The GTM Tech Stack: A Reference Architecture

The seven layers

How data should flow

The flow matters more than the vendors. Get the direction right and you can swap almost any component later; get it wrong and every swap is a migration.

1. 01
   Signal fires

   A company posts a relevant role, installs a competitor's tool, raises a round, or a known user hits a high-intent page. The signal carries an identifier — usually a domain — and a timestamp.

2. 02
   Enrichment resolves the account

   Waterfall across providers to build a complete record at a known cost per row. Cache with provenance so you never re-buy the same fact — see waterfall enrichment.

3. 03
   Orchestration applies the logic

   Fit filters, scoring, deduplication against existing accounts, and suppression against open opportunities and recent touches. This is where most plays either become precise or become spam.

4. 04
   Everything lands in the warehouse

   The signal, the enriched record, the score, and the decision — with timestamps. This is the durable store and the only place the full history survives a vendor change.

5. 05
   Qualified records resolve into the CRM

   Only what passes the threshold. The CRM is what sales sees, not a landfill for enrichment output. Records below the threshold stay in the warehouse.

6. 06
   Engagement executes

   Sequence, task, alert, or ad audience — with personalisation generated from the enriched record and validated before it sends.

7. 07
   Outcomes flow back

   Reply, meeting, and opportunity data returns to the warehouse and joins the original signal. Without this hop the scoring model has no feedback loop and never improves.

What to buy when

The most common sequencing error is buying orchestration first because it demos well. Orchestration coordinates other layers; with nothing to coordinate it is an expensive spreadsheet.

Three failure modes to design out

Silent failure

The default failure mode of this entire category. An enrichment job stops returning results, sequences keep sending with empty merge fields, and the first person to notice is a prospect. Alert on absence — a job that stops running produces no error at all — and reconcile record counts daily.

The CRM as a landfill

Pushing every enriched account into the CRM because it is where the tooling points by default. Within a year the database is full of accounts nobody has ever contacted, reporting is inflated, and reps do not trust their own views. Set an entry threshold and hold everything below it in the warehouse.

Tool-native data

History accumulating inside a vendor with no path out. Three years later the company stays on a tool it dislikes because the engagement history cannot be extracted usefully. Test the export path during the trial, not during the migration.

A worked reference stack

A mid-market B2B company, roughly $8M ARR, outbound-led, 12 revenue-facing staff:

• Signals — hiring and funding feeds, technographics, plus first-party site de-anonymisation.
• Enrichment — a three-provider waterfall with caching and a per-field TTL.
• Orchestration — one platform owning the fit filters, scoring, dedupe, and suppression logic.
• Warehouse — everything lands here with timestamps; scoring models built on top.
• System of record — the CRM, receiving only records above the entry threshold.
• Engagement — one sequencer, personalisation generated upstream and validated before send.
• Observability — absence alerts on every scheduled job, daily count reconciliation, alerts routed to a named person.

That stack runs roughly $4K–$6K per month excluding CRM seats and is operated by one GTM engineer. The layers a company genuinely needs are determined by how many plays it runs, not by its revenue — and the broader RevOps tool stack this sits inside is covered in the best RevOps tools.