AllodialStrategy

Strategy memo · October 2026

Four people, $2M, and a machine a county can own. How do we get paid?

We tested eight ways to make money against what the big consulting firms actually earn from AI and what our own research already settled. Three survive. Combined in the right order, they make the seed last long enough to raise on evidence.

The answer in three lines

  1. Fund the company with software takeover work now (HHI-type), and keep the IP so each engagement becomes a product.

    K01 HHI

    An ongoing development contract of about $25k a month, about 90% likely to sign this week, replacing SDSOL.

    Wilson, 13 Oct 2026

    Used in 3 places

    M08 Services in the model

    HHI-type services are modelled at $300k a year, 50% probability, off by default.

    Step 9

    Used in 2 places

    A15 Vertical beats horizontal

    Illustrative LTV to CAC: about 8x for vertical software, about 0.7x for a public app builder.

    Builder-vs-Vertical-GTM-Report_rev1.md §3

    Used in 3 places

  2. Sell governed AI as an annual service, priced by work items at a guaranteed turnaround, on hardware the customer owns: ours when they need sovereignty, theirs when they already have it.

    A01 Budget certainty is the motive

    No company in the public record has bought hardware to solve an AI cost problem; consumption billing is incompatible with annual public budgets.

    Who Buys the Box §5

    Used in 3 places

    A02 Nobody buys tokens

    No US state or local procurement is denominated in tokens; government buys work items at a turnaround.

    Round 9 bottom line 1 and 4

    Used in 3 places

    A08 At the small end, software is the moat

    Two DGX Sparks (under $10k) handle a mid-size city's document work; Furnace wins on concurrency, sealed custody, department isolation and lights-out operation.

    DGX Spark vs Furnace rev5

    Used in 5 places

    C01 Fixed price is the majority

    Accenture's fixed-price work, including outcome-based deals, is over 65% of bookings (Q4 FY26 call).

    HHI/research/Big-Consulting-AI-Monetization_rev1.md §1

    Used in 3 places

    C12 Hybrid pricing is the equilibrium

    Brynjolfsson and Petropoulos: where attribution is messy, a fixed base plus an outcome bonus wins; productised agents enable subscription and usage pricing.

    Big-Consulting..., §1

    Used in 2 places

  3. Build the rule modules once and deploy them many times. That is what separates a product company from a consultancy.

    A09 SNAP is parametric

    About 28 options across 53 agencies: one engine plus a table. Zoning varies structurally.

    Round 3 Part II

    Used in 3 places

    A10 The Catala precedent

    A four-person team reached a national-agency commitment in about six years.

    Round 3 Part III

    Used in 2 places

    A19 A marketplace is inbound code

    Fetching a rule module isn't exfiltration. Three-party warranty: Allodial warrants the runtime, the author the encoding, the jurisdiction the decision.

    Furnace-Marketplace-Overview_rev1.md

    Used in 2 places

Evidence chips External Folder analysis Model Known

The research

What the consulting research teaches

  1. L1

    Hours are dying; fixed and outcome pricing replace them.

    Accenture now books most of its work at a fixed price , about a quarter of McKinsey's fees ride on outcomes , and clients are already clawing back fees for hours that AI saved .

    C01 Fixed price is the majority

    Accenture's fixed-price work, including outcome-based deals, is over 65% of bookings (Q4 FY26 call).

    HHI/research/Big-Consulting-AI-Monetization_rev1.md §1

    Used in 3 places

    C02 Outcome-based fees

    About 25% of McKinsey's global fees are outcome-based (Nov 2025).

    HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

    Used in 2 places

    C03 Clients claw back AI savings

    KPMG cut its own audit fee from $416k to $357k, about 14%, citing AI.

    HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

    Used in 1 place

    For us: never price by effort. Price by the work item, at a stated turnaround.

    A02 Nobody buys tokens

    No US state or local procurement is denominated in tokens; government buys work items at a turnaround.

    Round 9 bottom line 1 and 4

    Used in 3 places
  2. L2

    The money is in running things, not advising on them.

    Managed services are half of Accenture's revenue .

    C04 Managed services are half

    $37.3B of Accenture's $74.2B FY26 revenue.

    HHI/research/Big-Consulting-AI-Monetization_rev1.md, §2

    Used in 1 place

    For us: the managed appliance and the governed workflows we operate are the recurring core.

    A13 Paid at delivery

    A third-party tax-exempt lease-purchase pays the vendor at delivery; the government must hold title.

    Round 1

    Used in 2 places
  3. L3

    Platforms get licensed, and the price is high.

    Accenture sells its agent platform at $405k a year, with infrastructure extra .

    C05 Platforms get licensed

    Accenture AI Refinery: $405,000 for 12 months on AWS Marketplace, with NVIDIA licences and AWS costs extra.

    HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

    Used in 2 places

    For us: our agent-control layer is licensable too, once it's ready .

    A17 What is proved

    The 8-gate admission rule holds over all 256 states with no axioms; a command needing Gater review can't be admitted directly; the kill-switch quorum is a Lean model.

    01_case_for_the_stack_rev1 §2.2, §3

    Used in 2 places

    A18 What is not yet shipped

    The proofs cover Lean models, not the shipped Haskell; the harness tool-call layer doesn't reach the runtime host at d55aabd4d6.

    01_case_for_the_stack_rev1 §3, §4.2

    Used in 3 places
  4. L4

    Nobody makes money on tokens.

    No big firm discloses a token markup , and no US state or local buyer purchases in tokens .

    C06 No token markup

    No big firm discloses a per-token markup. Falling token costs are framed as a driver of demand.

    HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

    Used in 2 places

    A02 Nobody buys tokens

    No US state or local procurement is denominated in tokens; government buys work items at a turnaround.

    Round 9 bottom line 1 and 4

    Used in 3 places

    For us: tokens are a cost line, never a price line.

  5. L5

    Accountability is what clients actually buy.

    Clients pay for a named party who owns the result and reclaim fees when that party fails . Insurers already back AI warranties for vendors with a track record .

    C08 Partnered projects succeed more

    MIT NANDA: about 95% of gen-AI pilots show no measurable P&L impact; projects bought from or built with partners succeed about 67% of the time, internal builds about a third as often.

    Big-Consulting... §3

    Used in 1 place

    C13 The action gap

    Forrester: value sits in agents that execute in systems of record, not in recommendations.

