AI OPERATING LEVERAGE AUDIT · FOR OWNER-LED COMPANIES

AI is easy to buy.Few know which five workflows to hand it.

In five business days I estimate and rank five of your finance and operations workflows by hours and dollars, then hand you two workflow designs and the arithmetic behind them. $7,500, fixed. You decide what gets built.

FIXED-SCOPE AUDIT
$7,500
DELIVERY WINDOW¹
5 business days
WORKING OUTPUTS
5 deliverables
¹ FROM KICKOFF, SCHEDULED ONCE THE DEPOSIT HAS CLEARED AND THE AGREED INPUTS ARE READY · TWO SCHEDULED CALL HOURS, PLUS INPUT PREPARATION · IMPLEMENTATION SCOPED SEPARATELY
THE INSTRUMENT

Five deliverables. Five days. $7,500, fixed.

One business unit, five priority workflows. A 60-minute kickoff, three days of analysis, a 60-minute readout. Your team prepares the agreed inputs; I do the rest.

01
AI Tool Stack Audit
Every tool your five workflows touch, with a verdict: keep, augment, retire or replace. Savings appear only where a contract supports them.
02
Process Leverage Map
Your five workflows ranked by hours, effort and modeled value. The rationale is written down and every override is logged.
03
90-Day Implementation Roadmap
What to build first, who owns it, what it depends on. 30/60/90 milestones and a build-or-buy call for each initiative.
04
Two Agentic Workflow Designs
Trigger, tools, review points, exception owner and manual fallback for two workflows. The build is scoped separately.
05
Working ROI Model
An Excel model with capacity and cash on separate lines and every cost included. Change an input and the case recomputes.
AFTER
Three good exits
Build it yourselves from the designs. Have me build it: fixed-scope implementation is typically $30,000–$45,000; retainers start at $5,500/month. Or wait, with the model showing what waiting costs. Sign a follow-on SOW within 30 days of the readout and half the audit fee paid is credited once toward it.

Unwrapped inventories, maps, validates, reconciles and orchestrates. It does not move money, post to the general ledger, contact your customers, build chatbots or act as an ERP implementer. Each design names its review points and a checker who did not write the prompt.

FEE: 50% ON SIGNATURE, DUE ON RECEIPT; 50% ON DELIVERY OF THE PACKAGE, DUE NET 15 · MUTUAL NDA BEFORE KICKOFF, INCLUDING AI-PROCESSING RESTRICTIONS ON YOUR DATA · THE 5-DAY CLOCK STARTS AT KICKOFF, SCHEDULED ONCE THE DEPOSIT HAS CLEARED AND THE AGREED INPUTS ARE READY
THE LAY OF THE LAND

The capability is real.
The evidence points to the workflow, not the purchase.

Three findings from the research, each with its population, its date and its source.

CAPABILITY

The capability is real, and uneven.

In a preregistered experiment with 758 BCG consultants, those given GPT-4 completed 12.2% more tasks, and finished 25.1% faster, on 18 tasks inside the AI’s frontier.1 On one task deliberately placed outside the frontier, consultants using GPT-4 were 19 percentage points less likely to reach a correct answer.1

On the right task the gain is real. Which task the tool gets matters as much as which tool you buy.

ORGANIZATION SCIENCE · 2026
USE

Use is real, and shallow.

In the Census Bureau’s Business Trends and Outlook Survey for late August to early September 2026, 34.1% of US businesses with 100 to 249 employees had used AI in any business function in the previous two weeks. That counts any use, not depth.2 In its AI supplement, collected November 2025 to February 2026, 53.8% of AI-using US firms with 100 to 249 employees made no changes to use AI in the previous six months.3 Among finance teams, in a June 2025 survey of 183 CFOs, 84% of finance organizations had implemented or planned to implement AI; only 7% reported a high or very high impact. That figure is Directional.4

The tool arrived, and most of the firms using it made no changes to use it.

US CENSUS BUREAU · GARTNER · 2025–2026
TIME

Undecided time is absorbed.

In the March 2026 version of a Danish study, 85% of chatbot users reported putting time saved into other job tasks, and fewer than 10% into breaks or leisure. That is self-reported, not proof of waste.5 Using Danish administrative registers, the same researchers rule out effects larger than 2% on workers’ earnings or recorded hours, in pooled estimates from the September 2025 version.6

The time doesn’t show up in pay or recorded hours, and only a decision changes that.

HUMLUM AND VESTERGAARD · 2025–2026

Unwrapped believes in AI. The evidence says where its result is decided, and that is where Unwrapped works.

