One plus one
makes eleven.
We map how your business actually runs, then build the AI agents and workflows that run it better — under a written reliability standard, with licensed professionals reviewing everything regulated before it reaches you.
You bring the question. We draw the machine behind it.
Nobody hires us for AI. They hire us because something takes eleven days that should take two, and no one can say which of the eleven is judgment. Here is what the diagnostic actually does with that.
We build into your stack. We don’t ask you to leave it.
Agents read from and write to the systems your team already lives in — practice management, ledgers, portals, property software, HRIS, the drive, the inbox. Nothing here requires a migration.
All logos, product names and company names are trademarks of their respective owners, shown solely to identify systems we integrate with. No affiliation, partnership, sponsorship, or endorsement is implied.
A.G.E.N.T. — five moves from “this is slow” to “this is proven.”
Every engagement runs the same playbook, in the same order, with the same artifact at the end of each stage. It is not ours — it is the A.G.E.N.T. playbook from The agent-centric enterprise in the Harvard Data Science Review, which is the standard we trained on and the standard we hold ourselves to.
Adapted from Hofmann, D. & Kruhse-Lehtonen, U., “The agent-centric enterprise,” Harvard Data Science Review (DAIN Studios). See how we run each stage →
Priced to the work we scope together. A national rollout and a six-person practice are never the same number — and never the same quote.
Of listed businesses never sell. Messy financials are a leading killer — we fix that before it costs you.
Source: IBBA Market Pulse reporting
Actively licensed US CPAs remain, down from ~1.9M in 2019. We built for the shortage: agents do volume, CPAs review and approve.
Source: AICPA Trends · NASBA
Every system we build is the same five-stage engine.
Signal
Watch the sources that matter — dockets, deadlines, ledgers, portals, pages. The signal layer is the moat.
Enrich
Pull the context that turns a raw event into a fact about your business.
Score
Decide if it matters, how much, and how confident we are — in numbers, not vibes.
Compose
Draft the artifact: the reconciliation, the filing, the follow-up, the report.
Act
File it, chase it, reprice it, escalate it — or hand it to a human, depending on the tier.
Only two things change between a legal docket watcher and a bookkeeping close agent: what stage one watches, and what stage five is allowed to do. That's why our builds compound — each engagement hardens the same spine. Errors compound in a system of agents; so do solutions. Which one you get is a design decision, made early.
Everything we sell carries one of four labels.
So you always know what you're buying, how it's priced, and who's accountable for it.
Consulting
Intake and diagnosis, workflow redesign, GAAP and compliance interpretation, exit readiness, audit prep. Judgment-heavy work, done with you — and the front door to everything else.
Agents
Software teammates that run a repeatable task end-to-end: bookkeeping and close, eligibility checks, document intake, monitoring. Reviewed by licensed professionals, measured against day one.
Dashboards
Your own data, queryable in plain English — with sourced answers or a refusal, never a guess. Persistent memory of decisions and documents for leadership.
Training
We teach your organization to use AI safely and well — executive briefings that book directly, and workflow-anchored training built on what we implement for you.
Where firms start, by industry.
Pick your world. Each tab shows what we typically take off a team's plate first — and the playbook behind it.
Trust is a design constraint, not a promise.
Per-client isolation. Nothing trains on your data. Licensed sign-off on everything regulated. Sourced answers or a refusal.
Diagnose. Propose.
Deliver. Compound.
Every engagement runs the same arc — and every stage produces something you keep, whether or not you continue to the next.
Diagnose
We map how work actually flows through your firm — the systems, the handoffs, the workarounds nobody wrote down. You get a written diagnostic: what to automate, what to fix first, and what to leave alone. It's a real deliverable you keep either way.
Propose
You receive a written proposal with package options — only what the diagnosis supports. If the honest answer is advice, an introduction, or nothing, that's what the proposal says. You choose; we never choose for you.
Deliver
Agents are built against our written reliability standard: autonomy tiers per task, a prohibited-actions register, and a separate examiner who checks what the builder built. Licensed CPAs review financial deliverables. A human approves anything that leaves the building.
Compound
We measure against your day-one baseline — hours, error rates, cycle times — and publish the delta to you monthly. What proves out, scales. What doesn't, we say so and stop. Proof, then scale; never the reverse.
What we will not do — printed, not implied.
An honest system is defined by its boundaries. These are contractual, not aspirational.
- No agent signs, files, or pays. Returns, government filings, and payments always carry a human signature.
- No tax positions, no valuation opinions. Judgment that belongs to a licensed professional stays with one.
- Nothing unsupported by the diagnosis. If you don't need it, it isn't in the proposal — even when it would be easy to sell.
- No claims we can't reproduce. Every number we publish about our own work traces to a real engagement record.