    Big-Consulting..., §3

    Used in 1 place

    C09 Accountability has teeth

    Deloitte made a partial refund on an AU$440k report containing AI-fabricated citations.

    Big-Consulting..., §2

    Used in 1 place

    C10 Insurers back AI warranties that have a track record

    Munich Re aiSure has backed vendor performance warranties since 2018.

    Accenture-CEO...

    Used in 2 places

    For us: sell scope of validity and service credits now, and priced warranties once we have deployment history .

    A14 Warranty demand ahead of insurance

    No insurer covers a system with no deployment history. Tech E&O and cyber cost about $2k to $12k a year. The GovAI Coalition addendum demands AI warranties.

    Round 6

    Used in 2 places
  6. L6

    Below $300M of revenue, nobody big is selling.

    Accenture's new mid-market unit starts at $300M , local government buys through cooperative contracts from dozens of small vendors , and federal AI seats now go for $1 .

    C07 Nothing below $300M

    Accenture Edge (June 2026) targets companies with $300M to $3B of revenue.

    Accenture-CEO-AI-Monetization-Playbook_rev1.md

    Used in 2 places

    C15 Local government buys from many small vendors

    TXShare awarded 77 AI suppliers on one cooperative contract.

    Accenture-CEO...

    Used in 3 places

    C14 Seats are worth $1 in government

    OpenAI and Anthropic offered federal agencies a year of access for $1 (GSA OneGov, Aug 2025).

    Accenture-CEO...

    Used in 1 place

    For us: the county, the credit union and the research lab are unserved by the big firms, and seats are no longer a revenue line in government.

The constraint

The seed is a bridge to evidence, not a business plan.

  1. 9 months / $2.5M for 18

    On our own model, $2M runs the Proof plan for about 9 months. The right-sized amount is $2.5M for 18 months .

    M01 The seed is short

    $2M runs the Proof plan for about 9 months; $2.5M reaches 18. Proof buys certification ($0.8M), a benchmark and one reference site.

    Allodial-Investment-Options_rev2.md

    Used in 4 places
    Model
  2. about $88k

    Without an implementation fee, every customer costs us about $88k up front .

    M02 The implementation fee is what keeps a customer profitable

    With no fee, a customer loses about $88k up front. Break-even is about $81k to $88k; quote $90k, floor $75k.

    07-Red-Team_rev1.md; Step 9

    Used in 5 places
    Model
  3. about 12 months

    Public buyers take about 12 months from letter of intent to purchase order .

    A20 Public cycles are long

    Government tech purchases average 22 months (Gartner); letter of intent to purchase order is about 12 months.

    02-Channel-Map_rev1.md

    Used in 4 places
    Folder analysis

The biggest swings in unit economics are customer acquisition cost, Emberholt's true cost and contractor rates .

M07 Where unit economics swing

Per customer: CAC $90k, Emberholt COGS $60k, contractor rates $58k.

07-Red-Team_rev1.md §3

Used in 2 places

So the $2M plan has to:

  1. 1

    earn revenue before any hardware ships;

  2. 2

    avoid carrying inventory;

  3. 3

    charge for implementation from the first deal;

  4. 4

    reach 8 to 10 signed Tier A letters before the next raise.

Strategies

Eight strategies, tested

Evidence lens

Choose a type to see which strategies rest on it.

Eight strategies by six criteria. Scores run 1 to 5, where 5 is best for Allodial.
StrategyCash in 12 monthsCapital requiredEvidence strengthDefensibilitySovereignty fitLoad on four people
S1Appliance first1 of 51 of 53 of 53 of 55 of 52 of 5
S2Software first: Allodial Control on the customer's hardware3 of 55 of 52 of 54 of 54 of 53 of 5
S3Services-funded vertical products5 of 55 of 54 of 53 of 52 of 52 of 5
S4Rule-module platform2 of 54 of 54 of 55 of 55 of 53 of 5
S5Outcome and warranty pricing2 of 54 of 53 of 54 of 54 of 54 of 5
S6Shared service through counties and councils of governments2 of 53 of 53 of 53 of 55 of 53 of 5
Rejected
S7Sell idle capacity into token marketplacesNot scoredNot scoredNot scoredNot scoredNot scoredNot scored
S8Sell boxes directly to small businessesNot scoredNot scoredNot scoredNot scoredNot scoredNot scored

Raised rows: the recommended combination.

S1

Appliance first

Sell Emberholt boxes to counties and cities, with a managed annual service.

How it makes money

Hardware at about 40% gross margin, an implementation fee, and annual service at $24k per customer plus $7k per box (proposed).

M03 Emberholt's price

$125,056 street, $75,034 COGS.

model rev9 and rev10

Used in 1 place

M06 The flat service fee loses money

$9k per box loses about $16.6k a year on a 3-box customer; proposed $24k plus $7k per box.

Step 4

Used in 2 places

M02 The implementation fee is what keeps a customer profitable

With no fee, a customer loses about $88k up front. Break-even is about $81k to $88k; quote $90k, floor $75k.

07-Red-Team_rev1.md; Step 9

Used in 5 places
Who buys

County human services in county-run SNAP states; cities over 50k in permit shot-clock states.

A11 The SNAP clock

The federal admin match fell from 50% to 25% on 1 Oct 2026 (about $2.7B); error-rate cost sharing starts 1 Oct 2027.

Round 2; Furnace-Automation-Target-List_rev1.md

Used in 4 places

A12 Permit shot clocks

Indiana: 7 business days, then forced issuance. Texas: 45 days, then no fees. Florida: fee cut 10% per business day late.

Furnace-Automation-Target-List_rev1.md

Used in 4 places
What $2M buys

Not enough on its own. Custom certification takes about $0.8M of the seed, and the model shows the Proof plan running out in about 9 months.

M01 The seed is short

$2M runs the Proof plan for about 9 months; $2.5M reaches 18. Proof buys certification ($0.8M), a benchmark and one reference site.

Allodial-Investment-Options_rev2.md

Used in 4 places
Evidence for

A01 Budget certainty is the motive

No company in the public record has bought hardware to solve an AI cost problem; consumption billing is incompatible with annual public budgets.

Who Buys the Box §5

Used in 3 places

A07 One card does the municipal work

Formalizable workloads fit one 96 GB card; reference model: dense 70B at FP8, gpt-oss-120b as alternate (decided 6 Oct).

Model-Tier-Hardware-Matrix_rev2.md

Used in 2 places

A11 The SNAP clock

The federal admin match fell from 50% to 25% on 1 Oct 2026 (about $2.7B); error-rate cost sharing starts 1 Oct 2027.

