Skip to main content
METHODOLOGY

How I measure ROI on AI work, including the intangible kind

Three numbers tracked monthly: time saved per workflow, money recovered from hidden leaks, decision speed. For intangible services like therapy or coaching, the value gets tied to a behavior the client can count.

I measure ROI three ways. Hard time saved per workflow, measured at every run. Money recovered from leaks the team did not see before. Decision speed, measured by how fast the team can answer a question that used to take hours. For intangible services, value ties to a behavior the client can count.

The two questions buyers ask, in two different vocabularies

A solopreneur psychologist asked me on a discovery call: "Como lo que yo vendo es tan intangible, te vendo claridad, te vendo paz, como hago yo para medirlo? Me ha costado mucho, vender lo intangible." Translation: what I sell is so intangible, I sell clarity, I sell peace, how do I measure that? It's been hard for me to sell the intangible.

A construction client at Concreto told me, in a different vocabulary: their projects lose between 10 and 15 percent of margin to inefficiencies they only see when the job is finished. Same root problem, different language. The buyer cannot price what they cannot count.

Most AI consultants dodge this. They sell the tool, not the outcome. They send invoices that read "AI strategy" and the client questions every renewal because nothing visible changed. I do not work that way.

THE THREE NUMBERS

The three numbers I track on every engagement

One. Time saved per workflow.

Before we automate a report or a quote or a proposal, I ask the person doing it today how long it takes. We write it down. After the change, we measure the new time at every run.

A wealth advisory firm I work with assembles quarterly investor materials. Before, the cycle was 30 to 45 days from books closed to investor delivery. After, the first draft is ready in minutes, with a partner reviewing for tone before it ships. That is the number I report monthly. Not "the AI is faster," but "5 minutes vs 35 days, every quarter." This is the pattern across wealth-firm quarterly reporting cycles where the diff between calendar weeks and minutes is the entire argument.

A construction firm runs a daily document search across job-site contracts. Before Capataz, finding a clause took 4 days because someone had to dig through 70-page subcontracts. After, it takes 30 seconds. Same number, different vocabulary.

Two. Money recovered from leaks the team could not see before.

In construction, this is the construction margin-leak case Concreto's team confirmed: 10 to 15 percent per project. The leak shows up after a job closes, when the post-mortem reveals the rebar order was 8 percent over what was needed, the foreman approved an unscheduled overtime block, the subcontractor billed an extra unit nobody caught. Eight revenue leaks in the first quarter we tracked. Each one a real dollar amount that used to disappear.

In packaging, a senior commercial operator at a packaging client put it: errors in a quote multiply across every unit shipped. We caught three quote-spec errors in the first month of running their workflow through Claude. The recovery is not theoretical. It is the difference between the wrong quote and the right one, multiplied by units shipped.

Three. Decision speed.

How fast can a foreman, an analyst, or a partner answer a question that used to require opening five files and calling someone? I time this before and after, with the actual user.

For one wealth firm partner, the question was "what is our exposure to a specific holding across all client portfolios." Before: 4 to 6 hours of an analyst pulling spreadsheets. After: under 60 seconds. The partner doesn't have to wait for the answer to make the next decision. That changes how the firm operates, not just how fast it operates.

A senior partner at a $14B wealth advisory firm put the goal in operator language: "make decisions and have confidence." Hours saved, money recovered, and decision speed all serve that single outcome. The partner can act on the information in front of them without waiting for someone to assemble it. That confidence is the actual product. The numbers are how I prove it.

INTANGIBLE SERVICES

How I measure intangible services

The trick is to tie value to a counted behavior.

A solopreneur psychologist sells "claridad." I asked her: when a client gets clarity, what do they DO differently? She said: they finish the week's most important task instead of avoiding it. So we count weekly anchor tasks completed. We count minutes of focused work logged in her Power Focus app. The intangible becomes a number she can put on a sales page.