- No surprise invoices. Whatever we agree becomes a written ceiling before work begins, and it holds for the term.
People are the other half of eleven.
Executive AI briefing
For leadership and boards: what AI is and isn't, the risks that apply to your organization, and the questions you'll be asked. Books directly — no diagnosis required.
Workflow-anchored training
Built on your diagnosis and what we implement, so your people learn on their own work — the kind of training that sticks. Governance, acceptable-use policy, and standing office hours where the diagnosis supports them.
We build capability — we don't certify it.
Two ways in: your industry,
or the function that hurts.
The same engine, cut two ways. Choose the vertical you live in, or the part of your business that needs to run better — the diagnosis meets you at either door.
Accounting — The Books Ladder
Wherever your books are today — shoebox, spreadsheets, or a close that slips every month — there's a rung for that, and a defined path up. Agents assemble; licensed CPAs review and approve; every step is measured. Explore the ladder below.
Finance & FP&A
A 13-week cash forecast that reconciles to the ledger — not to hope. Variance flagged with sources attached.
HR & people
Onboarding and offboarding chases that never lose a step; credential and license expiry watched continuously.
Operations
The exception board: one queue for everything stuck, with an owner and a clock on every item.
Marketing & growth
First-party lifecycle triggers on consented channels only. Relevance from your own data — no scraping, no purchased lists.
Sales & CRM
Pipeline hygiene enforced automatically; quote-to-close follow-up that's polite, relentless, and logged.
IT & data
Access reviews on schedule, unused licenses reclaimed, backups verified by restoring them — not by trusting the checkbox.
Compliance & risk
Obligations tracked from the governing documents, filings calendared with margins, evidence collected as the year happens.
Legal & contracts
Renewal and notice dates computed from the contracts themselves; obligations surfaced before they bite.
Customer service & front office
Intake triaged, routine questions answered from approved sources, the rest routed to a human with context attached.
Procurement & vendors
Invoices reconciled against contracts and price lists; vendor documents chased; renewals never auto-renewed unseen.
Data & reporting
Numbers assembled, sourced, and explained on schedule — your own data, queryable in plain English.
Five rungs. You enter wherever your books actually are.
Real estate & affordable housing
Development programs, property operations, resident notices, rent and subsidy reconciliation — the operational spine of a portfolio, run on time.
LIHTC compliance
Low-income housing tax credits, specifically: income certifications and recerts, agency deadlines, audit files — where documentation is everything and deadlines are federal.
General compliance — any industry
Whatever your regulator requires: obligations tracked from the governing documents, evidenced continuously, never late. Portable across sectors.
Dental & medical practices
Eligibility checks before visits, claim assembly, denial chase, recall outreach — the intake-heavy workflows that eat front-office hours.
Legal & policy firms
Docket and deadline tracking, matter intake, legislative monitoring fused with your client book — every alert lands as per-client positioning.
Restaurants & hospitality
Supplier invoices checked against agreed prices, inventory variance, license renewals, daily sales and labor reporting — margin defense, automated.
Municipal & public agencies
Permitting prescreens, program intake, records requests, constituent services — pilot-sized, procurement-aware, framework-aligned.
Owner-led firms preparing to sell
Recast financials, documented add-backs with evidence, a data room that survives diligence — engaged by the owner, welcomed by their broker.
Makes Eleven is a general consultancy — any firm, any industry, case by case. As engagements succeed, the winning playbooks get packaged. These are the verticals where our playbooks already run deepest: proof of range, not a fence around it.
Built for cities that
answer to everyone.
Government work is where careful AI matters most — and where it's already working: permitting prescreens that cut months to days, constituent services in dozens of languages, records processing that keeps pace with the law. We bring that discipline home to the Lowcountry and the District.
Pilot-sized by design
Diagnostics and pilots scoped to fit municipal small-purchase processes — proof before procurement, value before scale. No seven-figure platform bets.
Framework-aligned
Deployments follow the NIST AI Risk Management Framework and state guidance, with human approval, audit trails, and records-retention awareness built in from day one.
Equity, tested
Public systems must serve everyone. Bias and disparate-impact testing is standard in our deployments — an engineering step, not an afterthought.
Records-aware
FOIA and public-records obligations shape the architecture: what's logged, what's retained, what's producible — decided before the first document moves.
Start where the queue is longest.
- Permitting triage and prescreen
- Housing program intake and eligibility
- FOIA and records-request processing
- 311 and constituent services
- Grants management and reporting
- Forms digitization and legacy intake
DC and the Charleston region.
Where housing, growth, and service delivery are moving fast — and staff capacity isn't keeping up. The founder's background spans affordable-housing finance, compliance, and the rooms where initiatives actually move.