Round 2; Furnace-Automation-Target-List_rev1.md

Used in 4 places

A12 Permit shot clocks

Indiana: 7 business days, then forced issuance. Texas: 45 days, then no fees. Florida: fee cut 10% per business day late.

Furnace-Automation-Target-List_rev1.md

Used in 4 places

A13 Paid at delivery

A third-party tax-exempt lease-purchase pays the vendor at delivery; the government must hold title.

Round 1

Used in 2 places

A22 There is no AI budget line

Only 25% of state CIOs have dedicated GenAI funding; the first box is paid from an existing capital or modernisation line.

Round 9

Used in 1 place
Evidence against

A03 Most cities are below cloud break-even

Break-even is 170k to 1.5M tasks a month; a 500k-resident city generates about 76k.

Round 9 bottom line 7

Used in 3 places

A08 At the small end, software is the moat

Two DGX Sparks (under $10k) handle a mid-size city's document work; Furnace wins on concurrency, sealed custody, department isolation and lights-out operation.

DGX Spark vs Furnace rev5

Used in 5 places

A20 Public cycles are long

Government tech purchases average 22 months (Gartner); letter of intent to purchase order is about 12 months.

02-Channel-Map_rev1.md

Used in 4 places

M01 The seed is short

$2M runs the Proof plan for about 9 months; $2.5M reaches 18. Proof buys certification ($0.8M), a benchmark and one reference site.

Allodial-Investment-Options_rev2.md

Used in 4 places

Evidence: 0 external · 9 folder · 4 model · 0 known

Scores
Cash in 12 months
1
Public purchase orders arrive about 12 months after a letter.
Capital required
1
Certification plus working capital.
Evidence strength
3
Strong motive, no signed buyer yet.
Defensibility
3
The hardware alone is beatable by two DGX Sparks.
Sovereignty fit
5
This is the sovereignty product.
Load on four people
2
Install, certification and long sales cycles at once.

Keep it as the destination, not the starting line. Fund the custom box with the next round.

S2

Software first: Allodial Control on the customer's hardware

License the agent-control layer and governed workflows to run on hardware the customer already has, or on OEM hardware a reseller supplies.

How it makes money

Implementation fee plus annual license and service. No hardware margin, and no inventory.

M02 The implementation fee is what keeps a customer profitable

With no fee, a customer loses about $88k up front. Break-even is about $81k to $88k; quote $90k, floor $75k.

07-Red-Team_rev1.md; Step 9

Used in 5 places

C05 Platforms get licensed

Accenture AI Refinery: $405,000 for 12 months on AWS Marketplace, with NVIDIA licences and AWS costs extra.

HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

Used in 2 places
Who buys

Research computing groups, small towns through their counties, and credit unions with existing servers.

What $2M buys

Readiness work on the control layer, two or three sites, and a published benchmark.

Evidence for

A08 At the small end, software is the moat

Two DGX Sparks (under $10k) handle a mid-size city's document work; Furnace wins on concurrency, sealed custody, department isolation and lights-out operation.

DGX Spark vs Furnace rev5

Used in 5 places

C05 Platforms get licensed

Accenture AI Refinery: $405,000 for 12 months on AWS Marketplace, with NVIDIA licences and AWS costs extra.

HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

Used in 2 places

C07 Nothing below $300M

Accenture Edge (June 2026) targets companies with $300M to $3B of revenue.

Accenture-CEO-AI-Monetization-Playbook_rev1.md

Used in 2 places

A17 What is proved

The 8-gate admission rule holds over all 256 states with no axioms; a command needing Gater review can't be admitted directly; the kill-switch quorum is a Lean model.

01_case_for_the_stack_rev1 §2.2, §3

Used in 2 places
Evidence against

A18 What is not yet shipped

The proofs cover Lean models, not the shipped Haskell; the harness tool-call layer doesn't reach the runtime host at d55aabd4d6.

01_case_for_the_stack_rev1 §3, §4.2

Used in 3 places

A06 Small businesses can't carry a box

The capex is about ten times a small firm's annual AI spend; pooled demand hits the throughput ceiling first.

UK-SMB-Channel-Verdict_rev1.md

Used in 4 places

Evidence: 2 external · 4 folder · 1 model · 0 known

Scores
Cash in 12 months
3
Private buyers decide in months.
Capital required
5
No inventory.
Evidence strength
2
The control layer is partly Built, not yet a shipped product path.
Defensibility
4
Machine-checked admission rules are rare in this market.
Sovereignty fit
4
On-premises, but on hardware we don't seal.
Load on four people
3

The entry tier, and our answer to the DGX Spark. Until the tool-call path reaches the runtime host, sell it inside a managed service, not as a standalone license.

S3

Services-funded vertical products

Take over failing software for mid-size operators, fix it with receipts, rebuild it, and keep the IP so each rebuild becomes a product.

How it makes money

A monthly retainer now (HHI is about $25k a month), then vertical subscriptions.

K01 HHI

An ongoing development contract of about $25k a month, about 90% likely to sign this week, replacing SDSOL.

Wilson, 13 Oct 2026

Used in 3 places

A15 Vertical beats horizontal

Illustrative LTV to CAC: about 8x for vertical software, about 0.7x for a public app builder.

Builder-vs-Vertical-GTM-Report_rev1.md §3

Used in 3 places

A16 The Mews opening

Mews is worth about $167k per property; customers cite five complaints. Integrations, PCI compliance and 24/7 support are the costs the builder doesn't remove.

Builder-vs-Vertical-GTM-Report_rev1.md §5; mews-replacement-spec/

Used in 2 places
Who buys

Mid-size operators stuck with a bad vendor (wellness, hospitality, clinics, nonprofits), and the hotel design partner leaving Mews.

K02 Hotel design partner

A current Mews customer prepared to pay for a replacement.

Builder-vs-Vertical-GTM-Report_rev1.md §5

Used in 2 places
What $2M buys

Nothing extra. It funds itself in cash and shows the agent fleet working in production.

Evidence for

K01 HHI

An ongoing development contract of about $25k a month, about 90% likely to sign this week, replacing SDSOL.

Wilson, 13 Oct 2026

Used in 3 places

K02 Hotel design partner

A current Mews customer prepared to pay for a replacement.

Builder-vs-Vertical-GTM-Report_rev1.md §5

Used in 2 places

A15 Vertical beats horizontal

Illustrative LTV to CAC: about 8x for vertical software, about 0.7x for a public app builder.

