Objective

An SDR spends ~70% of pre-call time on research: who is this account, what did we already talk about, what similar customers have we won? In this recipe an account-research agent reads the CRM row, retrieves similar-shape wins from your historical close patterns, and outputs a structured briefing — in one SQL call.

Step 1: CRM accounts + activity + close patterns

CREATE TABLE IF NOT EXISTS recipe_sl_accounts (
  account_id    INTEGER PRIMARY KEY,
  name          TEXT,
  industry      TEXT,
  arr_band      TEXT,            -- '<1M' | '1-10M' | '10-100M' | '100M+'
  owner_email   TEXT,
  notes         TEXT
);

CREATE TABLE IF NOT EXISTS recipe_sl_activity (
  activity_id   INTEGER PRIMARY KEY,
  account_id    INTEGER,
  kind          TEXT,            -- 'call' | 'email' | 'demo' | 'reply'
  summary       TEXT,
  happened_at   TIMESTAMP
);

CREATE TABLE IF NOT EXISTS recipe_sl_wins (
  win_id        INTEGER PRIMARY KEY,
  customer      TEXT,
  industry      TEXT,
  acv_usd       DOUBLE,
  use_case      TEXT,
  embedding     VECTOR(384)
);

Step 2: Seed

INSERT INTO recipe_sl_accounts VALUES
 (101,'Northwind Logistics','Logistics','10-100M','jane@yourco.com','Mid-sized 3PL, scaling pain on returns processing'),
 (102,'BlueRiver Health',    'Hospital','100M+','sam@yourco.com','Multi-hospital system, RAG-over-policies pain'),
 (103,'PixelForge Studios',  'Software','1-10M','jane@yourco.com','30-engineer dev shop, ML team forming');

INSERT INTO recipe_sl_activity VALUES
 (1,101,'call', 'CIO: returns auto-triage is the #1 pain. Currently 30min/RMA, 200 RMAs/day.','2026-05-19 10:00:00'),
 (2,101,'email','Sent ROI calc spreadsheet. No reply yet.','2026-05-22 14:00:00'),
 (3,101,'reply','CIO: numbers look right, what about pilot scope?','2026-05-26 09:15:00'),
 (4,102,'demo', 'VP Innovation watched RAG-over-policies demo. Pushback on PHI boundary.','2026-05-21 11:00:00'),
 (5,103,'call', 'Founder/CTO: needs in-DB AI, currently glued together Pinecone + LangChain.','2026-05-15 16:00:00');

INSERT INTO recipe_sl_wins (win_id, customer, industry, acv_usd, use_case) VALUES
 (1,'Acme 3PL','Logistics',  90000,'Replaced manual returns triage with a SQL agent — 30 min/RMA → 5 min, 200 RMAs/day. Pilot to closed in 5 weeks.'),
 (2,'Helios Hospital','Hospital',180000,'In-DB clinical Q&A over policies + EHR — never crossed PHI boundary. Champion: CIO. Procurement took 6 weeks.'),
 (3,'Riverbed Devops','Software',62000,'Incident triage agent — replaced a LangChain + Pinecone stack. Champion: VP Eng. Pilot in 2 weeks, closed in 4.'),
 (4,'Civic Insurance','Insurance',220000,'Claims triage agent — same shape as Acme 3PL, larger volume. Champion: VP Innovation. Took 10 weeks (board approval).');

Step 3: Embed past wins (by use case)

UPDATE recipe_sl_wins SET embedding = EMBED(industry || ' — ' || use_case);

Step 4: Ground truth — similar-win retrieval

SELECT customer, acv_usd, COSINE_SIMILARITY(embedding, EMBED('logistics company with returns-triage pain')) AS sim
FROM recipe_sl_wins ORDER BY sim DESC LIMIT 2;

Expected: Acme 3PL ranks first; Civic Insurance second (same shape, different industry).

Step 5: Brief me before the call

SELECT AGENT_RUN(
  'aidb-assistant',
  'Brief me for a 2pm call with account 101 (Northwind Logistics). Use execute_query to pull the account record and the activity log, sorted most recent first. Use rag_search on recipe_sl_wins to find the 2 most similar past wins by shape (same pain, same buyer-title, similar industry). Output: a one-page briefing with (a) what we know about this account, (b) the last 3 touches and the most recent customer signal, (c) the 2 most similar wins with ACV + cycle-time, (d) recommended next step.'
) AS briefing;

Cleanup

DROP TABLE IF EXISTS recipe_sl_accounts;
DROP TABLE IF EXISTS recipe_sl_activity;
DROP TABLE IF EXISTS recipe_sl_wins;

Use it from your agent

REST/SDK: schedule the briefing on every meeting in your CRM — generate it 10 min before each call.
MCP: call query from your sales console with the Step-5 SQL.
Why in-DB: the briefing is only useful if it knows your pipeline state. A generic AI sales assistant doesn't know who on your team touched the account this quarter. SynapCores does — same DB.

Key Concepts Learned

The briefing pattern is account state (SQL) + similar-shape wins (RAG) — the agent fuses both in one call.
AGENT_RUN is composable: schedule it on a CRON to brief the entire team's day every morning.

Agentic Sales Account Research