AI-Native Service Business · Hard-to-Fool Blueprint

Climate Disclosure Engine — done-for-you California SB 253 GHG reports and the assurance-ready evidence file behind them

A productized compliance operation that ingests a large company's utility bills, fuel logs, fleet records, and facility data, builds a GHG-Protocol-conformant Scope 1 & 2 inventory, files the CARB-required report on the entity's behalf, and ships an audit-ready evidence binder that survives the limited assurance the same regulation phases in from 2027. We sell a filed, defensible disclosure — not a carbon-accounting dashboard the customer has to operate.

Sector: Corporate climate & ESG regulatory compliance · Buyer: $1B+ revenue companies doing business in California · Pricing: fixed per-entity annual report fee + assurance-readiness retainer

1 · Thesis

Roughly four thousand large companies must, for the first time in their history, file a government greenhouse-gas emissions report in California — the inaugural SB 253 Scope 1 & 2 deadline is now November 10, 2026, with civil penalties up to $500,000 per year for non-compliance Verified. Most of these companies have no internal carbon-accounting function, treat this as an unbudgeted fire drill, and face a choice between five-figure-a-year software they must learn and operate, or six-figure Big-Four consulting engagements. We sell the third option: a done-for-you service where our expert is the customer's single point of contact, our internal AI engine does the extraction, classification, emission-factor matching, and drafting, and the customer receives a filed report plus an evidence binder built to pass the limited assurance the law phases in from 2027 and the reasonable assurance it requires by 2030. The buyer experiences a sustainability controller who simply makes the obligation go away; behind that person sits a production machine that gets cheaper and faster every quarter. This is not a co-pilot because the customer never touches a calculation engine — they hand us shoebox data and receive a compliant outcome. The pain is acute and dated; the work is rule-governed and repeatable; and the escalating assurance ramp converts a one-time scramble into a recurring, defensible annuity.

2 · Discovery rationale

This run scanned regulatory-pressure terrain across healthcare administration, insurance ops, tax and audit, trade/customs, real estate, HR/benefits, banking/KYC, energy/environmental compliance, and government paperwork. The decisive signal was a dated, non-discretionary, first-of-its-kind filing obligation hitting thousands of large companies that lack the internal capability to meet it — the rare combination of a hard deadline, real penalties, and a near-universal capability gap.

California's climate-disclosure package cleared the noise. Three facts made it the winner over the other candidates: (1) the SB 253 first-year Scope 1 & 2 deadline is live and imminent — CARB moved it to November 10, 2026 on June 24, 2026, but did not cancel it Verified; (2) CARB published a preliminary list of roughly 4,160 entities with a likely reporting obligation, so the buyer set is enumerated rather than hypothetical Verified; and (3) the statute itself escalates the rigor over time — no assurance in 2026, limited assurance from 2027, reasonable assurance from 2030 Verified — which turns a one-off compliance task into a multi-year, deepening engagement. Crucially, today's market is bifurcated between self-serve software (Watershed, Persefoni, Sweep) and bespoke consulting; the productized done-for-you middle is thinly occupied, which is exactly the AI-native opening.

We explicitly rejected the partial-injunction trap: SB 261 (the climate-financial-risk report) is enjoined and unenforced for now, so we anchor the business on the non-enjoined SB 253 GHG obligation and treat SB 261 as an attach product carrying disclosed legal risk — rather than building the whole company on an enjoined statute.

3 · Candidate comparison

Six candidates generated this run; the winner was chosen on evidence quality, novelty vs. the seven prior blueprints, and fit to the AI-native rubric.

CandidateBuyerOutcome soldScore / evidenceDecision
California climate disclosure (SB 253) engine $1B+ revenue cos. doing business in CA; sustainability/finance lead Filed CARB-compliant Scope 1&2 GHG report + assurance-ready evidence binder High · live Nov 10 2026 deadline, enumerated buyer list, escalating assurance ramp, all Verified Selected
Unclaimed-property compliance & audit defense Corporate controller / treasury Compliant escheatment filings + minimized audit liability Med · aggressive state audits real, but no sharp 2026 catalyst Rejected — evergreen, weak "why now"; recovery mechanism overlaps prior runs
Cost-segregation study engine (post-OBBBA 100% bonus depreciation) CRE owners / CFOs / tax directors Engineering-based depreciation study that accelerates deductions Med-High · strong 2026 tax catalyst Rejected — buyer overlaps hour-07 property-tax-appeal; third tax-themed play, keep portfolio diverse
SOC 2 / ISO 27001 readiness-as-a-service B2B SaaS / AI startups Audit-ready security program by an SLA date Med · real demand Rejected — crowded (Vanta, Drata), weaker regulatory moat, commoditization risk
Workers'-comp premium audit & ex-mod recovery Mid-market employers / CFOs Recovered overpaid premium + corrected experience mod Med · evergreen Rejected — another recovery play, weak 2026 catalyst
OBBBA Medicaid work-requirement eligibility navigation Hospitals / FQHCs / SNFs Maintained patient Medicaid coverage = protected provider revenue Med-High · strong catalyst Rejected — vulnerable-population consumer exposure, navigator-licensing & heavier individual judgment