Builder-vs-Vertical-GTM-Report_rev1.md §3

Used in 3 places

A16 The Mews opening

Mews is worth about $167k per property; customers cite five complaints. Integrations, PCI compliance and 24/7 support are the costs the builder doesn't remove.

Builder-vs-Vertical-GTM-Report_rev1.md §5; mews-replacement-spec/

Used in 2 places

M08 Services in the model

HHI-type services are modelled at $300k a year, 50% probability, off by default.

Step 9

Used in 2 places
Evidence against

: integrations, PCI compliance and night support are not cheap even when the code is.

A16 The Mews opening

Mews is worth about $167k per property; customers cite five complaints. Integrations, PCI compliance and 24/7 support are the costs the builder doesn't remove.

Builder-vs-Vertical-GTM-Report_rev1.md §5; mews-replacement-spec/

Used in 2 places

Evidence: 0 external · 2 folder · 1 model · 2 known

Scores
Cash in 12 months
5
Capital required
5
Evidence strength
4
One signed client is likely this week, plus a paying design partner.
Defensibility
3
Strong once a product exists, weak as pure services.
Sovereignty fit
2
Mostly not about sovereignty.
Load on four people
2
Services eat founder hours.

Do it now, capped below half of revenue after month 12, and only on terms that leave Allodial the IP.

S4

Rule-module platform

Encode rule sets that are federal, stable and shared (SNAP eligibility first, then permit screening against model codes), and deploy each one many times.

How it makes money

Workflow bands inside the annual service at first; later a marketplace take rate on modules other jurisdictions author.

A09 SNAP is parametric

About 28 options across 53 agencies: one engine plus a table. Zoning varies structurally.

Round 3 Part II

Used in 3 places

A19 A marketplace is inbound code

Fetching a rule module isn't exfiltration. Three-party warranty: Allodial warrants the runtime, the author the encoding, the jurisdiction the decision.

Furnace-Marketplace-Overview_rev1.md

Used in 2 places
Who buys

County and state human services; building departments.

What $2M buys

The SNAP engine and one state's parameter table; a permit-screening subset for one code edition.

Evidence for

A09 SNAP is parametric

About 28 options across 53 agencies: one engine plus a table. Zoning varies structurally.

Round 3 Part II

Used in 3 places

A10 The Catala precedent

A four-person team reached a national-agency commitment in about six years.

Round 3 Part III

Used in 2 places

A11 The SNAP clock

The federal admin match fell from 50% to 25% on 1 Oct 2026 (about $2.7B); error-rate cost sharing starts 1 Oct 2027.

Round 2; Furnace-Automation-Target-List_rev1.md

Used in 4 places

A12 Permit shot clocks

Indiana: 7 business days, then forced issuance. Texas: 45 days, then no fees. Florida: fee cut 10% per business day late.

Furnace-Automation-Target-List_rev1.md

Used in 4 places

A19 A marketplace is inbound code

Fetching a rule module isn't exfiltration. Three-party warranty: Allodial warrants the runtime, the author the encoding, the jurisdiction the decision.

Furnace-Marketplace-Overview_rev1.md

Used in 2 places
Evidence against

: Catala took six years to reach a national agency. . : automated decisions against residents are where public AI fails, so every ambiguous case has to halt to a caseworker.

A10 The Catala precedent

A four-person team reached a national-agency commitment in about six years.

Round 3 Part III

Used in 2 places

A20 Public cycles are long

Government tech purchases average 22 months (Gartner); letter of intent to purchase order is about 12 months.

02-Channel-Map_rev1.md

Used in 4 places

C16 Never automate decisions against residents

Australia's Robodebt ended in a A$475M settlement.

Accenture-CEO...

Used in 1 place

Evidence: 1 external · 6 folder · 0 model · 0 known

Scores
Cash in 12 months
2
Capital required
4
Mostly engineering time.
Evidence strength
4
SNAP has been reimplemented independently five or more times.
Defensibility
5
Encodings plus proofs plus deployment history compound.
Sovereignty fit
5
Inbound code, not outbound data.
Load on four people
3

The compounding asset. Start the SNAP engine now, because the 2027 cost-sharing clock is what makes buyers move.

A11 The SNAP clock

The federal admin match fell from 50% to 25% on 1 Oct 2026 (about $2.7B); error-rate cost sharing starts 1 Oct 2027.

Round 2; Furnace-Automation-Target-List_rev1.md

Used in 4 places
S5

Outcome and warranty pricing

Price by work item at a guaranteed turnaround, backed by service credits now and a priced warranty later.

How it makes money

Turns S1, S2 and S4 into an outcome contract with a fixed floor; a warranty premium once an insurer will cover it.

C01 Fixed price is the majority

Accenture's fixed-price work, including outcome-based deals, is over 65% of bookings (Q4 FY26 call).

HHI/research/Big-Consulting-AI-Monetization_rev1.md §1

Used in 3 places

C02 Outcome-based fees

About 25% of McKinsey's global fees are outcome-based (Nov 2025).

HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

Used in 2 places

C12 Hybrid pricing is the equilibrium

Brynjolfsson and Petropoulos: where attribution is messy, a fixed base plus an outcome bonus wins; productised agents enable subscription and usage pricing.

Big-Consulting..., §1

Used in 2 places

A14 Warranty demand ahead of insurance

No insurer covers a system with no deployment history. Tech E&O and cyber cost about $2k to $12k a year. The GovAI Coalition addendum demands AI warranties.

Round 6

Used in 2 places
Who buys

Any buyer under a statutory clock: permit shot clocks, SNAP error rates.

What $2M buys

Contract language, tech E&O and cyber with no AI exclusion, and a turnaround measurement method.

Evidence for

C01 Fixed price is the majority

Accenture's fixed-price work, including outcome-based deals, is over 65% of bookings (Q4 FY26 call).

HHI/research/Big-Consulting-AI-Monetization_rev1.md §1

Used in 3 places

C02 Outcome-based fees

About 25% of McKinsey's global fees are outcome-based (Nov 2025).

HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

Used in 2 places

C10 Insurers back AI warranties that have a track record

Munich Re aiSure has backed vendor performance warranties since 2018.

Accenture-CEO...

Used in 2 places

C12 Hybrid pricing is the equilibrium

Brynjolfsson and Petropoulos: where attribution is messy, a fixed base plus an outcome bonus wins; productised agents enable subscription and usage pricing.

