Competitive Dossier · Sheet 2 of 2
Build vs. Buy / Offense

Compete: Build the Dataset Ourselves Audit Analytics' product is human-coded datasets assembled on top of our search. We sit upstream of them — which means we could produce the same data. What it would take, and whether we should.

Premise  We own the corpus + extraction They own  Analyst labor + brand Tension  They are a paying customer

01 The thesis

Their datasets are just our search results, read and coded by people. The pipeline that produces the candidates — SEC ingestion, full-text + section search, proximity grammar — is already ours. The only layer we don't run today is the extraction that turns a matched paragraph into a structured row. And we build exactly that layer elsewhere in this codebase.

Where the value actually sits

A restatement record — {cik, form, period, type: little-r, reason, $ amount} — is produced by (a) finding the filing, then (b) reading it and coding the fields. We already do (a) at scale. Step (b) is what their offshore team does by hand, and it's precisely what an LLM extraction pass does in our gpt_* pipelines.

02 The value chain — who owns what

We already own

The hard, capital-intensive infrastructure.
Real-time SEC ingestion
raptor feeds, sec4 / kitems3 / exh2 corpora
Full-text + section search
boolean · proximity (W/N, ADJ/N) · POS · faceting
Section extraction
kitems3 splits filings into part/item units
LLM field extraction
proven in gpt_compensation_summary & friends

They add on top

Labor and trust — replicable and not.
Analyst hand-coding
replicable with LLM extraction + QA
20+ yr historical backfile
we have the filings; coding history is the work
Methodology & accuracy reputation
the real moat — trusted by academics, PCAOB
Distribution & relationships
Wharton/WRDS, Bloomberg, insurers

03 What we'd build — dataset by dataset

Each line is: their query becomes our candidate filter; an extraction pass fills the fields. The search half is done today.

DatasetCandidate filter (from their own queries)ExtractBuild
Non-reliance restatements restatement / out-period / immaterial → 8-K 4.02, 10-K/A type (Big-R/little-r), period, reason, $ amount Medium
Out-of-period adjustments out-of-period · overstate · prior period → 10-K/10-Q direction, amount, account affected Medium
Changes in estimate estimate change · useful life · depreciation estimate type, effect direction Low
Insider pledging pledge W/20 shares · margin · collateral → DEF 14A person, # shares pledged, lender Low
Cyber incidents cyber · attack → 8-K Item 1.05 incident date, material?, systems, cost Low
Auditor / opinion changes auditor · independent accountant → 8-K 4.01 old/new auditor, reason, disagreements Low

"Build" rates the extraction + QA effort only — the candidate search is already live. Structured items (8-K with a named SEC item) are easiest; narrative restatements are the hardest to code accurately.

04 Our edge vs. their moat

Our structural edge

  • Real-time. We ingest as filings land; their coded datasets lag by weeks. We could ship "same-day restatements."
  • Marginal cost near zero. LLM extraction vs. a salaried analyst pool per record.
  • We own the funnel. No licensing the search from anyone — it's our infrastructure.
  • Breadth for free. Foreign (20-F/40-F) and fund (N-CSR) forms are already in-corpus.

Their real moat

  • Trust & methodology. 20 years of consistent, audited definitions researchers cite. Accuracy is the product, not coverage.
  • Ground truth is hard. Coding "is this a Big-R restatement" correctly at scale is genuinely difficult; errors destroy credibility.
  • Distribution locked. Embedded in WRDS, Bloomberg, insurer workflows.
  • Backfile depth. 2000s-era history coded and reconciled.

05 The strategic fork

Path A — keep them

Arms dealer

Sell them the extraction and alerts (Sheet 1). Higher margin than search, no channel conflict, and it deepens their dependence on us. We profit from every dataset they build without carrying the accuracy risk or distribution cost.

Lower risk, faster revenue. Recommended default. We win whether or not we ever compete.
Path B — replace them

Compete

Build and sell the datasets directly. Bigger prize (we capture the full dataset margin, not the search fee) but we'd be competing with a paying customer — they'd likely pull spend the moment they noticed — and we'd have to earn trust that took them 20 years.

Higher ceiling, real conflict. Only with a data line they don't already own, or as a deliberate pivot.
Channel conflict — read before choosing B

Audit Analytics/Ideagen is an active 40+ seat account. Competing head-on risks that revenue outright. The extraction tech serves both paths — so build the capability (Path A revenue funds it), and hold Path B for a net-new dataset where we're not cannibalizing a customer relationship.

06 If we test it — a contained pilot

One dataset. Their query as the filter. Measured against a known answer.

Pick cyber incidents (8-K Item 1.05) — structured, newly regulated, and one they're only ramping, so it's the least-conflicted place to prove the method.

Filter. Run their own cyber sweep as the candidate query over 8-K 1.05, last 12 months — search half is already live.

Extract. LLM pass on matched sections → {incident date, material flag, systems, est. cost}, reusing the gpt_* pattern.

Grade. Score precision/recall against a hand-checked sample. This is the whole question — can we hit dataset-grade accuracy?

Decide. If accuracy clears the bar cheaply, we have a repeatable machine; point it at the next line. If not, Path A only — and we've learned it for the price of one pilot.

07 Risk ledger

High

Channel conflict. Competing with a 40+ seat customer can cost the account. Mitigate by leading with Path A and reserving Path B for net-new data.

High

Accuracy bar. Dataset credibility is binary — a few wrong Big-R calls and researchers won't trust the feed. QA is the real cost, not extraction.

Med

Distribution. They're embedded in WRDS/Bloomberg. Even a great dataset needs a route to buyers we don't yet own.

Med

Backfile cost. Real-time is cheap; coding 20 years of history to match theirs is a project in itself.

Low

Tech feasibility. Lowest-risk piece — we already run every layer. The pilot de-risks it for one dataset's cost.