4 · Hard disqualifier check

#DisqualifierResultBasis / mitigation
1Primarily a co-pilot/SaaS the customer operatesPassWe deliver a filed report + binder; incumbents are software the customer runs. Our wedge is exactly the done-for-you gap.
2Requires physical labor / field crewsPassPure data/document work; no site visits required for Scope 1&2 inventory.
3Pricing depends on hourly / cost-plusPassFixed per-entity annual fee + assurance-readiness retainer; never hourly.
4Cannot reach 50%+ gross marginPass (inferred)Standards-based, automatable extraction/calc; margin path modeled in §15. Year-1 thinner, expands as automation rises.
5Buyer cannot be identifiedPassCARB published a ~4,160-entity preliminary list; buyer set is literally named.
6Workflow not decomposablePassGHG Protocol + CARB reg define discrete, repeatable steps (boundary → activity data → factors → calc → report).
7Fully automates regulated judgment w/o reviewPassHuman chokepoints: methodology QA + officer attestation; external limited-assurance provider (2027+) is separate and coordinated, not replaced.
8Duplicative of a prior blueprintPassNew buyer, regulator (CARB), inputs (energy/fuel data), and mechanism (report production, not recovery). See §3.
9Illegal / unworkable licensingPassNo professional license required to prepare a GHG inventory; assurance handled by accredited third parties we partner with.
10Core demand unverifiedPassLive statutory deadline + penalties + enumerated entity list, all Verified.
11Frontier models commoditize rather than strengthenPass (watch)Moat is the done-for-you ops layer + messy-data wrangling + regulatory QA; better models cut our COGS. Strongest threat (incumbent managed service) named in §17.
12Cannot be tested with a bounded pilotPass5–8 covered entities ahead of the Nov 10 deadline is a clean, time-boxed pilot.
No disqualifier fails. Two items pass on an inferred/watch basis (margin and model-commoditization) and are carried into the risk register and 90-day plan.

5 · Rubric scorecard

Gate 1 · Low trust burden — 4/5

Big companies already outsource ESG reporting to consultancies; buying the outcome from an external expert is normal. Slight friction: emissions data is sensitive and the report carries the company's name, so trust-building is required.

Gate 2 · Low task-level judgment — 4/5

Most steps — data extraction, unit normalization, emission-factor matching, calculation, report assembly — are deterministic or rules-bound. Judgment concentrates at organizational-boundary decisions and data-gap estimation.

Gate 3 · High intelligence threshold — 4/5

Requires synthesizing messy utility/fuel/fleet data against GHG Protocol, CARB rules, and emission-factor libraries, plus defensible treatment of estimates and exclusions. Not simple labor arbitrage; quality is the product.

Gate 4 · Regulation as moat — 5/5

Statutory mandate, civil penalties, named entity list, and an assurance ramp that raises the rigor bar yearly. Regulation creates the demand and deepens it over time.

Gate 5 · No physical labor — 5/5

Entirely digital: documents, spreadsheets, ERP/utility exports, PDFs. Remote delivery; no logistics.

Gate 6 · Sam Altman test — 4/5

Better models improve extraction accuracy, factor-matching, and narrative drafting — lowering COGS and raising throughput. Watch: well-funded software incumbents could fold equivalent automation into a managed service.

Composite: strong across all six gates, with regulation-as-moat and no-physical-labor maxed. The only sub-5 scores reflect data sensitivity and incumbent-commoditization risk, both actively managed.

