RevOps Automation Demo · CoinGecko Revenue Operations (L2)

01

Stale-deal & data-hygiene detector

JD: Data Integrity · Pipeline

Runs each weekday: flags deals missing an owner or amount, and deals that have gone quiet past a per-stage threshold (New 7d · Qualified 14d · Proposal 21d · Negotiation 30d), then nudges the owner in Slack. Kills the leaks that silently corrupt the forecast.

Sample output (from the 15 demo deals)

DEAL	STAGE	ISSUE
Veritas Fund	Proposal	stale 48d
Wallet9	Proposal	stale 52d
BlockMetrics	Qualified	stale 31d · no owner
Meridian DAO	Qualified	no owner · no amount
ChainScope	Qualified	stale 16d

// the rule, identical in Apps Script & n8n
const staleDays = { New:7, Qualified:14,
                   Proposal:21, Negotiation:30 };
for (const d of deals) {
  const issues = [];
  if (!d.owner)  issues.push('missing owner');
  if (!d.amount) issues.push('missing amount');
  const days = daysSince(d.last_activity);
  const th = staleDays[d.stage];
  if (th && days > th)
    issues.push(`stale: ${days}d in ${d.stage}`);
  if (issues.length) nudgeOwner(d, issues);
}

02

Agentic lead router

JD: Lead Management — "agentic routing"

// transparent score → capacity-aware assignment
let s = 0;
s += emp>=500?35: emp>=100?25: emp>=25?15:5;
if (['Institutional','Exchange','Fund']
      .includes(seg)) s += 25;
s += /booked a call|redistribution/.test(sig)?25:10;

// only reps with spare capacity + segment/region fit
let pool = reps.filter(r =>
  r.segments.includes(seg) &&
  load[r.name] < r.capacity);
pool.sort(byMostSpareCapacity);
const rep = pool[0]?.name ?? 'escalate';

Each inbound lead gets a transparent 0–100 score (size · segment · intent · source), then is assigned to the best-fit rep who still has capacity — so high-intent institutional leads get a 15-minute first-touch SLA and nobody gets overloaded. The score is plain code; an LLM only labels ambiguous cases.

Sample routing (from the 8 demo leads)

LEAD	SCORE	→ REP	SLA
Northwind Asset Mgmt	85	Aisha	15 min
Castell Securities	85	Priya	15 min
Stratos Exchange	85	Aisha	15 min
Petal Wallet	23	Marcus	1 day

03

AI weekly pipeline digest

JD: Revenue Analytics — "AI-assisted insights to leadership"

Every Monday, aggregate the pipeline into metrics (never raw rows — no PII, no hallucinated deals), hand them to Claude, and post a tight, human-approved digest to leadership. It writes the "so what," which is the part leadership actually reads.

// model sees aggregates only
{ open_pipeline_usd: 941400,
  target_usd: 600000, coverage_ratio: 1.57,
  pipeline_by_stage: { Proposal:331600,
    Negotiation:488000, ... },
  stale_deals: 5, deals_missing_fields: 2 }

📊 Claude's draft (sample)

Healthy on volume (1.57× coverage) — quality is the risk, not quantity.

$331.6k of Proposal pipeline is aging — 3 of 5 stale deals sit there; proposals are stalling after send.
$488k concentrated in Negotiation across 4 institutional deals — strong, but single-rep concentration risk on Priya.
2 deals missing owner/amount will distort the forecast until fixed.

Top 3 actions

AE Re-engage the 5 stale deals, Proposal-stage first.
RevOps Require owner + amount at Qualified to stop forecast leakage.
Leadership Rebalance Negotiation concentration on Priya.

Why it's built this way

Deterministic where it counts. Scoring & assignment are auditable code — an LLM can't silently overload a rep or drop a lead. The model only labels and narrates.

No raw data to the model. The digest is built from aggregated metrics only — no PII, no invented deals.

Human-in-the-loop. Claude drafts; a person approves before it reaches leadership. Same rules in Apps Script and n8n.

Prepared by Edward Tay · for the CoinGecko Revenue Operations Associate (L2) role · Jun 2026 · edwardtay.com · Edwardtay7@gmail.com · sample data is synthetic.