A read of how Basis goes to market, six data problems that motion creates, and a working model that measures each one on a synthetic world shaped like Basis's GTM.
New-logo ARR when the CRM's Type field decides, against the same deals measured on resolved firms.
Basis deploys agents that run long, end-to-end accounting work across CAS and advisory, tax, audit and corporate close, plus an offer for small firms, and it has just launched an end-to-end tax platform.
The buyer is rarely "the firm". Each practice has its own lead partner, workflow and budget. One firm is several buying groups, and growth is expansion from one practice to the next. A CRM built around accounts hides exactly that.
K-1-heavy returns went from 100+ prep hours to about 15–20.
Tax partner, Boulay (paraphrased from basis.ai)Basis is credited for 50% year-over-year growth in the CAS practice.
CAS lead partner, Clark Nuber (paraphrased)Top-25 and top-150 firms. Long cycles, pilots, security reviews.
Enterprise AE · Strategic Account PartnerNew: moving from reactive demand fulfillment to a proactive outbound engine.
Sales Manager · BDR · GTM EngineerAdvisory firms such as Sikich bring Basis into their client work. Partner credit is real money.
Channel Sales Leader · Partner ManagerThe site form asks firm size, use case, ERP and a self-reported "how did you hear about us".
Demand Gen · Growth · Field MarketingAdd Salesforce, Outreach, Gong, Clay, a partner portal and agents that now read and write the same records, and every tool thinks it is the source of truth. That is the job in the GTM Data and Systems Lead posting: define the model, make metrics consistent, keep syncs honest, give agents context they can trust.
Expansions land on duplicate accounts or get typed as new business. Fixing the field alone still leaves +30%. Only resolving the firm first gets it right, and the same fix moves customers from 24 to 20 and customers live in two or more practices from 6 to 9.
Domain matching catches 15 of 22 duplicate accounts with no false merges. Name matching catches all 22 but merges 3 different companies that share a name. The design that works is tiered: auto-merge on domain, send the 10 name-only matches to a person (7 real, 3 not).
13 partner-registered deals were also marked AE-sourced. One written rule, a registration dated before the opportunity was created wins, gives every deal exactly one motion and flags all 13 with no false flags.
150 webhook retries reused their event ID; 67 regenerated it. Dedup on event_id misses every regenerated one; dedup on message_id catches both. Then the definition: sends are attempts, not engagement.
For 10 accounts, the Salesforce last-activity sync stopped weeks ago while calls kept happening in Gong. A drift check comparing the field to its sources flags all 10 with no false flags.
The ontology records each entity's grain, which system owns each field, the metric rules in plain language, and what each agent may do. Reads carry lineage; governed changes go to a person with evidence attached.
sdr_agent task.create applied
revops_agent firm.merge escalated
sdr_agent last_engaged.update blocked
(owned by warehouse)The first version caught every failure with zero false merges, because every duplicate shared a domain. That was the generator agreeing with the detector, not a result. Adding duplicates under a legacy domain from an acquired firm produced the 15/22 and the false merges above, and the tiered design came out of that.
| Field | Owned by |
|---|---|
| opportunity.stage | Salesforce |
| opportunity.amount | Salesforce |
| opportunity.motion | warehouse, declared rule |
| opportunity.is_expansion | warehouse, declared rule |
| firm.firm_id | warehouse, identity resolution |
| firm.last_engaged_at | warehouse, max of Gong and Outreach replies |
| partner.registration | partner portal |
| activity.* | Outreach, Gong (append-only) |
Python standard library only, under a second. Three files: the synthetic world, the model, the report.
git clone https://github.com/sa-basharat-ali/accounting-gtm-data-model cd accounting-gtm-data-model && python3 run.py