Waterloo is a harder market to be useful in
In most places, an AI consultancy adds value by knowing more about AI
than the client does. That is not the situation in Waterloo. The
insurers, financial services firms, and technology employers here
employ people who can read a model card, and they have already sat
through the vendor pitches. Turning up with enthusiasm about large
language models is not a service.
What we've found these organizations are genuinely short of is
something less glamorous: a single accountable person deciding which
of the many possible AI initiatives is worth a quarter of engineering
time, which vendor claims hold up under contact with a real workload,
and what the organization's actual position is when a regulator, a
board member, or a large customer asks how AI is being governed.
The governance question arrives before the build question
Waterloo's employer base is unusually concentrated in insurance,
financial services, and health technology — three sectors where the
constraint on AI isn't capability, it's permission. Where can this data
go? Who reviews an automated decision? What happens when the model is
confidently wrong about a claim?
We start regulated engagements by mapping those boundaries, because a
design that ignores them has to be thrown away later. Sometimes that
means a self-hosted model instead of an API. Sometimes it means keeping
a human in the loop on anything consequential and automating only the
preparation. Occasionally it means concluding that a particular use case
isn't worth the exposure. All three are legitimate answers.
Why a fractional executive, rather than another consultant
A consultant delivers a recommendation and leaves. A
fractional CAIO or CTO stays on the
hook for whether it worked. For most mid-sized Waterloo organizations
the AI leadership problem doesn't justify a full-time executive salary,
but it does need continuity — someone in the room for the vendor
review, the roadmap argument, and the board update, quarter after
quarter.
That's the shape of most of our Waterloo work, and it's why this page
looks different from our Kitchener
page, where the ask is more often to automate a specific operational
process.
Our own bias, stated plainly
Since 2022 we've run HeitechSoft as a working experiment in applying AI
across our own business, which means our recommendations come from
having lived with the tooling rather than from reading about it. It also
means we've abandoned things that didn't hold up. We'll tell you about
those too — the failures are more instructive than the wins, and you'll
get a more honest read from a partner who admits to some.