6 · Opportunity

~4,160
Entities on CARB's preliminary SB 253/261 list V
Nov 10 2026
First SB 253 Scope 1&2 deadline (moved from Aug 10) V
$500k/yr
Max SB 253 civil penalty for non-compliance V
$1B
Revenue threshold to be covered by SB 253 V
2027
Limited assurance on Scope 1&2 begins V
2030
Reasonable assurance required V
$50k–$250k
Incumbent software annual list price band (Watershed/Persefoni) V

Illustrative TAM logic Inferred: if ~4,000 SB 253 entities pay a fixed done-for-you fee in the low-to-mid five figures for the inventory + filing, the California-only annual served market is on the order of $200M–$400M, before SB 261 attach, Scope 3 (2027), copycat-state expansion, and assurance-readiness retainers. Figures are inferred from the entity count and incumbent pricing, not a published market study; treated as directional only.

7 · Evidence quality & source-claim matrix

ClaimLabelSource / basisConf.Business impact
SB 253 covers companies >$1B revenue doing business in CA; SB 261 >$500MVerifiedCARB program page; Watershed; Baker Tilly; PwCHighDefines buyer set & ICP threshold
First SB 253 Scope 1&2 deadline moved to Nov 10, 2026 (from Aug 10)VerifiedLinklaters Sustainable Futures; CARB June 24 2026 bulletin (reported)HighSets the urgent sales trigger & pilot timing
No third-party assurance year 1; limited assurance from 2027; reasonable from 2030VerifiedWatershed; Davis Polk; Akin; ISS-CorporateHighYear-1 chokepoint is internal QA; assurance ramp = recurring revenue
Penalties up to $500k/yr (SB 253), $50k/yr (SB 261)VerifiedBaker Tilly; Greenberg Traurig; Nelson MullinsHighQuantifies cost of non-compliance = willingness to pay
CARB published preliminary list of ~4,160 covered entitiesVerifiedPersefoni; Hogan Lovells; Cleary GottliebHighEnumerated target list for outbound GTM
SB 261 enforcement enjoined by 9th Cir. (Nov 18 2025); CARB not enforcing Jan 1 2026 deadlineVerifiedWhite & Case; Wilson Sonsini; CARB Enforcement Advisory (reported)HighAnchor business on SB 253; treat SB 261 as risk-flagged attach
EU CSRD scope cut ~90% via Omnibus I (Feb 2026)VerifiedEU Council press release; Gibson Dunn; Accountancy EuropeHighEU is NOT a tailwind; durable demand is US/CA + value-chain pull
Incumbents (Watershed ~$50k+/yr; Persefoni $55k–$250k) are software-ledVerifiedNormative; Sweep; vendor pricing summariesMedValidates price ceiling & the done-for-you whitespace
Scope 3 (value-chain) reporting begins 2027 for FY2026 dataVerifiedPwC; BDO; CARB program materialsHighLarge expansion of scope & fees in year 2
Most covered firms lack an internal carbon-accounting functionInferredInferred from incumbent software/consulting market existing to fill the gapMedCore demand thesis for done-for-you; validate in pilot
Other US states (NY, NJ, IL, CO) advancing similar disclosure billsUnverifiedNot re-confirmed this run; widely discussed pre-2026LowExpansion optionality only; not used to select winner
Fixed done-for-you fee of low-to-mid five figures is acceptable to buyersUnverifiedPricing hypothesis benchmarked to incumbent software/consultingLowMust be proven in pilot pricing test (§27)

No claim is both "Verified" and uncited. Unverified items (state copycats, exact willingness-to-pay) are excluded from winner selection and routed to the validation plan.

8 · Why now

Verified regulatory change

  • SB 253's first-ever Scope 1&2 filing is due Nov 10, 2026 — a hard, near-term deadline with $500k/yr penalties.
  • CARB finalized initial regulations in early 2026 and named ~4,160 likely covered entities.
  • The assurance bar escalates on a fixed schedule (2027 limited → 2030 reasonable), so rigor only increases.

Inferred capability change

  • Frontier multimodal models can now reliably parse utility bills, fuel invoices, and fleet logs into structured activity data.
  • LLMs can match activity data to emission factors and draft GHG-Protocol-aligned documentation with expert review, collapsing labor per inventory.

Unverified / hypotheses

  • Other states replicate California within 2–3 years (optionality, not relied upon).
  • Value-chain pressure forces sub-threshold suppliers to report voluntarily, expanding the buyer pool.
Honest counter-signal: the EU narrowed CSRD by ~90% and SB 261 is enjoined — the global regulatory tide is not uniformly rising. The durable demand here is the non-enjoined SB 253 obligation plus customer/value-chain pull, not a presumption that every climate rule survives. This is built into the kill-criteria.

