Security certification body
≈ $10–15M revenue (approx.) · ~150 employees · ~100,000 CRM contacts
CRM contact database refresh
Roughly one hundred thousand contacts validated, deduplicated and enriched through a tiered waterfall of trust. Every changed value carries a source and a confidence tier, and a re-run kit stays with the customer.
- Organization
- Security certification body with a seventeen-year-old CRM
- Data
- Roughly one hundred thousand contacts, enriched once and never again
- Runs on
- Kirk-operated infrastructure, then handed back
- Scope
- Validate, deduplicate, enrich, enhance; three weeks
Where they started.
An event list pulled for one city came back mostly wrong: people who had moved, titles from years ago, work addresses for people who now work from home. The database had been enriched once by two successive vendors and then left alone. The alternative under consideration was months of manual lookups offshore.
What was built.
- A defined waterfall of trust: cheap, authoritative verification of every email and phone first; deduplication with exact and fuzzy matching; enrichment only where records fail or matter, starting with the subscription the customer already paid for.
- The model plays a small, bounded part: adjudicating junk and duplicate records, reading public web pages with the URL attached as evidence, and normalizing free-text titles.
- A residence-metro inference held in its own field, so event outreach can target where people live without overwriting the company address.
- Every changed value carries a source and a confidence tier; the pipeline is delivered as a re-run kit the customer keeps.
What changed.
- An import-ready update file with a trust tier on every value, in three weeks.
- Duplicates and junk identified with a merge file rather than merged blindly.
- Third-party costs shown line by line, so the margin is visible.
- Repeatable next quarter by pressing a button.
Contact data is personal data. Each service sees only the field it needs, every external call is logged per record, and all copies are deleted after acceptance.
Kirk infrastructure, delivered back to the customer
Components involved.
More like this.
Technology consultancy
≈ $8M revenue (approx.) · Boutique consultancy · ~25 consultants
- Use Case:
- Content production agent
- Results:
- Sales decks, one-pagers and campaign copy from a private agent with the brand system loaded, running on the company's own inference, with skills shared across the team rather than pasted between chats.
Sales team
≈ $120M revenue (approx.) · 35-person sales team · one CRM
- Use Case:
- CRM-connected agent
- Results:
- Pipeline questions, follow-up drafts and account summaries with the CRM connected through each user's own OAuth credentials, so nobody's agent can reach a deal its user cannot see.
Have one like it?
Every spotlight started as a conversation about a process nobody liked doing, under a data-locality constraint.