Company context
A B2B engineering services firm, engineering-led digital transformation across devices, data and digital. No ABM program, no Apollo instance and no account research process when I started.
My role
Program owner and systems architect. I own the structure and build the tooling that runs it.
- Six pillars over a shared governance layer of RACI, SLAs and stage definitions, never forked
- Dental as pilot vertical, one technical buying centre, eight segments, tiered 20 / 50 / 100+ with promotion earned by engagement
- MedTech sequenced behind dental, so nothing contends for domain reputation, LinkedIn budget or Apollo credits
- Dossier cut three times. v1 too generic to write from, v2 at 41 fields nobody would fill, shipped at 12 fields with a 45 minute hard stop
How it runs
Six executable skills rather than a slide deck. AI does research volume and drafting. It does not decide what is true, and it does not define done.
- No source URL means the field stays blank, never inferred
- An observation needs one signal, a diagnosis needs two
- Completion is an output, not a form. Research is done when the first sentence of the email can be written, which is what holds an account at 40 minutes instead of two days
- Cluster inheritance writes segment truth once, including a "do not say" column, removing about a third of research volume
Integrations
Apollo via MCP · Context Center · custom fields and sequences · HubSpot · LinkedIn Campaign Manager · Claude Code with six custom skills · Google Drive MCP · GA4 and GTM
Two limits found by inspection, not documentation: no field-creation endpoint in the Apollo MCP, and updating a contact with label names wipes list membership instead of appending to it.
I audit the vendor's AI as well as my own.
Apollo's Context Center had auto-generated a generic data-engineering profile. Dental named once in a list of eight industries, anti-personas null, proof points quoting a procurement engagement meaningless to a dental buyer. Every AI-written outbound email drew from it. Rewritten field by field.
Results
- 1,000+ accounts researched, 50 selected
- Research per account from ~2 days to ~40 minutes
- About a third of research volume removed by cluster inheritance
- Apollo, HubSpot and LinkedIn infrastructure standing where nothing existed three months earlier
- Plan endorsed by the VP of Sales and Marketing