How to measure ROI on an AI project
Without a baseline before you start, AI ROI is guesswork. Here is the practical framework for measuring time savings, quality improvement, and total cost return on an AI implementation — before and after.
To measure ROI on an AI project, record the current state of the target task before the tool goes live — how long it takes, how often it happens, and what it costs in staff time. After a defined trial period, measure the same metrics again, subtract the total cost of the tool (including setup labour and ongoing review time, not just the licence), and apply the standard formula: ROI = (Net Benefit ÷ Total Cost) × 100. The reason most businesses cannot answer whether an AI project is working is not that the calculation is complicated — it is that no baseline was captured before the tool was turned on. Without a before, there is no measurable after.
Set a baseline before the tool goes live
Before deploying any AI tool, record three numbers for the specific task you are targeting: the average time to complete one instance, the number of times that task occurs per month across your team, and the fully-loaded hourly cost of the people doing it. Fully-loaded cost means wages plus employment overhead — typically 20 to 25 percent above the base wage rate for a Canadian employer. Multiply those numbers: (minutes per instance ÷ 60) × monthly instances × hourly rate. That is your monthly cost baseline. A task that takes 30 minutes and happens 80 times per month at a $30 fully-loaded rate costs $1,200 per month. A tool that cuts that task to 10 minutes and costs $60 per month recaptures $800 per month in labour capacity — an ROI that is easy to defend. A task that takes five minutes and happens 15 times per month at the same rate costs $37.50 per month. A $60 tool cannot pay for itself there regardless of how good it is. The baseline tells you whether the math can work before any dollar is committed.
Count the full cost, not just the licence
- Licensing or subscription: the monthly or annual fee for the tool itself.
- Setup and integration labour: the staff or owner time spent connecting the tool to existing systems, writing initial instructions, and configuring outputs. For most small business implementations, this runs 6 to 20 hours — at a real labour cost.
- Training and onboarding: the time staff spend learning to use the tool effectively, including the productivity dip in the first two to four weeks.
- Human review overhead: for any AI tool that produces customer-facing or compliance-sensitive output, a person reviews before it goes out. Count that review time as part of the total cost.
- Prompt and workflow maintenance: instructions need updating as the business changes. Budget at least one hour per month for an actively used tool.
Four metrics worth tracking
- Time per task instance: measure 20 consecutive instances before implementation and 20 after a one-month trial. Average each set. The difference, multiplied by monthly volume and hourly rate, is the time-savings value.
- Error or rework rate: the percentage of AI outputs that require correction before use. A tool that saves 20 minutes per task but produces output requiring 15 minutes of correction is saving far less than it appears.
- Customer or internal response time: for anything that affects turnaround — booking confirmations, quote follow-ups, inquiry replies — response time before and after is a measurable proxy for service quality and recaptured capacity.
- Revenue attribution: only where there is a direct, traceable line. A tool that converts more inquiry responses into confirmed bookings has a revenue value. A tool that drafts internal reports faster generally does not, at least not directly.
The ROI calculation
ROI = (Net Benefit ÷ Total Cost) × 100. Net Benefit is the value of recaptured labour time plus quality improvements, minus the human review and maintenance labour the tool requires. Total Cost is the licence plus setup plus training plus ongoing oversight. To illustrate with a hypothetical: a service business in Northwestern Ontario automates first-draft responses to a standard set of client inquiry emails. Before: each inquiry takes 20 minutes to respond to; the business receives 60 per month; fully-loaded labour cost is $30/hour. Monthly baseline: $600. After implementation, review and send takes 5 minutes per inquiry. Time recaptured: 15 minutes × 60 instances = 15 hours per month, valued at $450. Tool licence: $45/month. Setup labour: 8 hours × $30 = $240, one-time. Annual Net Benefit: ($450 − $45) × 12 = $4,860. Annual Total Cost: ($45 × 12) + $240 = $780. ROI at 12 months: ($4,860 ÷ $780) × 100 = 623%. This is an illustrative calculation only — your task volume and baseline determine whether a similar result is achievable.
When to expect results
Simple workflow automation — booking responses, document drafts, follow-up messages — can produce measurable time savings within 30 to 90 days if a baseline was recorded. CFIB's September 2025 survey of 1,683 Canadian small business owners found that generative AI saves an average of 1.08 hours per day per active user, and that digital tools overall return $1.60 for every dollar invested — with a 29 percent average productivity boost among adopters. Those are averages across all tool types and adoption maturity levels; individual results depend on which task was chosen and how well the implementation was designed. Revenue impacts — better conversion, reduced churn, new capacity that fills with paying work — typically take six months or more to show clearly. Evaluate a new AI tool on time and quality metrics at 30 days; reserve revenue-impact judgements for a six-month review.
The only AI ROI you can measure with confidence is ROI on a project where a baseline was recorded before the tool went live. ISED Canada's SME AI Adoption Blueprint identifies difficulty assessing returns as one of the primary reasons Canadian SMEs get stuck after an initial pilot — they deployed something, cannot quantify what changed, and cannot decide whether to expand or cut it. The fix is simple: a 20-minute task measurement exercise before the next implementation, documented in a spreadsheet. Nothing in this post constitutes financial or business advice specific to your organization.
Sources: CFIB Digital Transformation Report (September 2025) — generative AI saves an average 1.08 hours per day per active user; digital tools return $1.60 for every $1 invested; 29 percent average productivity boost; survey of 1,683 Canadian small business owners conducted April–June 2025: cfib-fcei.ca/en/research-economic-analysis/digital-transformation-how-small-businesses-in-canada-are-leveraging-ai-and-technology-for-growth-and-productivity | ISED Canada SME AI Adoption Blueprint — identifies 'struggle to define use cases, assess returns or move beyond pilots' as a key barrier for Canadian SMEs: ised-isde.canada.ca/site/ised/en/sme-ai-adoption-blueprint
Frequently asked
What if I already deployed the tool without recording a baseline?
You can reconstruct a rough baseline from timesheets, invoices, calendar records, or staff estimates — how long did this task take before the tool existed? The number will be softer than a directly measured baseline, but it is better than nothing. Use it as a conservative floor estimate. For the next implementation, record the baseline before the tool goes in.
The ROI calculation looks positive, but the tool feels slow to use. Should we keep it?
Run the time-per-task measurement again, carefully, for 20 recent instances. If the math still shows net time savings after accounting for review overhead, the sense of slowness is likely friction at a specific step rather than a net negative result. Identify the slow step — usually a handoff, a login, or a review process — and improve it before deciding to cut the tool.
How do we measure ROI when the main benefit is quality improvement, not time savings?
Track error rate, revision cycles, or complaint rate before and after. Estimate the labour cost of fixing a typical error and multiply by the reduction in error frequency. A tool that cuts your document revision rate from four rounds to two on 30 documents per month, where each extra revision takes 20 minutes at $35/hour, recovers (2 rounds × 20 minutes × 30 documents ÷ 60 × $35) = roughly $700 per month in quality-driven ROI — a real dollar value even without a time-savings component.
Get the weekly Signal
The AI, funding, government, and tech moves that matter for Northwestern Ontario — one email a week, source-linked, read by a human before it reaches you.
One email a week from Thunder Bay AI. Unsubscribe anytime.