Big-Consulting..., §1

Used in 2 places

A14 Warranty demand ahead of insurance

No insurer covers a system with no deployment history. Tech E&O and cyber cost about $2k to $12k a year. The GovAI Coalition addendum demands AI warranties.

Round 6

Used in 2 places

A12 Permit shot clocks

Indiana: 7 business days, then forced issuance. Texas: 45 days, then no fees. Florida: fee cut 10% per business day late.

Furnace-Automation-Target-List_rev1.md

Used in 4 places
Evidence against

: no insurer covers a system with no deployment history. : attribution has to be clean.

A14 Warranty demand ahead of insurance

No insurer covers a system with no deployment history. Tech E&O and cyber cost about $2k to $12k a year. The GovAI Coalition addendum demands AI warranties.

Round 6

Used in 2 places

C12 Hybrid pricing is the equilibrium

Brynjolfsson and Petropoulos: where attribution is messy, a fixed base plus an outcome bonus wins; productised agents enable subscription and usage pricing.

Big-Consulting..., §1

Used in 2 places

Evidence: 4 external · 2 folder · 0 model · 0 known

Scores
Cash in 12 months
2
It's a pricing form, not a revenue line.
Capital required
4
Evidence strength
3
Defensibility
4
Few vendors can sign the GovAI warranty addendum.
Sovereignty fit
4
Load on four people
4

Adopt it as the pricing form across everything. Keep the fixed floor. Don't promise a priced warranty until we have deployment history.

S6

Shared service through counties and councils of governments

One county or regional council hosts a box and serves its towns under an interlocal agreement.

How it makes money

Same as S1, but one sale reaches many towns.

A03 Most cities are below cloud break-even

Break-even is 170k to 1.5M tasks a month; a 500k-resident city generates about 76k.

Round 9 bottom line 7

Used in 3 places

C15 Local government buys from many small vendors

TXShare awarded 77 AI suppliers on one cooperative contract.

Accenture-CEO...

Used in 3 places
Who buys

Counties over 100k and councils of governments.

What $2M buys

One reference deal, bought through a cooperative-contract reseller.

Evidence for

A03 Most cities are below cloud break-even

Break-even is 170k to 1.5M tasks a month; a 500k-resident city generates about 76k.

Round 9 bottom line 7

Used in 3 places

A07 One card does the municipal work

Formalizable workloads fit one 96 GB card; reference model: dense 70B at FP8, gpt-oss-120b as alternate (decided 6 Oct).

Model-Tier-Hardware-Matrix_rev2.md

Used in 2 places

C15 Local government buys from many small vendors

TXShare awarded 77 AI suppliers on one cooperative contract.

Accenture-CEO...

Used in 3 places
Evidence against

A20 Public cycles are long

Government tech purchases average 22 months (Gartner); letter of intent to purchase order is about 12 months.

02-Channel-Map_rev1.md

Used in 4 places

Evidence: 1 external · 3 folder · 0 model · 0 known

Scores
Cash in 12 months
2
Capital required
3
Evidence strength
3
Defensibility
3
Sovereignty fit
5
Load on four people
3

Our route to every town under 50k. Use it for the public reference site.

Rejected

S7

Sell idle capacity into token marketplaces (rejected)

The box earns revenue at night by serving outside workloads.

Killed by

A04 Idle resale is negligible

About $918 a year on a 2-GPU box at realistic prices, against an annual obligation of about $34k.

The Payback Model; Furnace-Excess-Compute-Recommendation_rev1.md

Used in 1 place

A05 Resale breaks the sovereignty claim

Third-party workloads on the records box; extra legal review for every public buyer.

Furnace-Excess-Compute-Recommendation_rev1.md

Used in 1 place

C06 No token markup

No big firm discloses a per-token markup. Falling token costs are framed as a driver of demand.

HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

Used in 2 places

Evidence: 1 external · 2 folder · 0 model · 0 known

Why

It nets about $918 a year on a 2-GPU box at realistic prices, against an annual obligation of about $34k. It breaks the sovereignty claim. And it adds a legal review to every public deal.

S8

Sell boxes directly to small businesses (rejected)

A mini data center for the 25-person firm.

Killed by

A06 Small businesses can't carry a box

The capex is about ten times a small firm's annual AI spend; pooled demand hits the throughput ceiling first.

UK-SMB-Channel-Verdict_rev1.md

Used in 4 places

A08 At the small end, software is the moat

Two DGX Sparks (under $10k) handle a mid-size city's document work; Furnace wins on concurrency, sealed custody, department isolation and lights-out operation.

DGX Spark vs Furnace rev5

Used in 5 places

Evidence: 0 external · 2 folder · 0 model · 0 known

Why

The capex is about ten times a small firm's annual AI spend. Pooled small-firm demand hits the box's throughput ceiling before it breaks even. Two DGX Sparks cost under $10k. Small businesses come in through S3 instead.

Parked, not rejected: grid demand-response aggregation and EU heat-reuse compliance (need $C scale); selling the software builder to the public against Lovable (Builder-vs-Vertical-GTM-Report_rev1.md: about 0.7x LTV to CAC); Crucible (a separate entity by design).

05 · The combination we recommend

S3 pays the bills, S2 opens the door, S4 compounds, S5 sets the price, S1 and S6 are where it lands.

  1. Now to December 2026

    S3

    S3: HHI signed; hotel build agreement with IP retained. Start reseller onboarding, recruit reps, run the pre-raise test, bind tech E&O and cyber.

    Gate

    Two services contracts; test data on conversion

  2. Months 1 to 6

    S2S4

    S2 at the first research-computing or credit-union site. S4: SNAP engine and one state table. Benchmark published.

    Gate

    One site live; 4+ Tier A letters

  3. Months 6 to 12

    S6S5

    S6 county reference through a cooperative contract. S5 pricing on every proposal.

    Gate

    2 to 3 reference sites; 8 to 10 Tier A letters; implementation fee accepted twice

  4. Months 12 to 15

    S1

    Raise $B (about $7.5M in the model) on letters and installs. It funds S1's custom box and certification.

    M01 The seed is short

    $2M runs the Proof plan for about 9 months; $2.5M reaches 18. Proof buys certification ($0.8M), a benchmark and one reference site.

    Allodial-Investment-Options_rev2.md

    Used in 4 places

    M05 What a sprint buys

    $500k over 90 days: P50 of 10 Tier A letters (P25 of 8).