Careful by architecture,
not by promise.
Most firms answer trust questions with adjectives. We answer with architecture — the constraints are built in, written down, and checkable.
Your data, isolated
Per-client credentials, storage, and memory. Agents read only sources you approve. Nothing you share trains any AI model — contractually. A full audit trail records every agent action.
Compliance built in
HIPAA Business Associate Agreements before any patient data moves. Confidentiality-first architecture for legal clients. Records-retention and FOIA awareness for public-sector work.
Licensed humans sign off
Financial deliverables are prepared by our agent systems and reviewed and approved by licensed CPAs in our delivery network — disclosed plainly in every engagement letter, because trust compounds too.
Sourced answers or a refusal
Our dashboards and research agents cite what they know and say so when they don't. No guesses dressed as answers — in your business or in ours.
Every agent ships against a reliability standard you can read.
Not a marketing page — an engineering document that governs every build.
- Autonomy tiers, per task. Each agent task carries an explicit tier — from fully checked to fully autonomous — assigned by risk, never by convenience.
- A prohibited-actions register. Thirty-six classes of action our agents are built to refuse, from signing filings to touching payments. It grows; it never shrinks.
- The examiner is never the builder. Every system is checked by someone who didn't build it, against test sets with known answers.
- Measured against day one. Your baseline is recorded at the start; every claim of improvement is a delta against it, shown to you monthly.
- Recomputable output. Where the work is arithmetic — reconciliation, close, eligibility — a second, independent computation checks the first.
What our agents never do.
- Never sign or file a tax return, government filing, or legal document. A licensed human signs; the agent assembles.
- Never move money. Payment initiation, transfers, and payroll runs require a human on the button — every time.
- Never take a tax position or issue a valuation opinion. Judgment calls belong to licensed professionals.
- Never send unreviewed work to a regulator, court, or counterparty. External release is a human decision.
- Never guess. Below the confidence threshold, the answer is a question to a human — not a plausible-sounding sentence.
The myths, examined.
The evidence, cited.
The most common concerns we hear from operators and public officials — taken seriously, and answered with primary sources. AI is already trusted where the stakes are highest: medicine, law, government, finance. If it's safe enough there, it's worth understanding here.
"AI will replace my people."
The Yale Budget Lab examined 33 months of labor data after ChatGPT's release and found no discernible disruption to overall employment — findings echoed by the Dallas Fed and RAND, which reports more businesses adding jobs from AI adoption than cutting them. The consistent pattern is augmentation: AI absorbs the volume; people keep the judgment, the relationships, and the accountability.
How we handle itOur engagements are designed around your team, not instead of it — and our enablement work exists precisely to build your people up. People are the other half of eleven.
"AI makes things up too much to trust with real work."
Unsupervised AI does invent — which is why serious deployments never run unsupervised. The FDA has authorized over 1,400 AI-enabled medical devices under formal validation; courts now formally require human verification of AI-assisted work. The fix isn't hope; it's structure.
How we handle itThe same structure: agents do volume, licensed professionals review financial deliverables, every answer is sourced or refused, and a human approves anything that leaves the building.
"Our data will end up training some model — or leaking."
That risk is real with consumer tools and absent with properly configured enterprise deployments: contractual no-training terms, tenant isolation, encryption, and audit logs are standard. California's court system draws exactly this line — prohibiting confidential data in public AI systems while permitting managed, internal ones.
How we handle itPer-client isolation, allowlisted sources only, no training on your data ever, and a written security posture you can hand to your counsel.
"AI is too legally risky for government."
Government frameworks don't prohibit AI — they govern it. NIST's AI Risk Management Framework, federal agency guidance, South Carolina's state AI strategy, and the SC judiciary's generative-AI policy all chart responsible adoption. Cities are already deploying: permitting prescreens, multilingual constituent services, records automation.
How we handle itFramework-aligned deployments with human approval, audit trails, records-retention awareness, and bias testing as standard — pilot-sized to fit procurement realities.
"It's a fad."
Legal AI reached in three years the adoption level cloud computing took a decade to hit. FDA authorizations of AI medical devices grew from a handful a year to nearly three hundred in 2025 alone. Nearly a third of practicing lawyers now use generative AI. Regulated professions are the slowest adopters by design — and they've adopted.
How we handle itWe don't sell momentum; we sell diagnosed fit. If AI isn't the answer to your problem, that's the report you'll get.
"We're too small — or not technical enough — for this."
The organizations winning with AI aren't the most technical — they're the ones that paired with people who know the terrain. Most successful deployments involve an outside partner, and modern pilots stand up in weeks, not years.
How we handle itYou don't need in-house engineers. The diagnosis tells us what fits your size and budget; a written ceiling means no surprises; and our enablement work leaves your team able to run what we build.
