Technology & AI

Should You Hire Someone or Automate the Work? A Canadian Operator’s Decision Framework

A Canadian employer pays up to CAD 6,218.75 in statutory contributions in 2026 before salary. The four tests that decide between a hire and an automation, and the work that should never be automated.

Cody WiseCody WiseFounder, Wise Media13 min read
Wise Media article card reading Hire Someone or Automate the Work, with the line statutory payroll adds 6,218 dollars before the first hour of work

Quick Answer

Automate the work first, then hire against what is left. A Canadian employer pays up to CAD $6,218.75 in statutory CPP, CPP2 and EI contributions in 2026 before salary, vacation or stat holiday pay, and that cost recurs whether the role is busy or idle. Automation carries the opposite profile: high setup, low marginal cost, zero judgment. The decision is not about money. It is about whether the task needs a decision.

Written by Cody Wise, founder of Wise Media, Calgary. Payroll figures are taken from the Canada Revenue Agency’s 2026 payroll deductions tables and Alberta’s published employment standards. They are general information for planning, not tax, payroll or employment advice. Confirm your own numbers with an accountant and your provincial employment standards office before you hire.

Summary

  • The decision rule in one line. If the task has a defined trigger, a defined output, and no judgment call in between, automate it. If any of those three is missing, a person does it.
  • The statutory floor is CAD $6,218.75 per employee in 2026. That is maximum CPP of $4,230.45, CPP2 of $416.00 and EI of $1,572.30 on the employer side, per the CRA’s 2026 tables.
  • Vacation and stat holidays are the second layer. In Alberta that is 4% of wages rising to 6% after five years, plus nine general holidays.
  • Automation fails on exceptions, not on volume. Count how often the task goes off script. Above roughly one in five, you are building a system that needs a human babysitter anyway.
  • The order matters. Document, delete, simplify, automate, then hire. Hiring into an undocumented process buys you a second person doing it inconsistently.
  • Hire for judgment, relationships and accountability. Automate for retrieval, routing, formatting, reminding and recording.
Founder working at a multi monitor home office workstation with dashboards and a laptop open
Most founders reach for a hire when what they actually have is an undocumented process running on one person’s memory.

Table of Contents

  1. Should you hire someone or automate the work?
  2. What a Canadian employee actually costs in 2026
  3. The four tests that decide
  4. What should never be automated
  5. Where automation actually pays, function by function
  6. The sequence that works: document, delete, automate, hire
  7. Common mistakes
  8. Frequently asked questions

Should you hire someone or automate the work?

Automate any task with a defined trigger, a defined output, and no judgment call in between. Hire when the work requires judgment under uncertainty, carries a relationship, or needs someone accountable when it goes wrong. Most founders get this backwards because they experience the pain as volume, and volume is what makes a hire feel obvious.

Volume is the wrong signal. Volume tells you the task is frequent, which is an argument for automating it. The signal that justifies a hire is variance: how often the task requires a different answer than the last time. A high volume, low variance task is the single best automation candidate you own. A low volume, high variance task is why you hire a senior person rather than a junior one.

This is the framework we run with clients inside our AI automation service, and the sequencing matters more than the tooling. Nearly every operator who tells us they need a hire is actually describing three separate tasks, two of which are mechanical and one of which is genuinely hard.

The cost profiles are shaped differently

HireAutomate
Setup costRecruiting time, onboarding, rampProcess mapping, build, testing
Marginal cost per unitFlat, the salary does not scale downNear zero, or metered per run
Cost when idleFullNear zero
Handles exceptionsYes, that is the pointNo, it escalates or breaks
FailsSlowly and visiblyInstantly and silently
Exit costNotice, severance, relationshipCancel the subscription
CompoundsThrough judgment and relationshipsThrough reuse across clients

The row that decides most real cases is “fails”. An automation that breaks silently at 2am and stops sending client reports is worse than no automation, because nobody notices for a week. That is an argument for monitoring, not against automating, but it is a real cost that spreadsheets leave out.