    03-Campaign-Portfolio_rev1.md

    Used in 1 place
    Gate

    Close

Customers

Customers, and how we reach them

Ideal versus likely

SegmentIdealLikely in 18 monthsStrategyEvidence
County human services, county-run SNAP statesYesOne pilot letterS4, S1

A09 SNAP is parametric

About 28 options across 53 agencies: one engine plus a table. Zoning varies structurally.

Round 3 Part II

Used in 3 places

A11 The SNAP clock

The federal admin match fell from 50% to 25% on 1 Oct 2026 (about $2.7B); error-rate cost sharing starts 1 Oct 2027.

Round 2; Furnace-Automation-Target-List_rev1.md

Used in 4 places
Cities over 50k in shot-clock statesYesTier B lettersS4, S5

A12 Permit shot clocks

Indiana: 7 business days, then forced issuance. Texas: 45 days, then no fees. Florida: fee cut 10% per business day late.

Furnace-Automation-Target-List_rev1.md

Used in 4 places
County or council of governments as hostYesOne reference dealS6

A03 Most cities are below cloud break-even

Break-even is 170k to 1.5M tasks a month; a 500k-resident city generates about 76k.

Round 9 bottom line 7

Used in 3 places

C15 Local government buys from many small vendors

TXShare awarded 77 AI suppliers on one cooperative contract.

Accenture-CEO...

Used in 3 places
R1 and R2 research computingYesFirst installS2

M04 Founder outbound converts best

Weighted letters per $100k: research computing 13.7, credit unions 8.0, county HHS 6.8.

02-Channel-Map_rev1.md

Used in 2 places
Credit unions over $500MYesOne installS2, S1

M04 Founder outbound converts best

Weighted letters per $100k: research computing 13.7, credit unions 8.0, county HHS 6.8.

02-Channel-Map_rev1.md

Used in 2 places
Mid-size operators with a failing vendorNoTwo to three contractsS3

K01 HHI

An ongoing development contract of about $25k a month, about 90% likely to sign this week, replacing SDSOL.

Wilson, 13 Oct 2026

Used in 3 places
Hotel design partnerNoFunded buildS3

K02 Hotel design partner

A current Mews customer prepared to pay for a replacement.

Builder-vs-Vertical-GTM-Report_rev1.md §5

Used in 2 places

A16 The Mews opening

Mews is worth about $167k per property; customers cite five complaints. Integrations, PCI compliance and 24/7 support are the costs the builder doesn't remove.

Builder-vs-Vertical-GTM-Report_rev1.md §5; mews-replacement-spec/

Used in 2 places
Health systemsYes, laterTier B lettersS1
Small businesses buying a boxNoNoRejected (S8)

A06 Small businesses can't carry a box

The capex is about ten times a small firm's annual AI spend; pooled demand hits the throughput ceiling first.

UK-SMB-Channel-Verdict_rev1.md

Used in 4 places

Channels, ranked by signed letters per dollar

  1. 1

    Founder outbound: research computing 13.7, credit unions 8.0, county HHS 6.8 weighted letters per $100k.

    M04 Founder outbound converts best

    Weighted letters per $100k: research computing 13.7, credit unions 8.0, county HHS 6.8.

    02-Channel-Map_rev1.md

    Used in 2 places
  2. 2

    Independent reps (former HHS directors and city CIOs), recruited in Q4 2026.

  3. 3

    A cooperative-contract reseller, so a county can sign without an RFP. Start now; onboarding takes 1 to 3 months.

  4. 4

    Three in-window events: NACo Legislative (19 to 23 Feb 2027), America's Credit Unions GAC (28 Feb to 4 Mar), NLC Congressional City Conference (15 to 17 Mar).

  5. 5

    LinkedIn matched audiences to named accounts.

  6. 6

    Referrals from services clients.

  7. 7

    HUBZone status, if the San Juan address and staff residency qualify: a 10% price preference on federal work.

    A21 HUBZone in Puerto Rico

    82.1% of Puerto Rico's census tracts qualify; a 10% price preference applies if at least 35% of employees live in a HUBZone.

    Opportunity-Register_rev1.md §5

    Used in 1 place

The opening line for each buyer

Order: budget certainty, then sovereignty, then local benefit. Never cost per token.

A01 Budget certainty is the motive

No company in the public record has bought hardware to solve an AI cost problem; consumption billing is incompatible with annual public budgets.

Who Buys the Box §5

Used in 3 places
County HHS

"Each determination cites the rule and the record field. Anything ambiguous goes to a caseworker. You buy the machine once from your capital budget and pay a fixed annual fee. There's no usage bill."

City permitting

"We screen submissions against your adopted code, cite every clause, and return anything out of scope to your reviewers. Screening, not approval."

Research computing

"Open-weight models on your data, under your governance, on hardware your team can run, with a control layer that keeps long agent runs within their authority."

Credit union

"Member data stays in the building, and every AI action leaves a signed receipt your examiner can check."

Software takeover

"We fix what your vendor left broken, in weeks, with receipts, then rebuild it without losing a record."

Decisions

Decisions this memo asks for

Decisions this memo asks for

  1. 1

    First sites: OEM-qualified hardware first and the custom Emberholt after the next round, or the custom box now? Recommended: OEM first.

    M01 The seed is short

    $2M runs the Proof plan for about 9 months; $2.5M reaches 18. Proof buys certification ($0.8M), a benchmark and one reference site.

    Allodial-Investment-Options_rev2.md

    Used in 4 places

    A08 At the small end, software is the moat

    Two DGX Sparks (under $10k) handle a mid-size city's document work; Furnace wins on concurrency, sealed custody, department isolation and lights-out operation.

    DGX Spark vs Furnace rev5

    Used in 5 places
  2. 2

    Small-business hardware: take it out of the plan? Recommended: yes.

    A06 Small businesses can't carry a box

    The capex is about ten times a small firm's annual AI spend; pooled demand hits the throughput ceiling first.

    UK-SMB-Channel-Verdict_rev1.md

    Used in 4 places
  3. 3

    Pricing: confirm the $24k + $7k service proposal and the $90k implementation quote.

    M02 The implementation fee is what keeps a customer profitable

    With no fee, a customer loses about $88k up front. Break-even is about $81k to $88k; quote $90k, floor $75k.

    07-Red-Team_rev1.md; Step 9

    Used in 5 places

    M06 The flat service fee loses money

    $9k per box loses about $16.6k a year on a 3-box customer; proposed $24k plus $7k per box.