  1. 1Dell’Acqua et al., “Navigating the Jagged Technological Frontier”, Organization Science 37(2), 2026 (version of record; working paper 2023). Preregistered experiment, 758 BCG consultants. The gains hold inside the frontier only.
  2. 2US Census Bureau, Business Trends and Outlook Survey, cycle 202619 (reference period 24 August to 6 September 2026), published 24 September 2026. Share of businesses, not of workers; any use in the two weeks, not depth.
  3. 3US Census Bureau, Business Trends and Outlook Survey, AI supplement, collected November 2025 to February 2026. The item counts changes made to use AI.
  4. 4Gartner press release, “Gartner Says CFOs Need Structured Finance AI Roadmaps”, 8 June 2026. Survey of 183 CFOs, June 2025. Directional.
  5. 5Humlum and Vestergaard, NBER working paper w33777, revised March 2026, p. 14. Self-reported; respondents could choose more than one answer.
  6. 6Humlum and Vestergaard, BFI working paper 2025-56, September 2025 version (the May 2025 version gives 1%). Danish administrative earnings and hours records; difference-in-differences; pooled estimates.
WHY UNWRAPPED

AI is sold wrapped: in a demo, an adjective, or someone else’s statistic.
Unwrapped opens it on your own work.

Unwrapped answers one question for owner-led companies, on their own records: where does AI change the work, and what is that change worth?

An AI claim is real for a company only once it lands in all four places.

01

The workflow

Does the task sit inside what the tool does well, on your records?

It is tested with known answers on your own work, and re-tested when the tool changes.

02

The system

Can the process be trusted, and can it stop?

Sources are dependable and checks are built in. Exceptions route to an owner. Nothing writes to the ledger. A manual fallback exists.

03

The people

Is use sanctioned, checked and safe to admit, and does someone stop the old task?

A named checker reviews the output and did not write the prompt. Role changes are clarified in writing. The stop is recorded.

04

The business

What is the released time for, and which ledger shows it?

Every block of hours gets a destination, an owner and a decision date, or is recorded as absorbed. Only invoiced or contracted value is called cash.

Not for a leader looking for a headcount number, in either direction, or for anyone who wants a return sized from someone else’s productivity statistic.

THE PRACTICE · FOUR CAPABILITY LANES

A reference library. Applied to your operation.

Four lanes of workflow designs. A starting point, not a catalogue: the audit decides which ones your operation actually needs.

THE PRACTICE PAGES →Data & AnalyticsStrategic ForecastingMergers & AcquisitionsChange Management

AUTOMATE STARTS WITH THE AUDIT · THE OTHER THREE LANES HAVE THEIR OWN PAGES · CHANGE MANAGEMENT RUNS UNDER ALL FOUR

DESIGN NAMES ARE PUBLIC; THE METHODS ARE NOT. SCOPED ESTIMATES AND REVIEW POINTS ARRIVE WITH THE AUDIT — OR IN THE DILIGENCE READ, IF YOU'RE MID-DEAL.
THE INSTRUMENT PANEL
SAMPLE MODEL — YOUR NUMBERS COMPUTE IN THE AUDIT

The analytics arrive looking like this.

Numbered exhibits, each answering one question a CFO actually asks and each tracing to a model cell. Below: the sample company’s month, decomposed.

FIG. 1Where the month goes
Hours per month across the top five workflows — before automation.
Reporting pack22h
Reconciliations15h
Budget-vs-actual packs10h
Board & lender packet8h
Contract & subscription renewals6h
FIG. 2The same five, ranked by dollars
Capacity at stake a year, per workflow: hours × 65% × $75 × 12.
Reporting pack$12,870/yr
Reconciliations$8,775/yr
Budget-vs-actual packs$5,850/yr
Board & lender packet$4,680/yr
Contract & subscription renewals$3,510/yr
FIG. 3The arithmetic, assembled
Each workflow stacks into the year. The total is capacity at stake at the loaded rate, before implementation costs.
+12.9K
+8.8K
+5.9K
+4.7K
+3.5K
$35,685
FIG. 4At stake, in hours and in dollars
The sample month as one pair: hours at stake at a 65% blended capture assumption, and what they are worth at $75 an hour.
MANUAL WORK
61h
A MONTH
AT STAKE
39.7h
A MONTH, AT 65%
CAPACITY AT STAKE
$35,685
A YEAR

None of this is cash. Cash is what an invoice or a contract shows, and it sits on its own line.