What a Canadian employee actually costs in 2026

Salary is the number people compare against a software subscription. It is the wrong number. The employer side of statutory payroll adds up to CAD $6,218.75 in 2026 before you account for vacation, general holidays, workers compensation coverage, equipment, software seats or management time.

The statutory floor, from the CRA’s 2026 tables

These figures are published by the Canada Revenue Agency in the T4032 payroll deductions tables effective 1 January 2026. All amounts are Canadian dollars. EI figures are the Canada except Quebec rates; Quebec operates QPP and QPIP separately and the figures differ.

ContributionCeilingEmployer rateEmployer maximum
CPP (base plus first additional)$74,600 YMPE, less $3,500 basic exemption5.95%$4,230.45
CPP2 (second additional)$74,600 to $85,000, a $10,400 band4.00%$416.00
EI, Canada except Quebec$68,900 maximum insurable earnings2.282%$1,572.30
Statutory employer total$6,218.75

Two details worth noting. The employer EI rate is 1.4 times the employee rate of 1.63%, which is why the employer pays more into EI than the employee does. And the employer remits both sides of CPP, so the cash leaving your account each cycle is roughly double the employer line above.

Vacation and general holidays, using Alberta as the worked example

Employment standards are provincial, so the second layer depends on where your employee works. Alberta’s published rules, from the provincial government’s vacation pay and general holidays pages, are representative of the structure if not the exact figures elsewhere.

  • Vacation pay. Two weeks and 4% of yearly wages for employees with one to four years of service, rising to three weeks and 6% after five consecutive years. Vacation has to be taken within the twelve months after it is earned.
  • General holidays. Alberta has nine, with employers free to recognise more. Easter Monday, Heritage Day, the National Day for Truth and Reconciliation and Boxing Day are optional additions many employers grant.
  • Eligibility. An employee must have worked for the same employer for at least 30 workdays in the twelve months before the holiday.
  • Working the holiday. The employer chooses between 1.5 times the normal wage for hours worked plus the average daily wage, or a future paid day off at average daily wage.

Workers compensation coverage is a third layer, priced by industry classification and payroll. We could not retrieve the current Alberta rate schedule while writing this, so verify your own classification rate directly with your provincial board rather than budgeting from a number you read anywhere, including here.

The costs nobody budgets

Add these before comparing a role to a tool. They are the reason a first hire usually underperforms the spreadsheet.

  • Ramp. Most operational roles take six to twelve weeks before output is net positive. Until then the hire consumes the founder’s time rather than freeing it.
  • Management overhead. One direct report is typically two to four hours a week of your time, permanently. That time comes out of the highest value work you do.
  • Tooling and seats. Every per seat subscription you run gains a seat.
  • The documentation you still have to write. You cannot delegate a process you have never written down, so the SOP cost lands either way. Automating at least means the SOP becomes the system.
  • Key person risk transfer. A hire moves single point of failure from you to them. That is progress, not elimination.

The four tests that decide

Run every candidate task through these four in order. A task has to pass all four to be worth automating. Failing any one of them is an argument for a person, at least for now.

1. The trigger test

Can you name the exact event that starts this task, in a sentence, without using the word “when someone notices”? A form submission, a payment, a calendar event, a status change, a file landing in a folder, a date passing. If the trigger is a human remembering, you do not have a process, you have a habit. Automate the trigger before anything else, even if a person still does the work.

2. The output test

Can you describe what “done” looks like precisely enough that two different people would produce the same artifact? If the answer is a document, a record, a message or a status, you are in automation territory. If the answer is “it depends what they say”, you are not.

3. The judgment test

Between trigger and output, is there a point where someone has to weigh competing considerations and make a call that carries consequences? Choosing which of three overdue clients to chase first is judgment. Sending a reminder on day 31 is not. Note that a judgment step does not kill the automation. It splits it. Automate up to the decision, surface the decision with everything needed to make it, automate everything after.

4. The exception test

Count the last twenty times this task ran. How many went off script? Below roughly one in ten, automate confidently. One in five, automate with a clear escalation path. Above one in three, the process is not stable enough to encode and you are better off running it manually until it settles, or hiring someone whose job is to make it settle.

