Measuring AI ROI: a simple framework for small businesses

The three buckets that actually matter

For a small or mid-sized business, AI ROI comes from three places. Pick the one your project leans on hardest and measure that.

1. Hours saved (cost out). The most common, easiest to defend. Someone was spending X hours on a task; they now spend Y. Value = (X - Y) × loaded hourly cost × frequency.
2. Throughput up (revenue / capacity). Same team, more output - more proposals sent, more leads worked, more calls answered. Value = (incremental units) × (margin per unit).
3. Error / risk reduction. Fewer mistakes, fewer callbacks, fewer compliance misses. Harder to measure but often the biggest dollar amount when something's gone wrong recently.

Set the baseline before you start

This is the step almost everyone skips, and it's why their AI projects produce a feel-good story instead of a number. Before the build:

• Pick one metric that the project should move (e.g. "average proposal turnaround").
• Measure it for two weeks. Write down the number.
• Decide what counts as a win - e.g. "30% reduction sustained for 30 days."

The 4-line ROI math you can defend

Annual benefit  = hours saved/week × 50 weeks × $loaded_hourly
              ( + new revenue captured )
              ( + cost of errors avoided )
Annual cost     = software + build + ongoing oversight
Payback (mo.)   = build cost / (annual benefit / 12)
ROI (year 1)    = (annual benefit - annual cost) / annual cost

Use a loaded hourly cost (salary × ~1.3) rather than salary alone. Use 50 weeks, not 52, to account for vacation. Round down. Defensible numbers beat impressive numbers.

A worked example

A 12-person services firm builds an AI intake assistant. Baseline: each new client intake takes the owner 45 minutes; they do 8 per week.

• After launch: 12 minutes each. Savings = 33 min × 8 = 4.4 hr/week.
• Loaded hourly = $180. Annual benefit = 4.4 × 50 × $180 = $39,600.
• Build cost $8,000. Software $2,400/yr. Oversight $1,500/yr.
• Payback = 8,000 / (39,600 / 12) = 2.4 months.
• Year-1 ROI = (39,600 - 11,900) / 11,900 = 233%.

That's defensible to a partner or a banker. "AI is the future" is not.

What to measure 30, 60, 90 days in

• Day 30: Did people actually use it? (Usage rate among the intended team.) If under 60%, fix that before measuring ROI.
• Day 60: Did the baseline metric move? Compare to your pre-launch number, not to last quarter.
• Day 90: Did the savings stick, or did people drift back to the old way?

AI projects that never measured a baseline almost never get the second round of funding. Projects with a clear baseline and a 60-day check almost always do.