The AI Scorecard – Mathew's Library

01 One job per AI use

On October 13, 2026, Wharton Human-AI Research and GBK Collective published their fourth annual study of AI at U.S. companies with at least 1,000 employees. Leaders who tracked at least one growth number said their AI returns were strongly positive 35% of the time. Leaders who measured only efficiency said it 13% of the time. The returns are self-reported, and a survey can't tell you which came first, so treat this as a cheap test, not a promise. Start by listing every place your team uses AI, and give each use one job.

List every AI use. Ask each person on the team where they use AI in a normal week. Write one line per use.

Give each use one job. Name it the way you would name a task: drafting proposals, answering support tickets, writing follow-up emails.

02 One growth number per job

Pick one growth number for each job, from the study's own list: net-new revenue, sales conversion, output volume and customer satisfaction. Be careful with output volume. More of something isn't growth unless it moves one of the other three. And pick a number the work can plausibly move: if AI drafts your blog posts, company revenue is too far away, but leads from those posts is close enough.

AI job Growth number

Drafting proposals Close rate

Answering support tickets Customer satisfaction

Writing follow-up emails Sales conversion

Drafting landing pages Conversion rate on those pages

Drafting blog posts Leads from those posts

03 Today's baseline

Write down today's baseline, before you scale anything. Without it, any improvement later is a story, not a result.

Pull the last eight weeks. From your own records: your CRM for close rate and sales conversion, your support tool for satisfaction scores, your analytics for page conversion and leads.

Write the number and the date in one place. One line per AI use, on this scorecard. A number with no date can't be compared later.

04 Where the freed hours go

This is the step people skip. One practitioner's rule, from a firm called Conversion System, says hours saved only converts to a return when the freed time is redirected to a named, revenue-connected activity. So write it down. If AI saves your team six hours a week on proposals, those hours go to follow-up calls within two days, and the growth number is close rate, not hours. For a content team: AI drafts the first version of landing pages, and the freed hours go to testing a second headline on every page.

Never add hours saved and revenue together. That's double counting. Keep hours saved on its own line as a signal that something is happening, and let the growth number say whether it mattered.

05 The 30- and 90-day checks

Check at thirty days and ninety days, and compare against the baseline. If you only have a few deals a month, the numbers will wobble, so compare before and after on the same kind of work, or keep one person on the old way for a while. Change one thing at a time: if you switch the tool, the offer and the pricing in the same month, no number will tell you which one worked. Copy one block per AI use.

AI scorecard: one line per AI use

AI use: [WHERE THE TEAM USES AI]
Its one job: [E.G. DRAFTING PROPOSALS]
Growth number: [NET-NEW REVENUE / SALES CONVERSION / CUSTOMER SATISFACTION / CLOSE RATE / ...]
Baseline (last 8 weeks): [VALUE] Written on: [DATE]
Freed hours go to: [NAMED ACTIVITY, E.G. FOLLOW-UP CALLS WITHIN TWO DAYS]
Hours saved per week (own line, not added to revenue): [HOURS]
Check at 30 days: [DATE] Value: [VALUE]
Check at 90 days: [DATE] Value: [VALUE]
What else changed in the same period: [NOTHING / LIST]

Sources

GBK Collective: Wharton-GBK AI study 2026, press release GBK Collective: Enterprise AI study 2026 Knowledge at Wharton: 2026 AI adoption report Conversion System: Hours saved is not ROI

Mathew's LibraryLibrary card

Free access

Keep reading free with your library card

One email opens this template and all 86 resources in Mathew's library. No account, no password.