9 · Customer & product-market fit

AttributeDetail
ICPUS-headquartered company with >$1B total annual revenue that "does business in California," no mature in-house carbon-accounting team. Sweet spot: $1B–$10B industrials, retail, food & beverage, real estate, manufacturing, and services firms — large enough to be covered, too small to have a sustainability department.
Economic buyerCFO or General Counsel (owns regulatory penalty risk and the officer attestation).
Champion / userHead of Sustainability/ESG, Controller, or EHS lead handed the obligation with no tooling.
Urgent triggerReceipt of CARB covered-entity notice; board/audit-committee question; the Nov 10 2026 deadline; a customer or investor demanding the disclosure.
Alternatives(1) Buy software and self-operate; (2) Big-Four / sustainability consultancy; (3) stretch an internal generalist; (4) do nothing and risk penalty + reputational exposure.
Jobs-to-be-done"Make this filing happen correctly and on time, defensibly, without me building a function for it — and keep it defensible when assurance kicks in."
Willingness to payInferred from the $50k–$250k software band and far higher consulting costs; a fixed done-for-you fee below consulting and at-or-near software, with zero internal labor, is plausibly attractive. To be proven in pilot.

10 · The outcome we sell

Deliverable

A submitted, CARB-conformant Scope 1 & 2 GHG emissions report for the covered entity, plus a structured assurance-ready evidence binder: organizational-boundary memo, data-source register, emission-factor citations, calculation workpapers, estimation/exclusion log, and an officer-attestation package.

Acceptance criteria

  • Report accepted by CARB's reporting mechanism without deficiency notice.
  • Every reported figure traceable to a source document and an emission factor.
  • Binder structured to the standards limited assurance will test from 2027 (ISO 14064-3 / ISSA 5000 mapping).

Customer promise

"You hand us your raw data; we hand you a filed, defensible disclosure and a binder that survives audit. One contact, fixed fee, on the deadline."

Exclusions & policy

  • Excludes legal opinions on coverage (we coordinate with the client's counsel) and third-party assurance opinions (separate accredited provider).
  • Rework: any CARB deficiency we caused is remediated at no charge before deadline; missed-filing-due-to-us triggers fee credit.
  • Success metric: 100% on-time, deficiency-free filings; binder passes assurance with zero material findings (2027+).

11 · Internal AI engine architecture

The customer never operates this. It is our production line; the expert is their interface.

1 · Intake

Secure portal + email-in. Utility bills (PDF/CSV), natural-gas & fuel invoices, fleet/telematics exports, refrigerant logs, facility lists, ERP/AP feeds, prior sustainability reports, org charts for boundary scoping.

2 · Normalization

OCR + LLM extraction into a canonical activity-data schema (site, meter, period, quantity, unit, fuel type). Unit conversion, de-duplication, period alignment, versioning, and gap flagging.

3 · Retrieval / knowledge

Indexed GHG Protocol Corporate Standard, CARB regulation text & FAQs, EPA/eGRID & IEA emission-factor libraries, GWP tables, prior-client boundary precedents, and assurance-standard checklists.

4 · AI workbench

Models classify each activity record to a scope/category, match it to the correct emission factor, compute CO₂e, draft the boundary memo and methodology narrative, and assemble the workpapers and exclusion log.

5 · Deterministic rules

Emission-factor application, unit math, GWP multiplication, eGRID subregion mapping, and report-schema validation are code, not probabilistic — every number is reproducible.

6 · Human chokepoint

A GHG-inventory specialist reviews boundary decisions, data-gap estimates, factor selection on edge cases, and sign-off readiness; a senior reviewer approves the officer-attestation package.

7 · QA layer

Automated completeness checks (every facility has data or a logged exclusion), variance vs. prior year, factor-version checks, and a red-team pass before delivery.

8 · Delivery

Filing to CARB's mechanism on the client's behalf (with authorization), plus the evidence binder and a one-page executive summary for the attesting officer.

9 · Learning loop

CARB deficiency notes, assurance findings (2027+), and reviewer edits feed back into extraction prompts, factor-matching rules, and gold-standard templates.

10 · Model portability

Provider-agnostic abstraction over extraction/drafting models; deterministic calc layer is model-independent, so we adopt better models without re-architecting.

12 · AI-vs-human operations pipeline

AI
Extract activity data from bills, invoices, fleet & refrigerant logs
AI
Classify records to scope/category; flag gaps
Rules
Unit conversion, eGRID/factor mapping, CO₂e math, GWP
Operator
Review boundary memo, estimates, edge-case factors
AI
Draft methodology narrative, exclusion log, workpapers
Rules
Completeness & report-schema validation
Senior reviewer
Approve attestation package & sign-off readiness
Customer
Officer reviews & attests; we file to CARB

Blue = AI-owned · grey = deterministic rules · amber = trained operator · red = senior reviewer chokepoint · green = customer touchpoint. The customer sees only the first and last green steps and their named expert.

