its monthly report card, compiled from online searches, but its franchise payment statements only need two existing systems connected. When AI is the answer, it needs checks: give a model the same prompt 100 times and about five answers go completely off, so our systems require structured output and have one AI check another's work.

Wins

The AI hiring tool we built for ourselves now screens for others

An HR consultancy has run its own branded copy since February 17 and used it for real hiring in March. Changes it asked for were live by the next morning. In April a fractional HR firm asked for a branded version to run a live search for a client in another province. We also used it for all of our own hiring this quarter, from a senior AI consultant to college interns. It judges video introductions from the transcript only, to reduce bias, and a person makes every decision.

A hands-on AI workshop for realtors, built on their own work

On April 15 we ran a three-hour, hands-on AI workshop for a network of realtors, organised by a local mortgage brokerage. Each module was built around the realtors' own tasks, from getting AI to write in each agent's own voice to a comparative market analysis from sample listing data. The organiser called the material "well presented" and "very detailed and thorough", and since some attendees asked for a more interactive session, we have proposed groups of six or fewer.

A finalist for the Guelph Chamber of Commerce's Innovation Award

On March 31 the Guelph Chamber of Commerce named us one of three finalists for its Innovation Award at the Guelph Awards of Excellence. It is independent recognition from a regional business community that our AI work is innovative. For owners deciding whom to trust with their AI, that outside view counts for more than our own claims.

By the numbers

93%
of the code changes in our own software projects this quarter (58 of 62) were made with an AI coding agent. Almost all our own software changes now come through an AI coding agent, so we recommend these tools from daily use, not theory.
6
AI engagements and project phases signed in three months. Signed work shows owners starting with defined phases they can review, rather than committing to one large AI project at once.
11
releases of a start-up's compliance platform in nine weeks, built with an AI app builder. Frequent small releases let a client see progress and redirect early, instead of waiting months for one big reveal.

AI for mid-market

BDC opens $500 million in loans to help smaller firms adopt AI

The Business Development Bank of Canada launched LIFT, with $500 million earmarked to help more than 1,000 small and medium-sized businesses adopt AI, with loans from $25,000 to $5 million and principal payments that can be postponed for up to two years (conditions apply). For a business of 20 to 200 people, that is financing built for AI projects that have been waiting on cash flow, and borrowers also get expert AI advisors to help identify opportunities and validate their value. Bring a specific use case with a measurable payback to that conversation, because that is what the advisors are there to test.

Source: BDC

CFIB: generative AI use rises with firm size, but half of businesses plan no AI spending

The Canadian Federation of Independent Business reported that 45% of businesses use generative AI, rising to 60% or more of those with 20 to 49 employees, and that 52% have no plans for AI spending in 2026. In a firm of 20 or more people, your staff are probably already using generative AI, whether or not there is a plan or budget for it. A short usage policy and some training turn that informal use into something you can manage and measure.

Source: CFIB

Statistics Canada: AI adopters are more productive, but data and skills explain most of the gap

A study by Innovation, Science and Economic Development Canada and Statistics Canada found that firms using AI had 16.8% higher labour productivity than non-adopters, but once earlier productivity and complementary capabilities such as data analytics and training were counted, the gap fell to 5.1% and was no longer statistically significant. For a business of 20 to 200 people, AI on its own is unlikely to move productivity much, because the gains go to firms that also have their data in order and their people trained. Budget for those alongside any AI tool, and measure results against how the work was done before.

Source: Statistics Canada

Microsoft's Copilot Cowork moves Copilot from answering to doing

Microsoft announced Copilot Cowork, which lets Microsoft 365 Copilot users describe an outcome and have Copilot plan and carry out tasks across email, meetings, messages and files, with users able to approve changes before they are applied. It started in research preview with a limited set of customers, and Microsoft said it would become more broadly available in its Frontier early-access program in late March. If your business runs on Microsoft 365, this reaches tools you already pay for, so test it first on one repetitive, low-risk task and tidy up who can see which files, since it works within each user's existing permissions.

Source: Microsoft 365 Blog

ChatGPT adds shared workspace agents for business teams

OpenAI launched workspace agents in ChatGPT, in research preview for its Business, Enterprise, Edu and Teachers plans. A team describes a workflow, and the agent runs in the cloud on a schedule or on triggers, with companies able to limit which data and tools it can reach and require human approval for sensitive actions. If your team built custom GPTs, this is their likely successor, and it can act in other apps, so watch the credit pricing that follows the free period until May 6, 2026, and require approval before any agent sends anything to a customer.

Source: Decrypt

If this quarter's conversations sound like your own, book a 20-minute conversation at https://heitechsoft.com/contact.

Get the next issue by email

Each quarter we share what we built, what we learned and the AI news that matters to mid-sized businesses.