    Step 4

    Used in 2 places
  4. 4

    Services: cap services below half of revenue after month 12, and require IP retention in every build agreement.

    A15 Vertical beats horizontal

    Illustrative LTV to CAC: about 8x for vertical software, about 0.7x for a public app builder.

    Builder-vs-Vertical-GTM-Report_rev1.md §3

    Used in 3 places

What would change our mind

  1. 1

    If buyers refuse the implementation fee, we need lower CAC or a smaller team.

    M02 The implementation fee is what keeps a customer profitable

    With no fee, a customer loses about $88k up front. Break-even is about $81k to $88k; quote $90k, floor $75k.

    07-Red-Team_rev1.md; Step 9

    Used in 5 places

    M07 Where unit economics swing

    Per customer: CAC $90k, Emberholt COGS $60k, contractor rates $58k.

    07-Red-Team_rev1.md §3

    Used in 2 places
  2. 2

    If founder outbound converts at half the estimate, we shift spend to reps and the reseller.

    M04 Founder outbound converts best

    Weighted letters per $100k: research computing 13.7, credit unions 8.0, county HHS 6.8.

    02-Channel-Map_rev1.md

    Used in 2 places
  3. 3

    If a statute moves (SNAP cost sharing delayed), S4's public timing slips and S2 and S3 carry longer.

    A11 The SNAP clock

    The federal admin match fell from 50% to 25% on 1 Oct 2026 (about $2.7B); error-rate cost sharing starts 1 Oct 2027.

    Round 2; Furnace-Automation-Target-List_rev1.md

    Used in 4 places
  4. 4

    If the control layer's tool-call path lands in the runtime host, S2 can be sold as a standalone license sooner.

    A18 What is not yet shipped

    The proofs cover Lean models, not the shipped Haskell; the harness tool-call layer doesn't reach the runtime host at d55aabd4d6.

    01_case_for_the_stack_rev1 §3, §4.2

    Used in 3 places
  5. 5

    If the labs' own delivery arms move down-market to counties and credit unions, S2 and S4 have to be differentiated on proofs and custody, not on integration work.

    C11 Labs are moving into delivery

    OpenAI's Deployment Company raised $4B (May 2026).

    Big-Consulting... §3

    Used in 1 place

Evidence

Findings register

Evidence chips External Folder analysis Model Known

External16

  1. C01

    Fixed price is the majority

    Accenture's fixed-price work, including outcome-based deals, is over 65% of bookings (Q4 FY26 call).

    HHI/research/Big-Consulting-AI-Monetization_rev1.md §1

    Used in Answer 2, L1, S5

  2. C02

    Outcome-based fees

    About 25% of McKinsey's global fees are outcome-based (Nov 2025).

    HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

    Used in L1, S5

  3. C03

    Clients claw back AI savings

    KPMG cut its own audit fee from $416k to $357k, about 14%, citing AI.

    HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

    Used in L1

  4. C04

    Managed services are half

    $37.3B of Accenture's $74.2B FY26 revenue.

    HHI/research/Big-Consulting-AI-Monetization_rev1.md, §2

    Used in L2

  5. C05

    Platforms get licensed

    Accenture AI Refinery: $405,000 for 12 months on AWS Marketplace, with NVIDIA licences and AWS costs extra.

    HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

    Used in L3, S2

  6. C06

    No token markup

    No big firm discloses a per-token markup. Falling token costs are framed as a driver of demand.

    HHI/research/Big-Consulting-AI-Monetization_rev1.md, §1

    Used in L4, S7

  7. C07

    Nothing below $300M

    Accenture Edge (June 2026) targets companies with $300M to $3B of revenue.

    Accenture-CEO-AI-Monetization-Playbook_rev1.md

    Used in L6, S2

  8. C08

    Partnered projects succeed more

    MIT NANDA: about 95% of gen-AI pilots show no measurable P&L impact; projects bought from or built with partners succeed about 67% of the time, internal builds about a third as often.

    Big-Consulting... §3

    Used in L5

  9. C09

    Accountability has teeth

    Deloitte made a partial refund on an AU$440k report containing AI-fabricated citations.

    Big-Consulting..., §2

    Used in L5

  10. C10

    Insurers back AI warranties that have a track record

    Munich Re aiSure has backed vendor performance warranties since 2018.

    Accenture-CEO...

    Used in L5, S5

  11. C11

    Labs are moving into delivery

    OpenAI's Deployment Company raised $4B (May 2026).

    Big-Consulting... §3

    Used in Part 9 · 5

  12. C12

    Hybrid pricing is the equilibrium

    Brynjolfsson and Petropoulos: where attribution is messy, a fixed base plus an outcome bonus wins; productised agents enable subscription and usage pricing.

    Big-Consulting..., §1

    Used in Answer 2, S5

  13. C13

    The action gap

    Forrester: value sits in agents that execute in systems of record, not in recommendations.

    Big-Consulting..., §3

    Used in L5

  14. C14

    Seats are worth $1 in government

    OpenAI and Anthropic offered federal agencies a year of access for $1 (GSA OneGov, Aug 2025).

    Accenture-CEO...

    Used in L6

  15. C15

    Local government buys from many small vendors

    TXShare awarded 77 AI suppliers on one cooperative contract.

    Accenture-CEO...

    Used in L6, S6, Part 6

  16. C16

    Never automate decisions against residents

    Australia's Robodebt ended in a A$475M settlement.

    Accenture-CEO...

    Used in S4

Folder analysis22

  1. A01

    Budget certainty is the motive

    No company in the public record has bought hardware to solve an AI cost problem; consumption billing is incompatible with annual public budgets.

    Who Buys the Box §5

    Used in Answer 2, S1, Part 6

  2. A02

    Nobody buys tokens

    No US state or local procurement is denominated in tokens; government buys work items at a turnaround.

    Round 9 bottom line 1 and 4

    Used in Answer 2, L1, L4

  3. A03

    Most cities are below cloud break-even

    Break-even is 170k to 1.5M tasks a month; a 500k-resident city generates about 76k.

    Round 9 bottom line 7

    Used in S1, S6, Part 6

  4. A04

    Idle resale is negligible

    About $918 a year on a 2-GPU box at realistic prices, against an annual obligation of about $34k.

    The Payback Model; Furnace-Excess-Compute-Recommendation_rev1.md

    Used in S7

  5. A05

    Resale breaks the sovereignty claim

    Third-party workloads on the records box; extra legal review for every public buyer.

