Most operators either don’t measure AI’s impact at all, or they track the wrong things — number of prompts run, tools subscribed to, features explored. None of that is ROI. ROI is what changed in the business: time recovered, errors eliminated, revenue influenced. Here’s how to measure it without a data team or a complicated setup.

Start With a Baseline

You cannot measure improvement without knowing where you started. Before you deploy an AI workflow, record the current state of the task it’s replacing or assisting with:

• Time: How long does this task take today, on average? (Be honest — include the time to context-switch, draft, and review.)
• Volume: How many times per week/month does this task happen?
• Quality indicator: Is there a quality signal you can track? Error rate, revision cycles, customer complaints, completion rate?
• Cost (if relevant): Is this task currently done by someone you pay? What’s the fully-loaded hourly cost?

Write this down before you start. A note, a cell in a spreadsheet, anything. You’ll need it in 30 days.

The Three Metrics That Actually Matter

• Time saved per occurrence: (Old time – New time) × frequency = hours recovered per month. If drafting a weekly report dropped from 90 minutes to 20 minutes, and it happens 4 times a month: 4.67 hours/month recovered. At $75/hour fully loaded, that’s $350/month in recovered capacity.
• Error or revision rate: Did AI-assisted outputs require more or fewer revision cycles than the old process? Track the number of edits required per output before it’s approved. Declining edits = quality improving.
• Revenue influenced (where applicable): If AI is helping with sales emails, proposals, or customer responses — are close rates, response rates, or retention numbers moving? This is harder to isolate but worth tracking if your use case is customer-facing.

Avoiding Vanity Metrics

Vanity metrics feel like progress but don’t tell you if AI is working:

• “We ran 500 prompts this month” — so what?
• “We use 7 AI tools” — irrelevant if none are producing measurable output
• “Everyone on the team has ChatGPT” — adoption isn’t impact

Replace these with: hours recovered, tasks completed per person per hour, cost per output, error rate. Numbers that connect to business operations.

A Lightweight Scorecard

Track this for each AI workflow you deploy. One row per workflow, reviewed monthly:

• Workflow name: e.g., “Weekly ops report draft”
• Baseline time: 90 min
• Current time: 22 min
• Frequency/month: 4×
• Hours saved/month: 4.5 hrs
• Tool cost/month: $20 (Claude Pro)
• Notes: Still requires one round of edits for tone

That’s it. Ten minutes a month to maintain. After 90 days you have enough data to make a real case for expanding AI use — or to identify workflows where the promise didn’t deliver and you should course-correct.

When ROI Is Harder to Quantify

Some AI value is real but hard to put a number on: faster decisions, better research quality, reduced cognitive load. Don’t ignore these — but don’t lead with them either. Anchor your case in the measurable numbers, and treat the intangibles as supporting evidence. The goal is to run the business better. The scorecard helps you know whether that’s actually happening. See Your First Daily AI Skill for the simplest way to start with a workflow you can measure from day one.