"AI in public systems invites bias and civil-rights problems."
The risk is documented — a state attorney general reached a $2.5 million settlement with a lender whose underwriting model disadvantaged Black and Hispanic applicants. That case also defines the fix: test models for disparate impact, keep humans on decisions, and document everything. Responsible deployment is a discipline, not a hope.
How we handle itDisparate-impact testing, human approval on consequential decisions, and full audit trails are standard in our public-sector and lending-adjacent work — not add-ons.
"AI consultants are just reselling tool tutorials."
Many are — which is why so many AI pilots die within a year. Deployments fail when nobody maps the workflow first; they succeed when the diagnosis comes before the prescription.
How we handle itWe're a consulting firm that uses AI, not an AI vendor with a pitch. The diagnostic is a real deliverable you keep either way, the proposal contains only what the diagnosis supports, and every claim we make in public traces to a real engagement.
Built by an operator,
not a demo reel.
The name is the thesis: paired with the right partner, a business doesn't add capability — it compounds it. One plus one makes eleven.
Leon Fields, Founder
An economist by training, an operator by habit, and a builder by conviction. Leon studied economics and community development at Howard University, earned his MBA at William & Mary with a dual specialization in finance and in entrepreneurship & innovation, and is completing advanced professional studies in agentic AI at Harvard.
Before founding the firm, he earned a banking innovation award for process improvement at one of the ten largest US banks, worked inside affordable-housing finance and LIHTC compliance, and used AI to successfully resolve two legal disputes of his own — including a federal matter opposite a major national law firm.
He has sat on your side of the table.
- Howard University — economics & community development
- William & Mary MBA — finance · entrepreneurship & innovation
- Harvard — advanced professional studies, agentic AI
- Banking innovation award — top-ten US bank
- Affordable-housing finance & LIHTC compliance
Some problems are operational. Some are technical. Some move only when the right people are in the room.
Operations
Workflow mapping, process redesign, the diagnosis itself. The lane everything else depends on.
Automation
The agents, dashboards, and systems — built to the written standard, measured against day one.
Relationships
Policy rooms, procurement processes, the introductions that move initiatives. Compounding requires every gear turning.
Three commitments, made before any work begins.
Not a platform, not a pilot programme, not a twelve-month roadmap. One process, one accountable person, one month of running both ways and timing them.
Not a platform. A single workflow — audited, redesigned, governed and measured. If it does not survive measurement, we say so.
Named before we start: the person who can stop it. Authority to halt the system belongs to someone specific, on day one, in writing.
Run the manual way and the agent way side by side, and time both. The comparison is against what would have happened anyway — not against a flattering baseline.
Let's find your eleven.
Step one is a conversation and a diagnosis — not a contract. Take the Workflow Mapper below and it becomes your pre-call assessment, so the call starts at the diagnosis, not the introductions.
The Workflow Mapper
Ten minutes, no account. Runs entirely in your browser — nothing is stored or sent until you choose to book.
Scoped before it is priced.
We do not publish a rate card, because the work is never the same twice. A national firm auditing every entity and a six-person practice fixing one workflow are different engagements, and pricing either from a template would be dishonest to one of them.
We scope the work before anyone quotes it
The diagnostic is quoted to the size of your firm and the ground it covers, and it is credited toward whatever engagement follows. You keep the written map either way — including the version that says you need a checklist and one hire, not agents.
The number is agreed before work begins
Your proposal names the scope, the price, and a ceiling that holds for the term. Nothing is billed that was not scoped, and nothing is scoped that the diagnosis did not support. Executive briefings book directly, no diagnosis required.
The conversation is free.
The diagnosis is where it starts.
Built for the way your
industry actually works.
The same five-stage engine, tuned to the documents, deadlines, and regulators of your world. Every engagement starts with the diagnosis — these are the worlds where our playbooks already run deepest.
Real estate & affordable housing
Development programs, property operations, and resident-facing workflows — the operational spine of a portfolio, run on time, with evidence attached.
Explore →LIHTC compliance
Low-income housing tax credits, specifically: income certifications, recert calendars, agency deadlines, audit files. The credit is won or lost in the paperwork.
Explore →General compliance — any industry
Whatever your regulator requires — obligations tracked from the source documents themselves, evidenced continuously, and never late. Portable across sectors.
Explore →Dental & medical practices
Eligibility before every visit, claims assembled and chased to payment, recalls that actually go out — the intake-heavy work that eats front-office hours.
Explore →Legal & policy firms
Docket and filing deadlines tracked from the documents, matters opened clean, legislative monitoring fused with your client book.