    Furnace-Excess-Compute-Recommendation_rev1.md

    Used in S7

  6. A06

    Small businesses can't carry a box

    The capex is about ten times a small firm's annual AI spend; pooled demand hits the throughput ceiling first.

    UK-SMB-Channel-Verdict_rev1.md

    Used in S2, S8, Part 6, Part 8 · 2

  7. A07

    One card does the municipal work

    Formalizable workloads fit one 96 GB card; reference model: dense 70B at FP8, gpt-oss-120b as alternate (decided 6 Oct).

    Model-Tier-Hardware-Matrix_rev2.md

    Used in S1, S6

  8. A08

    At the small end, software is the moat

    Two DGX Sparks (under $10k) handle a mid-size city's document work; Furnace wins on concurrency, sealed custody, department isolation and lights-out operation.

    DGX Spark vs Furnace rev5

    Used in Answer 2, S1, S2, S8, Part 8 · 1

  9. A09

    SNAP is parametric

    About 28 options across 53 agencies: one engine plus a table. Zoning varies structurally.

    Round 3 Part II

    Used in Answer 3, S4, Part 6

  10. A10

    The Catala precedent

    A four-person team reached a national-agency commitment in about six years.

    Round 3 Part III

    Used in Answer 3, S4

  11. A11

    The SNAP clock

    The federal admin match fell from 50% to 25% on 1 Oct 2026 (about $2.7B); error-rate cost sharing starts 1 Oct 2027.

    Round 2; Furnace-Automation-Target-List_rev1.md

    Used in S1, S4, Part 6, Part 9 · 3

  12. A12

    Permit shot clocks

    Indiana: 7 business days, then forced issuance. Texas: 45 days, then no fees. Florida: fee cut 10% per business day late.

    Furnace-Automation-Target-List_rev1.md

    Used in S1, S4, S5, Part 6

  13. A13

    Paid at delivery

    A third-party tax-exempt lease-purchase pays the vendor at delivery; the government must hold title.

    Round 1

    Used in L2, S1

  14. A14

    Warranty demand ahead of insurance

    No insurer covers a system with no deployment history. Tech E&O and cyber cost about $2k to $12k a year. The GovAI Coalition addendum demands AI warranties.

    Round 6

    Used in L5, S5

  15. A15

    Vertical beats horizontal

    Illustrative LTV to CAC: about 8x for vertical software, about 0.7x for a public app builder.

    Builder-vs-Vertical-GTM-Report_rev1.md §3

    Used in Answer 1, S3, Part 8 · 4

  16. A16

    The Mews opening

    Mews is worth about $167k per property; customers cite five complaints. Integrations, PCI compliance and 24/7 support are the costs the builder doesn't remove.

    Builder-vs-Vertical-GTM-Report_rev1.md §5; mews-replacement-spec/

    Used in S3, Part 6

  17. A17

    What is proved

    The 8-gate admission rule holds over all 256 states with no axioms; a command needing Gater review can't be admitted directly; the kill-switch quorum is a Lean model.

    01_case_for_the_stack_rev1 §2.2, §3

    Used in L3, S2

  18. A18

    What is not yet shipped

    The proofs cover Lean models, not the shipped Haskell; the harness tool-call layer doesn't reach the runtime host at d55aabd4d6.

    01_case_for_the_stack_rev1 §3, §4.2

    Used in L3, S2, Part 9 · 4

  19. A19

    A marketplace is inbound code

    Fetching a rule module isn't exfiltration. Three-party warranty: Allodial warrants the runtime, the author the encoding, the jurisdiction the decision.

    Furnace-Marketplace-Overview_rev1.md

    Used in Answer 3, S4

  20. A20

    Public cycles are long

    Government tech purchases average 22 months (Gartner); letter of intent to purchase order is about 12 months.

    02-Channel-Map_rev1.md

    Used in Part 3, S1, S4, S6

  21. A21

    HUBZone in Puerto Rico

    82.1% of Puerto Rico's census tracts qualify; a 10% price preference applies if at least 35% of employees live in a HUBZone.

    Opportunity-Register_rev1.md §5

    Used in Part 6

  22. A22

    There is no AI budget line

    Only 25% of state CIOs have dedicated GenAI funding; the first box is paid from an existing capital or modernisation line.

    Round 9

    Used in S1

Model8

  1. M01

    The seed is short

    $2M runs the Proof plan for about 9 months; $2.5M reaches 18. Proof buys certification ($0.8M), a benchmark and one reference site.

    Allodial-Investment-Options_rev2.md

    Used in Part 3, S1, Part 5, Part 8 · 1

  2. M02

    The implementation fee is what keeps a customer profitable

    With no fee, a customer loses about $88k up front. Break-even is about $81k to $88k; quote $90k, floor $75k.

    07-Red-Team_rev1.md; Step 9

    Used in Part 3, S1, S2, Part 8 · 3, Part 9 · 1

  3. M03

    Emberholt's price

    $125,056 street, $75,034 COGS.

    model rev9 and rev10

    Used in S1

  4. M04

    Founder outbound converts best

    Weighted letters per $100k: research computing 13.7, credit unions 8.0, county HHS 6.8.

    02-Channel-Map_rev1.md

    Used in Part 6, Part 9 · 2

  5. M05

    What a sprint buys

    $500k over 90 days: P50 of 10 Tier A letters (P25 of 8).

    03-Campaign-Portfolio_rev1.md

    Used in Part 5

  6. M06

    The flat service fee loses money

    $9k per box loses about $16.6k a year on a 3-box customer; proposed $24k plus $7k per box.

    Step 4

    Used in S1, Part 8 · 3

  7. M07

    Where unit economics swing

    Per customer: CAC $90k, Emberholt COGS $60k, contractor rates $58k.

    07-Red-Team_rev1.md §3

    Used in Part 3, Part 9 · 1

  8. M08

    Services in the model

    HHI-type services are modelled at $300k a year, 50% probability, off by default.

    Step 9

    Used in Answer 1, S3

Known2

  1. K01

    HHI

    An ongoing development contract of about $25k a month, about 90% likely to sign this week, replacing SDSOL.

    Wilson, 13 Oct 2026

    Used in Answer 1, S3, Part 6

  2. K02

    Hotel design partner

    A current Mews customer prepared to pay for a replacement.

    Builder-vs-Vertical-GTM-Report_rev1.md §5

    Used in S3, Part 6