TaskTriggerOutputJudgmentExceptionsVerdict
Sending the onboarding pack after a signed contractClearClearNoneRareAutomate
Chasing an overdue invoice at day 31ClearClearNone at first touchRareAutomate, escalate at touch three
Writing a proposal for a new prospectClearVariableHeavyConstantHire, template the mechanical parts
Qualifying an inbound leadClearClearLightModerateAutomate the scoring, human owns the call
Monthly client reportingClearClearIn the commentary onlyRareAutomate the data, human writes the read
Handling an unhappy clientUnclearVariableHeavyConstantHire, and make it senior

Notice how often the verdict is a split rather than a choice. That is the normal answer. The reason “hire or automate” feels hard is that it is almost never a whole role versus a whole system.

What should never be automated

Some work is cheap to automate and expensive to have automated. These are the categories where the saving is real and the damage is larger.

Operator taking a client call in a glass coworking phone booth, notebook open on the ledge
The save conversation, the price increase and the bad news call are the three moments a client remembers. None of them should arrive as a workflow.
  • Bad news. A missed deadline, a cost overrun, a mistake. Automated apologies read as contempt. Pick up the phone.
  • Pricing conversations. Rate increases and scope negotiations land on the relationship, not the inbox.
  • Retention. The signals that a client is about to leave are behavioural and early, and a person has to notice them. We broke down what those signals look like in why clients cancel retainers.
  • Anything a regulator or a court might read. Legal, tax, safety and compliance claims need a named human who checked them.
  • Hiring and performance conversations. Screening can be assisted. The decision cannot be delegated to a score.
  • First contact with a high value prospect. A templated first touch on a six figure opportunity is a false economy.

There is a fifth category worth naming: anything you do not understand yet. Automating a process you have never run manually encodes your assumptions rather than reality, and you will not find out until a client does.

Where automation actually pays, function by function

Pricing models matter more than list prices here, because the model determines whether your costs grow with headcount, with clients or with usage. Per seat pricing punishes hiring. Per run pricing punishes volume. Per conversation pricing punishes success. Choose the model that is misaligned with the thing you expect to grow fastest.

FunctionAutomateKeep humanTypical pricing model
Lead intakeCapture, routing, enrichment, scoring, instant acknowledgementThe qualifying conversationPer seat or per contact
Phone coverageAfter hours answering, triage, bookingComplex or upset callersPer minute or per conversation
OnboardingContract, invoice, access requests, kickoff scheduling, asset collectionThe kickoff call itselfPer run or flat platform fee
ReportingData pull, formatting, delivery, schedulingThe interpretation and the recommendationPer seat
Invoicing and collectionsIssue, remind, reconcile, flagAnything past the second reminderPercentage of volume or flat
Internal knowledgeSearch, retrieval, drafting from your own documentsDeciding what is truePer seat or per token

Published vendor pricing changes constantly and varies by region and contract, so price your own shortlist directly rather than planning against any figure quoted in an article. What does not change is the shape: setup is front loaded, marginal cost is low, and the real recurring cost is maintenance, not licence fees.

Two of these deserve their own read. Phone coverage is the most commonly mis specified, and we compared the options in AI receptionist versus answering service. Onboarding is the highest return single automation most service businesses can build, and the seven steps worth systemising first are in automating client onboarding.

The sequence that works: document, delete, automate, hire

Order of operations is where this decision is usually lost. Hiring into chaos produces two people in chaos. Automating chaos produces chaos at scale, faster, with no one watching.

  1. Document. Write the process as it actually runs, not as you wish it ran. Record yourself doing it once. The gap between the two versions is where the real problem lives.
  2. Delete. Kill the steps that exist because of a client who left in 2023. Most processes carry 20% dead weight that nobody has had a reason to question.
  3. Simplify. Reduce the number of tools the process touches. Every handoff between systems is a future failure point. Consolidating into one record of truth, usually the CRM, removes more work than any automation will. Our guide to choosing a CRM for a Canadian service business covers that choice.
  4. Automate. Now encode it. Start with the trigger and the record keeping, which are the least risky and the most reliably annoying parts.
  5. Monitor. Build the alert before you need it. An automation with no failure notification is a liability pretending to be an asset.
  6. Hire. Hire against what is left, which is now a smaller, clearer, more senior role than the one you were about to post. People are more willing to join a business that has its operations documented, and they ramp faster when they do.

