The best first automation is usually a repeated task with clear inputs, a recognisable good result, and someone who can check the output. A business does not need an AI strategy presentation to identify that task. It needs a close look at the work that keeps landing on a person’s desk.

For a small business in Winnipeg or elsewhere in Canada, a good starting point might be enquiry intake, preparing a weekly report, or collecting the information needed to start a customer project. The right choice depends on your volume and process. A task that is painful twice a year may be a worse first build than a small handoff repeated every day.

Compare three tasks before choosing one

Write down the trigger, the input, the person doing the work, the expected output, and the common exceptions. Then collect a representative week of timings. Include the time spent chasing missing information and correcting mistakes, not just the fast path.

Here is an illustrative example, not a client case study or a forecast of savings:

  • Lead intake: 40 enquiries a week, eight minutes each. Most need the same fields, but some are missing the project location or requested service. A person approves the outgoing reply. This is a candidate for a small pilot.
  • Monthly reporting: one report takes three hours, but the source files keep changing format. Standardising the inputs may be the first improvement. Building automation around an unstable input can create another maintenance task.
  • Quoting unusual projects: five quotes a week, with substantial judgement and different assumptions every time. Start with information gathering or a draft checklist; automatic pricing is a different and more consequential scope.

The point is to find a task whose boundaries you can explain. “Automate sales” is too large. “Create a draft CRM record from a complete enquiry and ask a person to approve it” is specific enough to test.

Decide whether AI is needed

Ordinary automation is good at explicit rules: a required field is absent, a date is due, a number exceeds a threshold, or a record needs copying between approved systems. Generative AI can help with more variable language, such as summarising a long message or preparing a draft response.

Keep authoritative values in their source system. If the customer’s contract says one thing and a generated summary says another, the summary should not silently become the source of truth. Give the reviewer a way to inspect the original information.

BDC’s guidance distinguishes improvements using existing business software from generative AI, connected automation, and custom development. That distinction is useful when you are deciding whether to configure something you already pay for or commission a new tool. Read BDC’s digital transformation guide.

Calculate recoverable time honestly

Using the example above, 40 enquiries multiplied by eight minutes is 320 minutes of current handling per week. Suppose a pilot handles 30 of those enquiries through a prepared draft. Each draft takes two minutes to review. The other ten still require the full eight minutes, and checking the automation takes another 20 minutes per week.

The new handling time would be 60 minutes of review, 80 minutes of manual work, and 20 minutes of monitoring: 160 minutes in total. The difference is 160 minutes, or about 2.7 hours of capacity recovered per week. At four weeks per planning month, that is about 10.7 hours. A four-week planning month is an explicit simplification; use your actual workload for the decision.

This is not automatically a payroll saving. The person may use that time to respond sooner, handle more enquiries, or do work that was being delayed. Put a dollar value on that capacity only if you can explain how the business benefits. Avoid counting the same benefit twice as both recovered staff time and extra revenue.

Subtract recurring software and model costs, support effort, and any additional review from the expected benefit. Keep the one-time setup cost separate. If the net monthly benefit is positive, dividing setup cost by that benefit gives a simple payback estimate. It still depends on the assumptions holding up in practice.

Write a pilot brief someone can build

The downloadable worksheet gives you a place to capture each of these decisions:

  • Trigger: what starts the task, and how often does it happen?
  • Input: which fields and documents are required, and who can approve access?
  • Output: what must exist when the task is finished?
  • Exceptions: what happens with missing information, duplicates, and failed updates?
  • Review: who can approve the result, and which actions must wait for approval?
  • Measurement: what will you compare with the current process?
  • Ownership: who checks failures and maintains the workflow after the pilot?

Use a small set of representative examples, including awkward cases. A pilot tested only on the cleanest enquiries can look excellent while avoiding the work that consumes the most time. Agree the sample and success criteria before judging the result.

Know when to leave the task alone

Wait if the process changes every week, nobody owns the result, access to the required data is unresolved, or the business cannot describe a correct output. Improve the process or clarify the responsibility first. An existing form, spreadsheet, or software feature may remove the bottleneck without a custom build.

Automation is worth pursuing when the task is stable enough to define, frequent enough to matter, and simple enough to evaluate. If the pilot needs as much correction as the original work, the result is useful evidence to change direction.

Turn the worksheet into a project

I build AI workflow automation and internal tools from Winnipeg for businesses across Canada. A filled worksheet is a better first message than a list of AI products you think you might need. It gives us a concrete task to discuss, with the assumptions visible.

You can inspect my product work, read about working with me in Winnipeg, or use the MVP scoping guide if the workflow needs a full application. When you are ready, send me the task you want to improve.