For a coaching client, the metric was "decisions made within 48 hours of receiving information." For a therapy client, it was "weeks without a relapse-into-old-pattern episode." Different domains, same architecture: the value follows from a behavior the client either does or doesn't do, and you count the behavior.

This is also why I refuse to invoice for workflows that don't produce a measurable number by month two. If we cannot count it, the engagement isn't working. I tell you to stop paying me for it.

THE MONTHLY REPORT

What goes in the monthly report

A one-page summary, every month, sent the same week the bookkeeping closes:

  • Hours saved this month per workflow
  • Money recovered from leaks identified
  • Decisions accelerated, with examples
  • What shipped (Skills, CLAUDE.md updates, automations)
  • What's next month

The report is generated from the Skills and CLAUDE.md that produce the monthly numbers. The retainer is justified by visible output, not by the calendar. If month three's report has the same numbers as month two with no new shipped work, the retainer drops. If the workflows compound and new ones come online, the retainer holds.

WHY THIS WORKS

Why this works for the Fractional Head of AI specifically

Most AI consultants get fired in month four because the visible value disappears after the initial setup. The architecture above prevents that. Every retainer month produces a Skill, an automation, or a CLAUDE.md improvement that ships, gets measured, and reports back as a number on the next month's one-pager. This is the Fractional Head of AI retainer architecture, end to end.

See how the retainer is structured around measured output and what you own at the end of every engagement for the contract-level details.

If you want to see what one of those monthly reports looks like, book a 30-minute call. I'll show you an anonymized one from a current client. You'll see the format, the numbers, and the language.

ROI Questions

Frequently Asked Questions

Three numbers, tracked monthly. Hours saved per workflow, measured at every run. Money recovered from leaks the team didn't see before. Decision speed, timed before and after with the actual user. The retainer is justified by visible output, not by the calendar.

Tie the value to a counted behavior. If a client gets clarity, what do they do differently? Finish a weekly anchor task. Make a decision within 48 hours. Spend more focused minutes on real work. Count the behavior. The intangible becomes a number on the sales page.

Depends on the workflow. For a wealth firm, 35-day quarterly reporting cycles dropped to minutes-plus-review. For construction, 4-day document searches dropped to 30 seconds, and 8 margin leaks worth 10 to 15 percent per project surfaced in the first quarter. The retainer is small relative to either number.

Before automation, I ask the person who does the work today how long it takes. We write it down. After the change, we measure the new time at every run. The diff is the number I report monthly. Not 'faster,' but '5 minutes vs 35 days.'

A margin leak is money lost to inefficiencies that only show up when a job closes. Wrong product spec quoted, unscheduled overtime, supplier overbilling. AI surfaces these by reading every project document and flagging anomalies the team is too close to see. Concreto found 8 in one quarter.

A wealth firm's quarterly reporting cycle. Before: 30 to 45 days from books closed to investor delivery, with three analysts assembling. After: first draft in minutes, partner reviews for tone, ships same week. Same investor-experience quality, ten weeks of analyst time freed per quarter.

I tell you to stop paying me for that workflow. If we can't count it by month two, the engagement isn't working on that workflow. Drop it, focus the retainer on the workflows that do produce numbers. The retainer should never coast on assumptions.

One page, sent the same week bookkeeping closes. Hours saved per workflow this month, money recovered from leaks identified, decisions accelerated with examples, what shipped (Skills, CLAUDE.md updates, automations), what's next month. The retainer is justified by visible output, not the calendar.

Tie value to a counted behavior. For coaching: decisions made within 48 hours. For therapy: weeks without an old pattern repeating. For psychology: weekly anchor tasks completed plus focused minutes logged. Different domains, same architecture: count the behavior the client does or doesn't do.

First measurable workflow ships in week 4 of a typical engagement. The number that matters lands by month two: hours saved or money recovered, with the diff documented. By month three, the monthly report has compounding numbers across multiple workflows.

If you want to see what one of those monthly reports looks like, book a 30-minute call. I will show you an anonymized one from a current client.

You will see the format, the numbers, and the language.