AI adoption, done well, is narrow and measurable. We look for repetitive, high-volume tasks where a tool can draft, sort, extract or summarise with a person checking the result, we test candidates against a sample of your real work, and we only recommend the ones whose payback holds up. Alongside that, we write plain guidelines so your team knows which AI tools it may use and which information must stay out of them.
Who this is for
This service suits businesses that want results from AI rather than experiments. Typical starting points:
- staff already using free AI tools with no agreed rules
- an office team spending hours on data entry from emails, invoices or forms customers fill in on paper
- quote, proposal or report writing that follows a predictable pattern
- an owner pitched an "AI platform" with no way to judge it
- a backlog of follow-ups, reminders and status updates done by hand
What is included
- Task inventory. We list repetitive tasks across the business and estimate weekly volume and minutes per item.
- Candidate scoring. Each task is scored on volume, rule clarity, error tolerance and data sensitivity.
- Payback model. For the top candidates we calculate hours saved, tool cost and the time needed for human review.
- Accuracy test. We run each shortlisted approach on a sample, typically 50 real items, and record what it gets right and wrong.
- Tool comparison. Where several products could do the job, we compare them on accuracy, cost, data handling and integration.
- AI use guidelines. A short document covering approved tools, permitted uses, prohibited data, review requirements and who to ask.
- Rollout plan with a named owner and a 60-day review date.
Our guide to automation payback walks through the arithmetic with a worked example, including what happens when you account for review time honestly.
Privacy and official guidance
Privacy comes up in every AI conversation, and we take it seriously without pretending to be lawyers. In December 2023, Canada's federal, provincial and territorial privacy commissioners issued principles for responsible, trustworthy and privacy-protective generative AI technologies, covering topics such as consent, appropriate purposes, openness, accountability and safeguards. The Office of the Privacy Commissioner of Canada also summarises how PIPEDA applies to private-sector organisations.
We use this guidance to frame questions for tool vendors and for your own policy. We do not give legal or compliance advice; your lawyer or privacy professional should confirm what applies to your business.
Timeline
| Weeks | Activity |
|---|---|
| 1–2 | Task inventory and interviews |
| 3 | Candidate scoring and payback models |
| 4–6 | Accuracy tests and tool comparison |
| 6–7 | Guidelines, rollout plan, handover |
Smaller scopes, such as writing AI use guidelines alone, can be done on an hourly basis in a few sessions.
How results are measured
- hours per week saved on each automated task, measured against the baseline we recorded
- error rate after human review, compared with the manual process
- tool cost per month against the value of hours saved
- staff awareness of the guidelines, checked with a short quiz or review at 60 days
A short example of candidate scoring
Imagine an office that handles 300 supplier invoices a month, at about 6 minutes each for data entry. That is 1,800 minutes, or 30 hours, a month. If a tool cuts entry to 1 minute and review adds 1.5 minutes, each invoice now takes 2.5 minutes: 750 minutes, or 12.5 hours. The saving is 17.5 hours a month. At a loaded cost of $40 an hour that is $700 a month, which is then weighed against the tool's price and setup time. The figures are illustrative; the structure is how we compare candidates.
Mistakes to avoid
- Automating a broken process. If the manual process is unclear, the automated version will be unclear faster. Sometimes the right first step is process redesign; ProcessImprovement.ca covers that discipline in depth.
- Ignoring review time. A tool that saves four minutes but needs three minutes of checking saves one.
- Testing on cherry-picked examples. Accuracy on five tidy samples says little about the messy fiftieth.
- No owner. Every automation needs someone who notices when it stops working.
To talk through a task you suspect could be automated, text (604) 632-4959 or call or email. See the pricing page for fee options, or the full list of services for related work such as data and reporting.
Frequently asked questions
Do you build AI tools?
No. We identify and test use cases, compare available products, and set the rules around them. Where custom work is needed, we help you scope it and choose who should build it.
Is it safe to put client information into AI tools?
That depends on the tool, its settings, its contract terms and your obligations. We point you to official guidance, including the privacy commissioners' principles on generative AI, and build those questions into tool reviews. Conclusions about legal obligations belong with your lawyer or privacy lead.
How quickly do automation projects pay back?
It varies widely. We calculate an expected payback for each candidate before anything is purchased, using your own volumes and time estimates, and we recommend dropping ideas that do not clear a reasonable threshold.
Our staff already use AI tools informally. Is that a problem?
It is common, and it is a good reason to write guidelines. A short policy that says which tools are approved, for what, and with which data usually reduces risk more than a ban.