13 · Operations as product

Standardization

  • One canonical activity-data schema across all clients.
  • Boundary-scoping SOP with a decision tree (operational vs. financial control).
  • Required-evidence checklist per facility type generated at intake.

Variance elimination

  • Automated completeness gate: no facility ships without data or a logged, reasoned exclusion.
  • Confidence scoring on every extracted record; low-confidence routes to an exception queue.
  • Year-over-year variance flags catch anomalies before filing.

Control & auditability

  • Full version control and immutable audit trail on every figure and edit.
  • Reviewer-assignment logic by sector/complexity.
  • Gold-standard binders as templates and training data.

Continuous improvement

  • Root-cause analysis on every CARB deficiency or assurance finding.
  • Postmortem loop converts each recurring manual fix into a rule, template, or model-prompt change.

14 · No-holes quality engine

The failure we must prevent is a wrong or unfilable disclosure carrying the client's name and officer attestation. Defenses, in layers:

  • Source traceability: every reported tonne traces to a specific document, factor, and version; numbers with no provenance cannot enter the report (deterministic gate).
  • Completeness gate: facility roster reconciled against data received; any missing site is either resolved or appears in the exclusion log with a documented rationale.
  • Dual-path calc check: AI-proposed classification is recomputed by the deterministic calc layer; mismatches are quarantined for operator review.
  • Hallucination control: emission factors and regulatory citations are retrieved from indexed authoritative libraries, never free-generated; the model cites the library record ID.
  • Red-team pass: a checklist adversarially probes boundary, double-counting, factor-vintage, and unit errors before release.
  • Human sign-off: senior reviewer approval is mandatory before any filing; the attesting officer receives a plain-English summary of methods, estimates, and exclusions.
  • Assurance-forward design: binders are built to the evidence standard limited assurance will test from 2027, so quality is validated by an external party on a known schedule.

15 · Pricing, legality & unit economics

Primary model

Fixed per-entity annual report fee (tiered by facility count / data complexity), plus an assurance-readiness retainer that grows as the 2027/2030 assurance bar rises. Scope 3 (2027) priced as an add-on module.

Why not hourly

Hourly punishes our own automation and caps margin. A fixed outcome fee lets every efficiency gain accrue to us while the customer gets price certainty — the core AI-native flywheel.

Pricing legality

No contingency/percentage issues here (this isn't recovery). The only boundary: we don't sell legal coverage opinions or assurance opinions — both are separate, so no unauthorized-practice or independence conflicts. Clean fixed-fee service.

Compliance-safe structure

We prepare; the client's officer attests; an independent accredited firm assures (2027+). Keeping preparation and assurance separate preserves assurer independence and is itself a selling point.

Unit economics (illustrative, mid-tier entity)Year 1After 90 daysAfter 1 year
Model inference / doc-processing per inventory Inf$300–$700$200–$450$120–$300
Hosting, storage, third-party data/APIs Inf$150$150$150
Human review minutes per inventory Unv~600 min~360 min~180 min
Expected automation rate Unv~45%~65%~80%
Target gross margin Inf~40%~55%60%+

COGS drivers tracked from day one: model inference, document processing, storage/hosting, third-party emission-factor data, human-in-the-loop review, senior QA, CARB filing handling, and rework. Year-1 margin is intentionally thinner while the engine learns; margin expands as automation rises and review minutes fall. Review-minute and automation figures are unverified launch targets to be measured in the pilot.

16 · Nonlinear scaling plan

$700k–$1M
Revenue per FTE target at maturity Unv
~80%
Target automation rate after year 1 Unv
3–4×
Inventories per specialist vs. manual baseline Inf

Revenue decouples from headcount because the marginal inventory is mostly machine work: extraction, classification, factor-matching, and drafting scale with compute, while the specialist's time per inventory falls each cycle as the exception queue shrinks and templates harden. The same client recurs annually (and pays more as Scope 3 and assurance phase in), so the book compounds without proportional hiring. The constraint is reviewer capacity at the chokepoint — addressed by pushing the automation rate up and reserving humans for boundary judgment and sign-off. Margin expands along the curve: thinner in year 1, 60%+ as automation crosses ~80% and review minutes per unit roughly halve.