Running the sequence usually changes the role. A founder who was about to hire a coordinator ends up hiring an account manager, because the coordination is now a workflow and what remains is the client relationship. That is a better role, a better hire and a better business.

Common mistakes

  • Comparing salary to subscription. The honest comparison is loaded employment cost against setup plus licence plus maintenance plus the cost of silent failure.
  • Automating the task instead of removing it. Ask whether the output is still read by anyone before you build the thing that produces it faster.
  • Treating a contractor as the cheap version of an employee. Worker classification is decided by the substance of the relationship, not the label on the invoice. Get it wrong and the reassessment lands on you, not on them. Confirm the position with an accountant before structuring a role that way.
  • Building the automation yourself at founder rates. Your time is the most expensive line in the business and it does not appear on any invoice.
  • No owner. Every automation needs a named human responsible for it. Unowned systems rot quietly.
  • Scoping the build like a project instead of a system. Automations change as the business changes. The same discipline that keeps a client project from drifting applies here, and we set it out in writing a proposal that does not invite scope creep.
  • Hiring to avoid learning the process. If you hire because you do not want to understand the work, you will not be able to tell whether it is being done well.

Frequently asked questions

At what point do automation tools beat hiring another person?

When the work is high frequency and low variance. Frequency makes the build worth amortising and low variance means the system will not need a human standing behind it. If the task goes off script more than about one time in five, a person is still the right answer.

How much does an employee really cost on top of salary in Canada?

Statutory employer contributions alone reach CAD $6,218.75 in 2026 at the maximums: $4,230.45 CPP, $416.00 CPP2 and $1,572.30 EI, per the CRA’s 2026 tables for Canada outside Quebec. On top of that sit provincial vacation pay, general holidays, workers compensation premiums, equipment, software seats and your own management time.

Should a solo founder automate before the first hire?

Yes, with one condition: document the process first. Automation locks in whatever process you give it. A solo founder who automates intake, scheduling, invoicing and reporting typically delays the first operational hire by months and makes it a better role when it comes.

What if the automation breaks while I am asleep?

Assume it will. Build a failure notification that reaches a human, a fallback path that degrades to manual rather than silently dropping work, and a weekly check that something actually ran. Treat an unmonitored automation as unbuilt.

Is a contractor the middle ground between hiring and automating?

Sometimes, for variable volume and specialist skills. It is not a way to avoid employment costs. Classification depends on the real nature of the relationship, including control, tools, financial risk and integration into your business. Structure it with an accountant rather than assuming a contract settles it.

How long should an automation take to pay back?

Use a 90 day rule for operational workflows. Estimate hours saved per month, multiply by a realistic internal rate, and compare against build plus three months of running cost. If it does not clear inside a quarter, the process is probably not stable enough yet.

The short version

Hiring buys judgment, relationships and accountability, and it costs more than the salary by a margin most founders never calculate. Automation buys consistency and throughput, and it costs attention rather than payroll. The task tells you which one it needs: trigger, output, no judgment, few exceptions means build it. Anything else means hire, and hire for the part a system cannot do.

Run the sequence in order and the question usually answers itself. By the time you have documented, deleted and automated, the role you were going to post has become a smaller and more senior one, and you can afford to pay properly for it.

Build the system before you build the team

Wise Media maps the processes a service business actually runs, removes the steps that should not exist, and builds the automations that make the next hire a smaller problem. If you are weighing a role against a system and want an outside read on which one you need, tell us what is eating the week and we will scope it.

Cody Wise, founder of Wise Media

Written By

Cody Wise

Founder of Wise Media, the Calgary software development and product design studio. Cody builds brands, websites, apps and AI systems for founders across Canada and the United States, and writes here about what works in live client builds.

Reading About It
Or Getting It Built?

Thirty minutes with Cody. The bottleneck named, what we would build first, and what it costs.