Explore →Restaurants & hospitality
Supplier invoices checked against agreed prices, inventory variance flagged weekly, licenses renewed with time to spare. Margin defense, automated.
Explore →Municipal & public agencies
Permitting prescreens, program intake, records requests, constituent services — pilot-sized, procurement-aware, framework-aligned.
Visit the practice →Owner-led firms preparing to sell
A recast P&L, add-backs with evidence, a data room that survives diligence — engaged by the owner, welcomed by their broker.
Explore →Of listed businesses never sell — and messy financials are a leading killer.
Source: IBBA Market Pulse reporting
What AI permitting prescreens have done to review queues in cities already deploying them.
Source: public-sector deployment reporting
Actively licensed US CPAs remain, down from ~1.9M in 2019 — the shortage every regulated industry feels.
Source: AICPA Trends · NASBA
Three things stay constant.
Diagnose first
The diagnostic maps your workflow before anything is proposed. If the honest answer is advice, not software — that's the report you get.
A written ceiling
Scope and price are agreed before work begins, and the ceiling holds for the term. No industry, no exception.
Measured against day one
Your baseline is recorded at the start; every claim of improvement is a delta against it, shown to you monthly.
Asked by operators in every industry.
Do you only work in these industries?
No — Makes Eleven is a general consultancy: any firm, any industry, case by case. These are the verticals where our playbooks already run deepest. If yours isn't listed, the diagnostic works exactly the same way.
What does an industry engagement actually start with?
The diagnostic. You get a written map of how work flows through your firm, what to automate, what to fix first, and what to leave alone. It is scoped to your firm, credited toward whatever follows, and it is a deliverable you keep whether or not you continue.
Who reviews the regulated output?
Licensed professionals. Financial deliverables are prepared by our agent systems and reviewed and approved by licensed CPAs in our delivery network — disclosed plainly in every engagement letter. Agents never sign, file, or pay; that's contractual.
Will this replace the systems we already use?
No. The agent layer sits on top of your existing systems by API — we never migrate your ledger, your practice management system, or your CRM. Your data stays where it lives; the agents do the work between the systems.
How is pricing set?
By scope, never by template. We do not publish a rate card, because a national rollout and a single-office practice are not the same work. The diagnosis establishes what the work actually is; the proposal names the price and a written ceiling that holds for the term.
Your industry, mapped in ten minutes.
Every function of the business.
One engine underneath.
Accounting is the flagship — but the same signal → enrich → score → compose → act machine runs any function where documents move, deadlines bind, and follow-up decides outcomes.
Accounting — The Books Ladder
Five rungs from clean records to diligence-ready books. Agents assemble; licensed CPAs review and approve; every step is measured against your day-one baseline.
Finance & FP&A
A 13-week cash forecast that reconciles to the ledger — not to hope. Variance flagged with sources attached.
Explore →HR & people
Onboarding and offboarding chases that never lose a step; credential and license expiry watched continuously.
Explore →Operations
The exception board: one queue for everything stuck, with an owner and a clock on every item.
Explore →Marketing & growth
First-party lifecycle triggers on consented channels only. Relevance from your own data — no scraping, no purchased lists.
Explore →Sales & CRM
Pipeline hygiene enforced automatically; quote-to-close follow-up that's polite, relentless, and logged.
Explore →IT & data
Access reviews on schedule, unused licenses reclaimed, backups verified by restoring them — not by trusting the checkbox.
Explore →Compliance & risk
Obligations tracked from the governing documents, filings calendared with margins, evidence collected as the year happens.
Explore →Legal & contracts
Renewal and notice dates computed from the contracts themselves; obligations surfaced before they bite; drafts assembled for counsel.
Explore →Customer service & front office
Intake triaged, routine questions answered from approved sources, everything else routed to a human with context attached.
Explore →Procurement & vendors
Invoices reconciled against contracts and price lists, vendor documents chased to completion, renewals never auto-renewed unseen.
Explore →Data & reporting
The numbers assembled, sourced, and explained on schedule — plus your own data, queryable in plain English, with sourced answers or a refusal.
Explore →Enablement & training
Executive briefings for leadership, workflow-anchored training for teams — people are the other half of eleven.
See enablement →How function engagements work.
Can we start with just one function?
That's the recommended path. The diagnosis picks the function where proof is fastest, we run it to a measured result, and expansion follows evidence — never enthusiasm.
Why is accounting the flagship?
Because books are where value, financing, and exits are decided — and because bookkeeping is recomputable: a second, independent calculation can verify every number the agents produce. It's the function where our verification architecture works hardest for you.
Do these connect to our existing tools?
Yes — by API, on top of what you already run. We never migrate your ledger or CRM to something new; the agents work between your systems and leave the source of truth where it is.
What never gets automated?