17 · Moat & Sam Altman test

Does model progress strengthen or commoditize us? Strengthen — on net. Each model generation improves messy-document extraction, factor matching, and narrative drafting, directly cutting our COGS and lifting throughput, while the deterministic calc layer and the assurance-grade evidence discipline remain ours regardless of model. Our compounding assets are proprietary: the corrected-data corpus, boundary-precedent library, gold-standard binders, and the CARB-deficiency/assurance-finding feedback loop — none of which a raw model possesses.

Strongest commoditization threat: a well-funded incumbent (Watershed, Persefoni, a Big-Four practice) bolts a done-for-you managed-service tier onto existing software and undercuts us at scale. Mitigation: win the productized mid-market they under-serve, move faster on the assurance-readiness niche, and lock in multi-year recurring relationships before they pivot. We monitor this as a live, named risk.

18 · Buyer-specific go-to-market

Buyer behavior dictates the motion: large enterprises with a dated regulatory obligation respond to founder-led outbound timed to the trigger plus channel referral from advisors — not waitlists or creator content.

Why this GTM

The buyer is enumerated (CARB's ~4,160-entity list), penalty-motivated, and deadline-bound. This rewards precise outbound and warm intros from the lawyers, auditors, and consultants already advising these companies on the rule.

First 50 prospects

$1B–$10B firms on/adjacent to CARB's covered list with no visible sustainability hire (LinkedIn signal) and a California nexus — concentrated in industrials, retail, food & bev, and real estate.

Trigger events

CARB covered-entity notice; audit-committee agenda items; investor/customer ESG-data requests; approaching Nov 10 deadline; competitors' first disclosures appearing.

Channel strategy

Referral partnerships with ESG/securities counsel, mid-tier audit firms (who can't staff prep but will assure later), and benefits/risk consultancies — they hand off prep, we hand assurance back to them.

Outreach wedge

"You're likely on CARB's list with a Nov 10 filing and no team for it. We file it for you, fixed fee, and the evidence binder is built to survive the assurance starting in 2027."

Conversion path & metrics

Free 20-minute coverage-and-readiness triage → fixed-fee proposal → data intake → filing. Track: triage-booked rate, triage→proposal, proposal→close, time-to-file, and net revenue retention (Scope 3 + assurance expansion).

19 · Pilot design & early-demand-trap mitigation

Pilot cap

Hard cap of 5–8 covered entities for the Nov 10, 2026 cycle. Refuse more even if demand exceeds it — the goal is a hardened engine, not maximum logos.

Pilot ICP

$1B–$5B firms with 5–40 facilities, mostly Scope 1&2-relevant operations, and an internal champion who will give us real data fast.

Success criteria

  • 100% on-time, deficiency-free CARB filings.
  • Measured automation rate ≥ target by client #5.
  • ≥3 references willing to vouch + 1 channel partner activated.

Demand-trap discipline

  • Instrument every manual workaround; log minutes by step.
  • Refuse bespoke client asks that don't generalize.
  • Convert each repeated human fix into a rule/template/prompt before scaling.
  • Define the process-hardening gate that must close before client #9.
Idea-killer evidence: if specialist review minutes per inventory don't fall materially across the 5–8 pilots, the automation thesis is wrong and the margin won't reach 50%+ — that is a stop signal, not a "hire more people" signal.

20 · Competitive landscape

AlternativeWhat they areGap we exploit
Watershed / Persefoni / Sweep / NormativeCarbon-accounting software ($50k–$250k/yr) the customer largely operatesCustomer must staff and run it; we are done-for-you with no internal labor required
Big-Four & sustainability consultanciesBespoke, high-touch ESG engagementsExpensive, slow, partner-leveraged; we are productized, fixed-fee, AI-native
Boutique ESG consultantsManual prep shopsLinear labor model; we out-automate on cost, speed, and consistency
Internal generalist (controller/EHS)Stretched employee with a spreadsheetNo methodology, no audit trail, high penalty risk; we de-risk it
Do nothingBet on non-enforcement$500k/yr exposure + board/investor pressure; our triage quantifies the risk

21 · Regulation, compliance & licensing boundary

What our AI/operators do

Extract activity data, classify to scope, apply emission factors, compute CO₂e, draft methodology and workpapers, validate the report schema, and assemble the evidence binder.

What stays with humans

Operators review boundary and estimation decisions; a senior reviewer approves sign-off readiness. The client's officer attests to the report; the client's counsel opines on coverage; an independent accredited firm provides assurance (2027+).