The sample audit shows which builds come first →
ILLUSTRATIVE COMPOSITE — NOT A CLIENT ENGAGEMENT · 61 MONTHLY HOURS × 65% BLENDED CAPTURE ASSUMPTION = 39.65 HOURS A MONTH AT STAKE, SHOWN AS 39.7 · × $75/HOUR × 12 = $35,685 A YEAR OF CAPACITY AT STAKE · CAPACITY AT THE LOADED RATE EXCLUDES IMPLEMENTATION COSTS AND IS NOT CASH
YOUR MODEL · ASSUMPTIONS IN VIEW

Your hours. Your arithmetic.

The defaults are the sample company’s five workflows. Replace them with a workflow you know, then add documented cost changes to build a separate cash case.

Capacity assumptions

Starting values: 61 hours and $75 an hour from the sample company (an illustrative composite, not a client engagement), and 65% as a capture assumption, not a measured rate. Include review, exceptions and maintenance in your estimate.

Add a cash scenario

Use cost reductions you can act on, such as a cancelled subscription. Do not enter the capacity value above as cash savings.

Zero is a starting input, not a cost estimate.

Capacity at stake / year
$35,685
Hours at stake / month
39.7

Time available for other work. This value does not mean payroll falls or cash is saved.

Monthly net cash change
$0
Upfront audit + implementation
$7,500
Simple cash payback
No positive cash payback
First-year net cash, after upfront cost
-$7,500

Cash model assumes constant monthly savings and costs from month one. It excludes tax, financing, ramp time, and the time value of money.

How to read this model →
CAPACITY AT STAKE = MONTHLY HOURS × CAPTURE ASSUMPTION × HOURLY COST × 12 · CASH PAYBACK = ($7,500 AUDIT + BUILD COST) ÷ POSITIVE MONTHLY NET CASH · ESTIMATES, NOT GUARANTEED RESULTS
SIGNAL & NOISE · THE SHOW

One claim per episode.
The arithmetic included.

Signal & Noise is a show about AI in owner-led companies, told with the arithmetic. Each episode opens one claim about AI in business on real work and keeps the assumptions in view: what the tool does, what the people do, what the hours become, and whether any cash moves. One useful finding per episode, with the number that held and the number that didn’t. Hosted by Paul Mattson.

All five episodes publish together. Subscribers to the Letter hear the date first.

  1. E01

    The model before the claim

    Where AI stands in business, why the way it’s sold skips the part that matters, and the arithmetic that has to happen before a capacity figure becomes a decision.

  2. E02

    Two pieces of good news that added to zero

    Price up, volume up, margin flat: the calculation that finds the explanation, and the decision it leaves with the person who runs the business.

  3. E03

    The order of operations

    Which workflow goes first, and how you prove it before you promise anything.

  4. E04

    Where the hours go

    Where released hours go, who decides, and why writing down “absorbed” is the control.

  5. E05

    A saved hour is a stopped task

    How you know an hour has actually left.

FIVE EPISODES · EVERY FIGURE ON AIR IS SOURCED IN THE SHOW NOTES
FIELD NOTES · HOW THE WORK IS APPROACHED

The method, in the open.

Practical notes on reliable reporting, honest business cases, and useful AI. Each is a method you can inspect before a conversation.

THE OPERATOR

Senior judgment.
Direct involvement.

Paul Mattson, founder of Unwrapped, seated at his desk with his glasses in his hands
PAUL MATTSON · FOUNDER, UNWRAPPED

A career in operations, project management, data analytics and consulting, including six years advising organizations from the Fortune 10 to owner-operated businesses. Unwrapped brings that discipline to owner-led companies, and I do every audit myself, from the fit call to the readout.

Unwrapped is new, so I show the method instead of borrowing results: the sample audit, the arithmetic and the assumptions are all on this site. Client results will appear here once they’re measured and the client agrees. Unwrapped is part of Intelligence Unleashed LLC’s growing portfolio of brands.

HOW THE ENGAGEMENT RUNS

  • Direct work with Paul from the fit call to the readout
  • Agreed scope and written acceptance criteria
  • Assumptions, exceptions and sources in view
  • Specialist involvement agreed before it begins
Discuss your workflow →
PAUL MATTSON · LANCASTER, SC · CHARLOTTE REGION · REMOTE ENGAGEMENTS

THE LETTER · BY PAUL MATTSON

Dated. Signed.
Every figure sourced.

One finding about AI in owner-led companies, every two weeks, with the number that held and the number that didn’t.