Signatures, filings, and payments; tax positions and valuation opinions; anything leaving for a regulator, court, or counterparty without human review. The full refusal list is printed on the Method page — it's contractual, not aspirational.
Ten minutes. Your map. No account.
Notes from the founder —
and the reading behind them.
Positions, industry updates, and the primary sources worth your time. Everything here follows the house rule: sourced answers or a refusal — in your business and in ours.
The Makes Eleven letter
Occasional, short, and specific: what changed in AI for regulated work, what it means for operators, and what we'd do about it. No volume commitments, no filler.
Positions, in writing.
Why the diagnosis comes first — and why we publish no rate card
Most AI engagements fail before the first line of configuration, because nobody mapped the workflow the software was supposed to run.
Read the note
Every failed automation project I've examined shares one property: the prescription came before the diagnosis. A tool was chosen, then a problem was found for it. We run the arrow the other way. Nothing gets proposed until we have mapped how work actually moves through the firm, and the map is a written deliverable you keep whether or not you continue.
That sequence is also why there is no rate card on this site. A published number is a promise made before the work is understood, and it fails in both directions: a national firm auditing every entity gets quoted as though it were a single office, or a six-person practice gets quoted as though it were an enterprise. Both are wrong. The scope has to exist before the number does, so we scope first and price to what we find.
What replaces the rate card is a discipline, not a mystery. The diagnosis is quoted up front and credited toward whatever follows. The proposal names the scope, the price, and a ceiling that holds for the term. Nothing gets billed that was not scoped, and nothing gets scoped that the diagnosis did not support.
The honest finding — "you don't need agents, you need a checklist and one hire" — costs us the engagement and still gets written down. That finding is in more diagnostics than you would think. If a proposal ever contains something the diagnosis doesn't support, you're holding the wrong proposal. That's the whole method.
We built for the CPA shortage on purpose
Roughly 650,000 actively licensed CPAs remain in the US, down from about 1.9 million in 2019. That's not a staffing problem — it's an architecture problem.
Read the note
The profession isn't coming back to its old headcount, and every firm that needs reviewed financials is competing for the same shrinking bench. You can respond by paying more for the same hours — or by changing what the hours are spent on.
Our architecture does the second thing: agents do the volume — categorization, reconciliation, assembly, chase — and licensed CPAs do what only licensed CPAs can do: review, judge, approve, sign. The disclosure is printed in every engagement letter, because the model only works if the client knows exactly who did what.
The shortage is the market evidence for the whole design. A CPA reviewing agent-prepared workpapers covers a multiple of the clients they could serve preparing everything by hand — and the work they keep is the work their license was actually for.
GAAP is a ladder, not a leap
Full GAAP on day one is over-prescription for most owner-led firms. The honest version is five rungs — and an honest stopping point.
Read the note
GAAP is the reference standard buyers, lenders, and bonding agents actually use — which is exactly why it gets oversold. A firm with no exit on the horizon, no institutional financing, and no outside investors doesn't need ASC 842 workpapers; it needs reconciled accounts and a close that happens on a date.
So we sell a ladder: records → reconciled → accrual → GAAP-ready → diligence-ready. Clients enter wherever their books actually are, and full GAAP is triggered by something real — an exit inside two years, a covenant, a bond, an investor — not defaulted. No trigger? Stopping at rung three is the "never sell what you don't need" doctrine, applied to our own flagship.
The rungs above stay lit for the day something real turns on. That's the difference between a ladder and a leap: you can stand on a ladder.
The sources worth your time.
The primary documents behind the claims we make — read them yourself.
- NIST AI Risk Management FrameworkThe governance framework our public-sector deployments align toNIST.GOV ↗
- FDA AI-enabled device list1,400+ authorized devices — regulated AI adoption, documentedFDA.GOV ↗
- The Budget Lab at YaleThe labor-market evidence on AI and employment, measuredYALE.EDU ↗
- AICPA Trends reportThe pipeline data behind the CPA shortageAICPA ↗
- IBBA Market PulseQuarterly brokerage data on the businesses that sell — and the ones that don'tIBBA.ORG ↗
Curation rule: primary sources and measured data only — no vendor decks, no hype cycles.
The engine arrived. The road didn't change.
Almost every firm has bought the tools. Very few have changed the work. That gap — not the technology — is the whole story of the last three years, and it is the reason a small firm can now outrun a large one.
Rockets on a horse.
“We are, in many cases, bolting a jet engine onto a horse carriage — and wondering why it doesn’t fly.”
This is what most AI adoption looks like. The model is genuinely powerful. It is strapped onto a process that was shaped entirely by human limits — an eight-hour shift, a sequential handoff, an approval that waits for Monday, a form designed to be read by a person.
