Fast or finished: measuring AI by the finished work

BDC’s Small Business Week starts October 19, and this year’s theme doesn’t leave much room for comfort: “fast or finished.” (source)

I read BDC’s June study on SME digital transformation with that line in my head. (source) It surveyed 1,500 Canadian SMEs earlier this year, and it says a lot about the gap between using AI and getting something out of it.

The plateau

30% of Canadian SMEs used generative AI in 2025. In BDC’s 2024 research it was 28%. BDC describes adoption as having changed little since 2024. (source)

I’d have guessed higher, given how often assistants and agents now appear in software pitches. But adoption alone tells us little about whether the work is getting better.

Using it vs using it well

BDC gives each surveyed SME a digital maturity score out of 100. It’s built from the technology a business uses, how it manages and integrates data, its planning and culture, and how intensely it uses AI. Only 23% of SMEs land in high or very high. (source)

BDC also estimates the relationship between digital maturity and productivity, measured as sales per employee. Low to moderate maturity is associated with 41% higher productivity, and moderate to high with 36%. (source)

Those estimates cover digital maturity across the business. They do not isolate the effect of AI or show that a particular AI workflow caused a productivity gain. My takeaway is practical: adoption is a starting point. We still need to check what happens to the work.

That’s how I read “fast or finished.” Speed here isn’t the number of tools you’ve bought. It’s how quickly you get from “people are using it” to “we know it’s working.”

What this means for a services firm

A caveat up front. The BDC study doesn’t break its results out by industry, so there’s no professional services number in it. What follows is me applying general SME findings to the firms I spend most of my time with: agencies, consultancies, accounting and advisory practices.

With that said, the pattern seems to fit services firms well. A lot of what they deliver is documents. Proposals, decks, reports, memos, models. Generative AI is very good at producing documents.

BDC flags a version of this risk. In the section on generative AI for personal productivity, it describes an “adoption paradox”: producing more emails, documents and messages can leave businesses with less of a productivity gain than they expected. Its advice is to focus on faster turnaround and fewer errors rather than more activity. (source)

For a firm whose product is documents, more documents doesn’t automatically mean more value. Think of a proposal. AI produces a draft in ten minutes. A senior person then spends two hours checking the facts, fixing the argument and making it useful to the client.

If you only time the draft, you miss most of the work. To know whether AI saved time, you need to compare the whole job with how it was done before, including review and rework, and check that the quality held up.

That’s why I’d ask teams about the finished deliverable. How much total time did it take? What needed fixing? Was the result ready for the client? Those answers tell you where AI helps and where it needs more work.

Getting from fast to finished

BDC’s advice for getting more out of AI is short: plan, train, and improve the quality of your internal data. (source)

For a services firm I’d add one thing. You need some way to see what AI actually produced and whether it was any good. Prompt counts and active seats tell you about activity. The proposal that went out and the report the client signed off on tell you about the work.

That’s the angle we’re taking at Sabia: AI spend vs Work Produced, with the evaluation done on the finished work itself (the document, the commit, the message) rather than on the agent’s trace of producing it. We’re in waitlisted early access and still learning a lot about what firms want to see first.

Pick one recurring piece of work, such as a proposal, a report or a client deck, and follow it all the way through. Compare the total time and the finished quality with how the job was done before. Use what you find to decide what to keep, what to improve and where to focus review.

Ahead of Small Business Week, that’s the question I’d bring to the team: where is AI shortening the path to client-ready work, once review and rework are included?