DEF 14A filings read by encoded rules, not analysts.
We encode your study’s rules — disclosed trumps plan tables, targets never mix with actuals — extraction runs across the coverage universe, every value cited.
Official services partner of the platforms defining AI
NVIDIA
Anthropic
The proxy isn’t the problem. The reading is.
From the partners, analysts, and research directors we have sat with: boutique exec-comp advisors and executive compensation benchmarking teams covering dozens to thousands of public companies — not shopping for software, drowning in the reading.
150-page proxy statements read by hand
Board and pay data comes out of DEF 14A filings one company at a time, some running 200 pages, while the study waits on the reading.
150 granular extraction rules per study
Each pay study needs roughly 150–160 fine-grained field rules, derived by hand from a high-level requirements document before extraction can even start.
Disclosed figures trump plan tables
The rulebook says non-disclosed must never default to zero — salary, STI, and LTI targets land in the workbook with per-field confidence flagged.
Target vs actual comp values
Analysts need target values for salary, STI, and LTI — mixing in actuals corrupts the study, and the rebuild happens every single study.
Same prompt, different numbers
AI outputs that vary run to run cannot be defended to clients — consultants need reproducible numbers with a citation behind every value.
One analyst, one CD&A, one company at a time.
What the manual way looks like at a consultancy whose studies span dozens to hundreds of proxies, with coverage universes up to the Russell 3000. Your numbers will differ — the calibration study puts figures on yours before anything scales.
From anonymized engagements — exec-comp consultancies and benchmarking teams, studies spanning dozens to hundreds of proxies
We encode your study’s rules.
No new system for your team to learn. The extraction rules your senior consultants carry become how every filing gets read.
- 01
Calibrate against ground truth
We run the encoded rules against your past hand-built studies and match the analysts’ answers line by line before anyone trusts the output.
- 02
Encode the study’s rules
Disclosed figures trump plan tables, non-disclosed never defaults to zero, targets never mix with actuals — roughly 150–160 field rules, documented and yours to keep.
- 03
Exception-only review
Extraction runs across the whole coverage universe; low-confidence fields queue for an analyst, and approved rows lock and carry into the next study.
The data subscriptions sell everyone the same numbers. Your study’s rules aren’t in the package.
The data subscriptions sell everyone the same pre-packaged numbers. The dev tools hand an engineer raw filings. Neither encodes your study’s rules — disclosed over plan tables, target never mixed with actual. We encode those.
Our AI does the reading of every DEF 14A, CD&A, and 8-K in your coverage universe; your encoded rules do the judging — which fields count, which table wins a conflict, what gets flagged for an analyst.
The review surfaced real violations a senior consultant had not flagged manually.
Led by Peter Enestrom.
Asked by comp consultants and research teams.
The straight answers, before you book anything.
A DEF 14A is the definitive proxy statement a public company files with the SEC before its annual shareholder meeting. It discloses board nominees, the matters up for a vote, and executive compensation — including the CD&A narrative and the summary compensation table. It is the primary public source for named-officer pay data.
The Compensation Discussion and Analysis is the narrative section in which the compensation committee explains its pay decisions: the pay philosophy, the peer groups, and how salary, incentives, and equity were set. It is where target values and plan structures are described, which is why comp studies read it alongside the tables.
Executive pay appears in the CD&A narrative and in the required tables: the summary compensation table, grants of plan-based awards, outstanding equity awards, and option exercises and stock vested. Director compensation has its own table. The figures for the named executive officers sit across these sections, not in one place.
Target values are the amounts a plan is designed to pay at expected performance — base salary, target bonus, and target award values. Actual values are what was ultimately paid or realized. Benchmarking studies are built on targets; mixing actuals into a target-based study corrupts the comparison, so the two must be extracted as separate fields.
Yes. DEF 14A filings are public on EDGAR, and the fields a study needs can be read directly from the filings themselves. The hard part is not access — it is applying a study’s rules consistently across hundreds of filings, so that every extracted value is reproducible and cited to its source location.
Start with one workflow.
Tell us where your team loses hours. We will come back with a straight answer on whether AI can help, what it would take, and what it would pay.