The thrust is real. The animal underneath it is the constraint. You get a faster horse, a louder one, and an expensive one — and the finish line does not move.
The tell is easy to spot in your own firm: the tool got adopted, everyone says it helps, and no line on the P&L moved.
A Formula One engine in city traffic.
“The engine’s power is there, but the full potential is squandered by the limits of the road.”
The engine’s power is real and entirely unavailable. The limit is the road, not the driver — every stoplight is a hand-off, every intersection an approval queue. You do not fix this by tuning the engine.
Suppose the technology is right and the deployment is competent. It still sits inside a process with a red light at every intersection: a queue, a weekly meeting, a shared inbox, a person who has to notice something before anything else can happen.
An agent that can work through the night is worth nothing if the next step waits until someone opens a folder on Tuesday. Throughput is set by the slowest gate, not the fastest actor — which is why so many pilots post a real speed gain on one step and no change at all in cycle time.
This is the diagnosis most firms never get: not which tool, but which intersection.
Fig. 2 — Throughput is set by the gate, not the engine
This is not a hunch. It is the most consistent finding in the field.
Three independent bodies of research, all pointing at the same gap between adoption and earnings. We publish the sources so you can check them.
Of organizations report using AI in at least one business function — up from 78% the year before. The tools are, effectively, everywhere. Source: McKinsey Global Survey on AI, 2025
Report any measurable effect on enterprise-level EBIT — and among those, most attribute under 5% of it to AI. Widespread use, narrow result. Source: McKinsey Global Survey on AI, 2025
Of enterprise generative-AI pilots studied delivered no measurable P&L impact. The authors attribute the divide to approach, not to model quality. Source: MIT Project NANDA, The GenAI Divide: State of AI in Business 2025
Two operating models. The gap is not the tools — it is six design decisions.
Every firm sits somewhere on this table. Almost none of it is about which model you bought; nearly all of it is about how the work was shaped before any model arrived.
Comparison and the reported productivity ranges from Hofmann & Kruhse-Lehtonen, “The agent-centric enterprise,” Harvard Data Science Review, citing Noy & Zhang (2023), Riedl et al. (2023), OECD (2025), McKinsey (2025) and BCG (2025). The second column is the destination, not a promise — which row you can honestly move is exactly what a diagnosis is for.
Rebuild the road.
Nothing above is an argument against the technology. It is an argument about where the work happens. When the process itself is redesigned so an agent can act — data reachable, decisions written down, gates placed deliberately rather than by accident — the same engine finally gets to run.
That redesign is a specific, ordinary, repeatable exercise. It has five moves, it takes weeks rather than quarters, and it starts with one workflow, not a transformation program.
It also has a boundary that never moves: the gates you keep are the ones that carry judgment, liability, or a signature. Those are not inefficiencies to be removed. They are the reason anyone trusts the output.
The point was never headcount. It was where your best people spend Tuesday.
When agents handle routine execution, people move up to problem framing, trade-off decisions, stakeholder alignment, and leadership under uncertainty. You hired them to think and reason, not to retype. Your best people become considerably more valuable, not less necessary.
Manual work produces variation in quality. A designed workflow sets a floor that holds no matter who is in the office, on vacation, on leave, or brand new. Consistency is usually worth more to a regulated firm than raw speed.
The measure quietly shifts from return on investment to return on inference: what each unit of machine reasoning actually produced for the business. It is a harder number to fake than a licence count, which is precisely why we track it.
Autonomy is how much a system may do alone. Agency is how much your people can direct it. We build for agency and raise autonomy only where the evidence earns it — never the reverse, and never quietly.
Pick one workflow. Map it. Then decide.
The Workflow Mapper takes about ten minutes and needs no account. You leave with your own map — the steps, the gates, and where a machine could carry weight — whether or not you ever hire us.
A.G.E.N.T. — Audit, Gauge, Engineer, Navigate, Track.
Five moves that take a workflow from “this is slow and nobody knows why” to “this is measurably better and the CEO believes the number.” Every engagement runs them in this order. Each stage ends in an artifact you keep.
The playbook doesn’t start with a workflow. It starts with what the business is for.
Picking a workflow first is how firms end up automating something that works fine. Four questions come before Audit, and they take an afternoon, not a quarter.
What is the firm actually trying to do this year? Everything downstream has to trace back to a line on this list or it doesn't get built.
What role should machines play here at all — and, just as important, where you have decided they should not. This is where refusals get written.
Systematically, across the value chain, not from a vendor list. In most firms the biggest pain points are already common knowledge — the work is confirming which are tractable.
Strategically important, but safe enough to experiment on. One workflow, chosen deliberately — not a transformation program.
Audit
“What is actually happening, and why does it fail?”