Prohibited claims

We do not provide legal opinions, do not issue assurance, and do not guarantee a specific regulatory interpretation. We prepare a defensible inventory; we don't certify it ourselves.

Controls

Data-processing agreement and confidentiality on sensitive operational data; immutable audit logs; clear authorization before any filing; independence firewall between our prep work and any assurer.

Regulatory risk is real and disclosed: SB 261 is enjoined; the SB 253 deadline has already slipped once; litigation and political change could narrow or delay the regime. We anchor on the live SB 253 obligation, diversify toward voluntary/value-chain and multi-state demand, and keep the cost base variable so a delay is survivable.

22 · Compact founding team & expert map

RoleWhy neededFT / fractionalFirst hire timing
GHG-inventory domain expertOwns methodology, boundary judgment, assurance-readiness; the customer-facing trust interfaceFull-timeFounder / hire #1
AI/automation engineerBuilds extraction, factor-matching, deterministic calc layer, QA gatesFull-timeFounder / hire #1
Operations lead / senior reviewerRuns the production line; final sign-off chokepointFull-time by client #3Pilot
Sales / channel leadOutbound to CARB-list entities; advisor referral networkFractional → FTPost-pilot
ESG/securities counselCoverage interpretation, disclaimers, filing authorizationFractional (outside counsel)Pilot
QA ownerRed-team checklist, deficiency root-cause loopFractional at launchPilot

23 · Exhaustive risk register

1 · SB 253 is repealed, delayed again, or further enjoined

Likelihood: Medium · Impact: High · Verified precedent (SB 261 enjoined; SB 253 deadline already moved). Mitigation: anchor on live obligation, diversify to voluntary/value-chain and multi-state demand, keep cost base variable. Owner: CEO. Leading indicator: CARB bulletins, 9th Cir. docket.

2 · Incumbent software adds a done-for-you managed tier and undercuts us

Likelihood: Medium-High · Impact: High · Inferred. Mitigation: own the under-served productized mid-market, lead on assurance-readiness, lock multi-year contracts. Owner: CEO. Leading indicator: incumbent "managed service" launches.

3 · Automation rate stalls; review minutes don't fall

Likelihood: Medium · Impact: High · Unverified until pilot. Mitigation: instrument every step; kill-criteria if minutes/inventory don't drop across pilots. Owner: Eng + Ops. Leading indicator: per-inventory review-minute trend.

4 · A filed report contains a material error under the client's attestation

Likelihood: Low-Medium · Impact: Very High · Inferred. Mitigation: source-traceability gate, dual-path calc check, red-team pass, mandatory senior sign-off, E&O insurance. Owner: Senior reviewer. Leading indicator: QA defect rate, deficiency notices.

5 · Willingness-to-pay below our fixed-fee target

Likelihood: Medium · Impact: High · Unverified. Mitigation: pilot pricing test vs. software/consulting anchors; tier by complexity. Owner: CEO. Leading indicator: proposal win-rate vs. price.

6 · Client data is incomplete, late, or low-quality

Likelihood: High · Impact: Medium · Inferred. Mitigation: structured intake checklist, completeness gate, documented estimation methods, early data-readiness call. Owner: Ops. Leading indicator: intake-completeness score.

7 · Emission-factor or standard changes break calculations

Likelihood: Medium · Impact: Medium · Inferred. Mitigation: versioned factor library, deterministic calc layer, regression tests on factor updates. Owner: Eng. Leading indicator: eGRID/IEA/CARB factor releases.

8 · Data-security or confidentiality breach of sensitive operational data

Likelihood: Low · Impact: Very High · Inferred. Mitigation: encryption, least-privilege access, DPAs, SOC 2 roadmap, no training on client data without consent. Owner: Eng. Leading indicator: access-audit anomalies.

9 · Assurance providers (2027+) reject our binder format

Likelihood: Medium · Impact: High · Inferred. Mitigation: co-design binder to ISO 14064-3 / ISSA 5000; pre-validate with a partner assurer in 2026. Owner: Domain expert. Leading indicator: assurer feedback in pilot.

10 · Reviewer-capacity bottleneck caps growth

Likelihood: Medium · Impact: Medium · Inferred. Mitigation: raise automation rate, exception-only review, tiered reviewer pool. Owner: Ops. Leading indicator: queue depth, reviewer utilization.

