Test reliable firmographic enrichment on a narrow task.
From form submission
to sales-ready context.
A Contact or Demo request is already high intent.
The opportunity was not to generate another lead, but to understand it, qualify it and get it to the right person with enough context to act.
97%company matching + complete firmographics
<2 MINsubmission → assignment
<$0.26per run
Updated
01ASK LESS. KNOW MORE.
Better data should not mean
more form fields.
Qualification originally depended on information visitors had to type themselves. That added friction and created inconsistent data, especially company names, where one organisation can arrive as “Acme”, “A.C.M.E.” or “ACME Inc.”. The better route was to ask only for what the visitor uniquely knows, then resolve the rest elsewhere.
BEFORE / USER-DECLARED
WORK EMAIL ✓
COMPANY NAME ✓
COUNTRY ✓
INDUSTRY ✓
COMPANY SIZE ✓
JOB TITLE ✓
VOLUME / MONTH 500–1,000 ▾
MESSAGE ✓
More typing · inconsistent naming · duplicate risk
→
AFTER / MINIMUM NECESSARY
WORK EMAIL ✓
VOLUME / MONTH500–1,000 ▾
MESSAGE ✓
EMAIL DOMAINCompany resolution
CRM + WEBFirmographics
MESSAGEIntent signals
Less friction. Better context.
WORK EMAIL→real-time validation+DOMAIN→canonical company resolution
02THE BUILD PATH
The system didn’t start
as a system.
It started with a focused proof of concept: could one agent reliably complete missing firmographic data? Each iteration exposed what should be specialised, controlled or connected next.
Split responsibilities to make prompts, outputs and testing more reliable.
AI interprets. Code cleans, prepares, validates and applies deterministic rules.
DATACODEAGENTCODERULE
Reuse CRM context, connect specialised outputs and converge validation, research, AI and rules into qualification and routing.
Confidence, refresh logic, rules and agents continue to evolve from real usage.
AI performs the tasks that need interpretation.
Code controls the rules that should stay predictable.
03THE INFORMATION ROUTE
Watch missing context
find its way in.
A submission starts with only what the visitor should have to provide. The system validates, reuses, interprets and researches in parallel, then converges those routes into one sales-ready context.
VALIDATEREUSERESEARCHINTERPRETDECIDE
ILLUSTRATIVE PROCESSING / SAMPLE REQUESTSTARTING
WHAT WE ASK
WORK EMAILalex@acme.exampleVALIDATING…
VOLUME / MONTH500–1,000
MESSAGEWe need to automate customer onboarding before January…
KNOWN DATA?WAITING
CompanyYESREUSE → ACME Inc.
IndustryYESREUSE → Software
HQ countryNOSEARCHING…
RevenueNOSEARCHING…
Job titleNOSEARCHING…
WHAT WE KNOW
EMAILValid ✓
COMPANYACME Inc. ✓
HQ COUNTRYFrance ✓
REVENUERange found ✓
INTENTB2B · structured ✓
QUALIFICATIONMQL
ASSIGNMENTRight owner ✓
SALES CONTEXTHook ready ✓
Illustrative processing: values shown are sample dataREAL SYSTEM: <2 MIN
UNDER THE HOODForms & CRM · email validation · workflow orchestration · specialised agents · web research · deterministic rulesCurrent implementation: HubSpot · LangDock · ZeroBounce
03TURN LANGUAGE INTO SIGNALS
The message is not a note.
It is qualification data.
Structured fields only tell part of the story. The free-text message is read against commercial dimensions so useful intent can join the rest of the scoring model instead of staying buried in a text field.
04THE HARD PART
Research where data
providers stop.
The objective was not simply to fill fields. It was to reach a usable level of confidence and completeness while keeping time and cost low.
ANNUAL REVENUE
Research first.
Estimate only when needed.
A dedicated sub-agent searches for reported revenue. When no reliable figure is available, a coded fallback estimates a range using weighted signals including industry, country, company age and size.
COMPANY MATCH + COMPLETE FIRMOGRAPHICS97%vs ~40% on a comparable sample
05DON'T ENRICH TWICE
Reuse what the system
already knows.
Useful information is propagated between Contact and Company records. Known data takes the reuse route; only missing, uncertain or stale context takes the enrichment route, which reduces cost, duplication and unnecessary research.
KNOWN DATA / YES
Reuse.
Contact and Company context already in the CRM is propagated and reused before any new research is triggered.
or
MISSING / STALE DATA
Enrich.
Only missing or refresh-worthy information is sent down the research route, then merged back into the canonical record.
06BUILT TO STAY USEFUL
Freshness matters as much
as enrichment.
Firmographics age. The architecture therefore stores metadata that can later decide when a company deserves another look instead of treating enrichment as a one-off event.
TIMESTAMPWhen was this learned?
CONFIDENCEHow much should we trust it?
COMPANY STAGENew · growing · mature · decline…
REFRESHTrigger research when context becomes stale.
IMPACT
Less than two minutes
between intent and action.
97%company matching + complete firmographics
<2 MINfull process to assignment
<$0.26per workflow run
The system combines AI-assisted research with deterministic business rules. The goal is not AI for its own sake: it is faster, more consistent commercial decisions with better context.