How to Measure the ROI of AI in Your Business

"Is it worth it?" is the right question to ask about any AI investment, and too few businesses can answer it. Many adopt tools on enthusiasm, see a vague sense of improvement, and six months later can't say whether the money was well spent. When budgets tighten, projects without clear numbers are the first to be cut, even the ones that were working.

Measuring the return on AI doesn't need a finance department or complicated models. It needs a baseline, a few honest measures and the discipline to check them. This guide shows you how.

Why AI ROI is tricky

A few things make AI harder to measure than, say, a new machine on a production line.

• The benefits are spread out. Ten minutes saved here, twenty there, across many people and tasks.
• Time saved doesn't automatically become money saved. If someone saves five hours a week and spends it on better work, that's real value, but it doesn't show up as a lower payroll.
• Quality effects are real but harder to count. Faster responses, fewer errors, more consistent proposals.
• Costs are easy to underestimate. Licences are visible; setup, training, maintenance and review time often aren't.

None of this makes measurement impossible. It just means you need to be deliberate about it.

Step 1: set a baseline

You can't measure improvement without knowing where you started. Before you introduce any AI tool or workflow, record how the process performs today.

For each process you plan to change, capture:

• Volume: how many times it happens per week or month
• Time: how long each instance takes, and total hours
• Cost: hours multiplied by the loaded hourly cost of the people doing it
• Turnaround: how long from request to completion
• Quality: error rates, rework, customer complaints
• Outcomes: conversion rates, win rates, response rates, where relevant

Rough numbers are fine. Ask people to track a typical week, or estimate from system data. What matters is that you measure the same way before and after.

Step 2: choose the right measures

Different projects create different kinds of value. Pick the two or three measures that matter most for each.

Time and capacity

The most common benefit. Measure hours saved per task and in total. Then decide what happens with that time: more volume handled with the same team, time redirected to higher-value work, avoided hiring, or reduced overtime.

Revenue and growth

For sales and marketing projects, look at leads generated, meeting rates, proposal turnaround, win rates, deal size and customer retention. Be careful about attribution: if several things changed at once, don't credit AI with all of it.

Quality and risk

Error rates, rework, compliance issues and customer complaints. A document intake workflow that cuts errors in half may save more than the time it frees up, once you count the cost of fixing mistakes.

Speed and customer experience

Response times, turnaround on quotes and support tickets, and customer satisfaction scores. Faster responses often win business that slower competitors lose.

Employee experience

Harder to put a dollar figure on, but important. Removing tedious work improves morale and retention, and replacing an employee can easily cost a significant share of their annual salary.

Step 3: count all the costs

Be honest about what AI really costs, including:

• Licences and subscriptions, including usage-based charges that grow with adoption
• Setup and integration, whether done internally or by a consultant
• Training time for staff
• Review time, because people still need to check AI output
• Maintenance, since workflows and agents need updating as your systems and processes change
• Management time spent on governance, policy and vendor management

Forgetting review and maintenance time is the most common way businesses overestimate their return.

Step 4: calculate the return

The basic calculation is simple:

ROI = (value created minus total cost) divided by total cost

Here's a worked example for a professional services firm that introduces an AI-assisted proposal process.

Baseline:

• 20 proposals per month
• 6 hours per proposal, so 120 hours per month
• Loaded cost of $70 per hour, so $8,400 per month
• Win rate of 25 percent

After three months:

• Drafting time drops to 2.5 hours per proposal, including review, so 50 hours per month
• That saves 70 hours per month, worth about $4,900
• Faster turnaround helps win rate rise to 28 percent. On an average project value of $15,000, that's roughly 0.6 extra wins per month, or about $9,000 of additional revenue. To be conservative, count only the margin on it, say 30 percent, which is about $2,700

Costs:

• Setup and training: $6,000 one-time
• Tools and maintenance: $400 per month

First-year result:

• Value: about $7,600 per month, or $91,200 per year
• Cost: $6,000 plus $4,800, so $10,800
• ROI: roughly 740 percent

The numbers in your business will be different, and you should be sceptical of examples that look too neat, including this one. The point is the method: baseline, value, full cost, and a conservative calculation.

Step 5: use conservative assumptions

AI business cases get into trouble when they assume the best case. To keep yours credible:

• Count only a portion of time saved as real value, for example half, unless you can show where the time went
• Include review time in the new process
• Credit revenue gains only at margin, and only the share plausibly caused by AI
• Include every cost, even small ones
• Revisit after three and six months with real data

A conservative business case that turns out better than expected builds trust. An optimistic one that falls short damages the next project.

Step 6: track over time

Measure at the start of a pilot, after one month, after three months and after six. Adoption usually rises as people learn, then levels off. Some benefits, such as better data and faster onboarding, grow over time. Some costs, such as usage-based charges, grow too.

Keep the tracking simple: a short monthly update on your chosen measures is enough. The goal is to know whether to expand, adjust or stop.

Signs a project isn't paying off

• Low usage. If people aren't using it, investigate why before paying for more.
• Time saved is eaten by review. The output may need too much correction, which usually points to poor instructions or poor information.
• Errors are rising. Tighten review and fix the underlying cause.
• Nobody owns it. Unowned workflows decay quickly.

Stopping a project that isn't working is a good decision, not a failure. It frees budget for the ones that are.

Building the business case before you start

If you need to justify an AI investment to a partner, board or bank, a one-page business case is usually enough:

1. The problem: what process, how much time, what cost, what frustration
2. The proposal: what you'll change and which tools you'll use
3. The expected value: time, revenue, quality and speed, with conservative assumptions
4. The costs: one-time and ongoing
5. The measures: how you'll know it worked
6. The pilot: a small, time-limited test before committing fully

Where the biggest returns usually are

Across the businesses we work with, the highest returns tend to come from high-volume processes with messy inputs: document intake, email triage, proposals and quotes, reporting, knowledge lookup and sales preparation. We explain why in AI and business processes. The lowest returns tend to come from buying licences for everyone without training or a clear purpose.

How Line49 helps

Every Line49 engagement starts with a baseline and ends with measured results. In our Foundation package we map your processes, estimate the time and cost involved, and prioritize projects by expected return. When we build workflows and agents, we measure them against the baseline so you know exactly what you got for your investment.

If you'd like help building a business case, or checking whether your current AI tools are paying off, book a consultation with our Ottawa team. New to all this? Start with how to use AI in your small business.