11 · Scope 3 (2027) is far harder and erodes margin

Likelihood: High · Impact: Medium · Verified (Scope 3 begins 2027). Mitigation: price Scope 3 as a separate module; supplier-data automation; phase carefully. Owner: CEO. Leading indicator: CARB Scope 3 rulemaking.

12 · Sales cycle to large enterprises is too slow for a deadline-driven model

Likelihood: Medium · Impact: Medium · Inferred. Mitigation: trigger-timed outbound, channel referrals, fast fixed-fee proposals, "we file in N weeks" framing. Owner: Sales. Leading indicator: triage→close time.

24 · Tech stack & build plan

Stack

  • Intake: secure upload portal + email ingestion; object storage with per-client isolation.
  • Extraction: multimodal LLM + OCR into a typed activity-data schema (Pydantic-validated).
  • Knowledge: vector index over GHG Protocol, CARB text, emission-factor libraries (eGRID/IEA), GWP tables; retrieval returns record IDs for citation.
  • Calc: deterministic Python service (no model in the math path) with unit tests + factor-version pinning.
  • Workbench: reviewer UI with confidence scores, exception queue, diff view, audit trail.
  • Delivery: report generator + CARB filing handoff; binder export (PDF/structured).

Build sequence

  • Wk 1–3: schema + extraction + factor library + deterministic calc core.
  • Wk 4–6: reviewer workbench, completeness/QA gates, binder template.
  • Wk 7–9: run pilot client #1 end-to-end; instrument every manual step.
  • Wk 10–12: harden the top recurring manual fixes into rules/templates; onboard clients #2–5.
  • Throughout: regression tests on factor updates; deficiency/finding feedback loop.

No "just use agents" hand-waving: the math is deterministic and tested; models are confined to extraction, classification, and drafting, each behind a validation gate.

25 · Metrics & KPIs

Throughput
Inventories filed per specialist per cycle
Cycle time
Intake → filed report (days)
Rework rate
Reports needing post-review rework
Gross margin
Target 60%+ at maturity
COGS / unit
Model + review + filing cost per inventory
Rev / FTE
$700k–$1M target
Escalation rate
% records hitting exception queue
Automation rate
% steps with no human touch
Evidence completeness
% figures fully source-traceable
Quality failure rate
CARB deficiencies / assurance findings
Acceptance rate
% filings accepted without deficiency
Pilot conversion
Triage → proposal → close

26 · What could kill this

  • Regulatory rollback: SB 253 is repealed, gutted, or indefinitely enjoined like SB 261 before recurring demand is established.
  • Automation never arrives: review minutes per inventory stay high; the business is just a consultancy with worse branding and sub-50% margins.
  • Incumbent crush: Watershed/Persefoni/Big-Four launch a cheaper done-for-you tier and out-distribute us.
  • A high-profile error: a material misstatement under a client's attestation destroys trust in a reputation-driven market.
  • Price compression: buyers anchor to "it's just a report" and won't pay enough to clear our COGS.
  • Enterprise sales drag: deals take longer than the deadline allows, starving the pilot of reference logos.

27 · 90-day validation & launch plan

WeeksFocusProof / kill-criteria
1–2Confirm CARB covered-entity list; build target list of 50 $1B–$10B firms with CA nexus and no sustainability hire. Recruit a partner assurer + ESG counsel.≥50 qualified targets; assurer + counsel engaged.
3–4Build extraction + deterministic calc core + factor library; draft binder template mapped to ISO 14064-3 / ISSA 5000.End-to-end calc on sample data reproducible & cited.
3–6Outbound to target list; run free coverage-and-readiness triage; sign 5–8 pilot entities at fixed fee (pricing test across tiers).≥5 signed; price holds at target band → else revisit model.
5–9Run pilot inventories end-to-end; instrument every manual workaround; log review minutes per step.Review minutes/inventory trend down client #1→#5 → core thesis; flat = stop.
8–11Harden top recurring manual fixes into rules/templates; pre-validate one binder with the partner assurer.Assurer: no material gaps; automation rate ≥ target.
10–13File all pilot reports ahead of Nov 10 deadline; capture references; sign first channel partner.100% on-time, deficiency-free; ≥3 references; 1 channel partner.

The plan optimizes for proof — automation curve, pricing, and assurance-readiness — not logo count. The pilot cap is a deliberate constraint, not a limit we expect to regret.

28 · Sources

Hard-to-fool blueprint · evidence-labeled (Verified / Inferred / Unverified) · generated 2026-06-28 09:00 run. Claims about present-day regulation reflect sources retrieved this run; verify against CARB before acting.
↑ Top