Market Requirements Document

The category. Regulated enterprises — companies whose obligations change because a government body issues an order and a clock starts, not because they chose to change them. Utilities, banks, insurers, pharma, telecoms, pipelines.

The beachhead, and only this one. Investor-owned utilities and independent power producers. ADR-001 names one user and refuses a persona list, for the same reason this refuses the category: a product for “regulated enterprises” is a product for nobody.

Why the problem is structurally hard, which matters more than its size. A company's obligations are set proceeding by proceeding, docket by docket, across one or more state commissions and, for wholesale power or interstate transmission, FERC. A multi-state holding company does not run one open docket. It runs one per jurisdiction, each on its own procedural clock, each amended independently. The unit of regulatory obligation is fragmented by design, and no single filing tells a company everything it now owes. That fragmentation is the market opening. It is not a temporary inefficiency anyone is about to remove.

Supply side, sourced. The EIA counts 168 investor-owned utilities serving an average of 654,600 customers each, inside roughly 3,000 electric distribution companies (EIA, 2017 data). The beachhead is a few hundred buying entities, not thousands.

Demand side, unknown, and it is the part that decides the answer. Nobody here knows the average regulatory-affairs headcount at one of those utilities, or what one would pay per seat or per docket.

So there is no TAM figure in this document, and that is a decision rather than an omission. Multiplying 168 by two invented numbers yields three significant digits and no evidence — a confident assertion whose citation does not verify, which is the exact failure this product exists to refuse. Two questions asked of one real analyst close it: how many of you are there, and what do you pay today for anything adjacent. Until then the honest statement is that the beachhead is measured in hundreds of buyers and its size in dollars is unknown.

Sourcing note. The three forces below are asserted from industry knowledge and are not sourced here. They are load-bearing for the argument, so they are flagged rather than deleted.

Load growth. Data-centre demand is pushing utilities to file for generation, transmission and rate changes at a pace not seen in decades. Every filing opens or amends a docket.

Interconnection reform. FERC Order No. 2023 and the state proceedings after it rewrote how projects join the queue. Reform at that scale does not land as one final rule — it lands as orders on rehearing, compliance filings and tariff amendments, each a new version to re-read.

Energy-transition rulemaking. Clean-energy targets, resource planning cycles and state implementation of federal programmes are producing more proceedings, each revising more often before it settles.

Why they compound rather than add. None is new alone. What is new is that all three land on one desk at once: more proceedings, more amendments per proceeding, and analyst headcount that has not moved. The volume of change is rising faster than the headcount tracking it, and that gap is structural rather than a one-year spike.

User. The regulatory-affairs analyst who reads the docket, works out what changed, and tells the business what it must do.

Buyer. Almost certainly their manager — a VP of Regulatory Affairs, a chief compliance officer, general counsel. Someone measured on whether the company missed a deadline, not on hours spent reading. Hypothesis, not a finding.

Why it changes the product and not just the pitch. The user buys time and trust: did it read the docket correctly, and can I verify a claim without redoing the work. The buyer buys risk: can we show an auditor that this order was seen, assessed and acted on inside the deadline, with a record of who reviewed it. A demo that only shows the analyst's daily win stalls at the buyer's desk; a pitch that only shows the audit trail builds a tool nobody opens. The product has to carry both, which is why the audit chain is not a compliance checkbox bolted on late.

Four categories already sit near this problem and each does something real. None does the join this product is built around: an exactly-cited change, in a specific proceeding, mapped to this company's own obligations, with an action and a reviewer attached.

Categories, not named vendors. This document holds no sourced, current facts about any specific product and states none. If an incumbent has already closed part of this gap, a competitive trial surfaces it; guessing here would not.

The wedge. Between the alert and the memo. One tells you something moved; the other tells you slowly and expensively what it meant. Verbatim's bet is the middle: an exactly-cited account of what changed, mapped to this company's obligations, cheap enough to run on every amendment rather than only the ones that already look expensive — and honest enough to escalate when it does not know.

Not a moat: the model call. Any funded competitor can wire an LLM to a docket feed next quarter. Every incumbent above will add a model this year regardless of what this company does. A defensibility pitch resting on having used AI first is not a pitch.

The floor: the correctness engine. Citation verification against exact source offsets (ADR-003) and deterministic diffing before model interpretation (ADR-004). These are table stakes — the work a buyer must see before trusting the tool with anything. A competent team rebuilds both in a few weeks. They earn the right to be trusted; they keep nobody out.

The moat: the obligation graph. The mapping between how a regulator words a requirement and how one specific company words its own obligation, built proceeding by proceeding and jurisdiction by jurisdiction. It compounds per customer, and it cannot be bought — a competitor starting today rebuilds it customer by customer.

The slower, stronger candidate. If a compliance officer starts pointing at Verbatim's record to answer “what did we know and when”, removal stops being a vendor swap and becomes a legal-risk decision. That lock is stronger than the taxonomy and slower to earn than either.

Step 1 — same regulator, adjacent players. Independent power producers and retailers sit under the same state-PUC-plus-FERC structure. Engine transfers; the obligation vocabulary shifts toward interconnection agreements and power-purchase contracts. Cheapest real expansion, and it is also the test: if this takes a rebuild rather than a re-tag, the generality claim weakens considerably.

Step 2 — adjacent regulator. Telecom, under the FCC and state commissions in a structurally similar docket-and-order pattern. Diff and citation transfer; the taxonomy and the materiality judgment need rebuilding, because “material” in a telecom tariff and in a rate case are different questions.

Insurance — the portability argument, and nothing is being built. Held as an argument from regulation rather than a roadmap item. A carrier files rates and forms with each state before use, almost all through SERFF, and the shape is submission, objection, amended response, disposition — docket-and-amendment exactly. Utah Admin. Code R590-227-11 obliges a filer answering an objection to supply an underline-and-strikeout version, which is the artefact app/diff/engine.py already produces: a regulator mandating the output. The fragmentation argument is worse here than for utilities, not weaker — one product filed across fifty jurisdictions, each amended on its own clock.
The objections, before anyone raises them. Perr&Knight, Oden and Vertafore already sell into it, so incumbency is heavier than the utility beachhead. Rate filings are more form-structured, so less free-text diff value transfers than the similarity suggests. SERFF public access varies by state, making ingestion fifty questions rather than one. No filings analyst has been asked.

The gate on every step. Expansion is earned by repeating the user discipline in the new vertical, not by copying the schema and assuming it holds.

The buyer needs a felt cost, not an inferred one. Someone must be able to point at what a missed deadline or a wrongly-timed action actually cost — a penalty, a rate case, a filing redone. Whether buyers reason and budget in those terms is unconfirmed.

Usage has to be continuous. Dockets run on an ongoing calendar and do not stop when a company buys a tool, which is what a subscription needs. Inference from how dockets work, not evidence from a customer.

The escalation queue has to stay usable at scale. ADR-006 names the risk itself: a queue nobody empties is its own failure. The eval scores correctness on a fixed corpus and says nothing about throughput under load.

The sale has to land in a tool budget, not against a hire. A pitch that reads as “hire fewer analysts” invites resistance from the user adoption depends on. A positioning judgment made without a buyer conversation.