Understand how the work is done today and what outcome really matters — not what the process document says, and not what the org chart implies. Most of the value of the whole engagement is decided here, because everything downstream inherits whatever this stage got wrong.
- Map goals, data, systems, roles
- Interview users; capture pain points
- Document desired outcomes, not just outputs
- Sit with the people doing the work, not only the people managing it
- Start from the relevant Workflow Atlas board so you edit a draft rather than face a blank page
- Record the gates that exist by design and the ones that exist by accident
A clear picture of the as-is and the target outcome, with no unknowns — written down, in your hands, useful even if you stop here.
Gauge
“What would success look like for the business?”
Evaluate each workflow on impact, repeatability, complexity and outcome potential. This is the stage that decides what we will not build. A step that is high-risk, low-volume and dependent on judgment is a step that should stay exactly where it is.
- Score steps on repeatability and risk
- Estimate outcome upside
- Every step lands in one of five categories: automate, draft & review, assurance, watch & alert, or human-always
- The human-always register is written before anything is built, and it may grow but never shrink
- We pick the success number here — one number, agreed in writing, before a line of work starts
A prioritized shortlist of agent opportunities and, just as importantly, a documented list of what we declined to touch and why.
Engineer
“If you started fresh with AI, what would this look like?”
Redesign and build the flow so an agent can act: data accessible, decisions explicit, handoffs deliberate. This is where the road gets rebuilt rather than repaved — and where most programs quietly skip ahead to tooling instead.
- Refactor the process for straight-through flow
- Build or configure the agent
- Define success metrics and guardrails
- Everything is built on the same five-stage spine — signal, enrich, score, compose, act
- Autonomy is set per task against our written reliability standard, not per system
- A separate examiner checks what the builder built; licensed professionals review anything regulated
A working agent-first process ready to pilot, with measurable success criteria attached to it.
Navigate
“Who decides what, and what happens when AI is wrong?”
Shape the human–agent relationship: transparency, intervention paths, governance. Assume the system will be wrong at some point, and design the moment of being wrong before it happens rather than after.
The framing that matters here: the system is not doing the regulated work. It is preparing expert work for controlled approval — which is why accountability can stay exactly where the risk is created.
- Add explainer UI, override and escalation paths
- Train staff on the new oversight roles
- Embed compliance checks
- Every output carries its sources, its confidence and the record of what it touched
- Signatures, filings and payments stay human — printed in the contract, not implied
- The people who will supervise the agents are in the room while it is designed, not briefed afterward
- Human review is a catalyst, not just a brake — validated output is what earns a system the right to more autonomy later
High trust and real human control that holds even as autonomy rises — because the escalation path was built, not promised.
Track
“Would your CEO believe these numbers?”
Prove value against the outcome chosen back in Gauge — throughput up, resources down, reach up. And show the honest comparison: not just before and after, but what would have happened anyway versus what the redesign actually did.
- Instrument KPIs and dashboards
- Run A/B or before-and-after comparisons
- Feed learnings into the next Audit cycle
- Measured against your day-one baseline and published to you monthly
- A short dashboard on purpose: source, freshness, confidence, exceptions needing a human
- What proves out scales. What doesn't, we say so and stop — proof, then scale, never the reverse
Tangible outcome gains that fund the next move — or a documented reason to stop, which is a legitimate result and one we will hand you.
Five kinds of agent. The question is never “where can we use AI.”
It is “which kind of agent should own this step, and where does a human stay in control.” A taxonomy is what connects the technology to your actual pain points and bottlenecks — without it, every conversation collapses into tool shopping.
Taxonomy adapted from Hofmann & Kruhse-Lehtonen, “The agent-centric enterprise,” Harvard Data Science Review. The Guardian row is where our reliability standard lives: a separate examiner checks what the builder built.
One workflow. Roughly eight weeks. Then you decide again.
The published sprint runs in three phases. We keep that shape because it forces a visible result early, while the people who will live with the system are still in the room.
Identify the high-value workflow, map its current state, and find where the time actually goes. Choose the one number that will decide whether this worked.
Build and deploy the first agent workflow, with attention on data accuracy and the routine work that consumes the day. Guardrails ship with it, not after it.
Scale what works, learn from what didn't, wire the oversight, prove the number — then repeat the cycle on the next workflow.
Sprint structure adapted from Hofmann & Kruhse-Lehtonen, “The agent-centric enterprise,” Harvard Data Science Review. Scope, duration and price are set per engagement — a six-person practice and a national rollout are never the same shape.
The Audit starts with a map. Draw yours now.
The Workflow Mapper is the first half hour of Audit, run by you, for free. It becomes the pre-call assessment if we do speak — and a useful document if we never do.