Every downstream system in a revenue organization - routing, sequences, audiences, forecasting, the Monday report - consumes the CRM and inherits its condition. Three jobs decide that condition. Scoring decides which records deserve attention. Verification decides which contacts can be reached without burning the sending domain. Hygiene decides whether the records are true at all: deduplicated, suppressed where they must be, standardized, current, and carrying the provenance that makes any field trustworthy at a glance. Treated as three projects, they fail on schedule; treated as one standing system, they turn a CRM from a liability into the asset the rest of the stack assumes it is.
This is the pillar for the CRM and RevOps cluster. It covers fit scoring that recomputes from live data, the verification verdicts and the action each one deserves, dedupe on confidence tiers and suppression of the records you must never work, the five standing hygiene missions, the provenance rules that make trust inspectable per field, and the handoff into routing. It reflects how the verification agent works a CRM.
Three jobs, one system
The jobs interlock in a fixed order. Hygiene comes first, because scoring a duplicate produces two scores for one company and verifying a record that should have been suppressed spends credits on a customer. Verification comes second, because a fit score is only actionable when the contact behind it can be reached. Scoring comes last and reads both: a verified contact at a high-fit account with a fresh signal is the record that deserves a person's next hour. Run in that order on a schedule, the three produce a CRM where the top of any sorted view is genuinely the best next action, which is the entire purpose of the system.
Fit scoring against a live ICP
Static point schemes - ten points for a director title, five for a target industry - encode one quarter's opinion and decay from the day they ship. A fit score worth acting on has three properties. It reads an executable ICP: filters and weights expressed as data, derived from closed-won, closed-lost and churned history, the method in ICP definition with live data. It is computed from enriched fields rather than from whatever happened to be typed into the record, which means the waterfall runs before the score does. And it is recomputed on a schedule, because size, stack, hiring and funding all move, and a score frozen at import time is a score about a company that no longer exists.
Two components belong in the score and one does not. Fit - how well the enriched firmographics, technographics and buying committee match the ICP - is the stable component. Momentum - the converging signals from the signals pillar - is the timely one, decayed on each family's own clock. Engagement with your own marketing is a separate score with a separate purpose, and blending it into fit is how a scoring model learns to love whoever clicked a newsletter. Validate quarterly: component analysis against what closed, weights adjusted with the evidence attached, the model versioned and the change noted where reps see it. The mechanics are in lead routing and scoring that reflect reality.
Verification: verdicts and what to do with each
| Verdict | What it means | Action |
|---|---|---|
| Deliverable | The mailbox exists and accepts mail | Eligible for sequences and audiences |
| Catch-all (accept-all) | The domain accepts everything, so the mailbox cannot be confirmed | Send only under a per-domain cap, or after a secondary confirmation |
| Undeliverable | The mailbox rejects mail or does not exist | Suppress from sends; re-run the waterfall for a replacement address |
| Unknown | The server did not answer in time | Re-verify later; never send on an unknown |
| Role or disposable | A shared or throwaway address | Exclude from person-level outreach |
| Phone: connected | The number rings and is assigned | Eligible for call notes and dialer loads |
The rule that protects the domain is short: only deliverable addresses enter a sequence, catch-alls enter under a cap, and everything else is suppressed or re-verified. The temptation to include the risky tier to make a list look bigger is how a clean sending domain inherits a bounce rate, and the mechanics of what that costs are in email verification and deliverability for outbound. Verification also has a deadline: addresses decay as people move, so active segments re-verify on a monthly cycle and any list older than a quarter re-verifies before it sends.
1 in 3B2B contacts that change role, company or address within a year - the reason verification has a cycle, never a checkbox
Dedupe and suppression
Duplicates arrive continuously: form fills with a different email, list imports with a different company spelling, a partner's CSV, a rep creating what already existed. Dedupe runs on confidence tiers. A pair matching on domain plus a normalized name plus a verified email is a high-confidence merge that executes automatically, with both prior states logged so any merge is reversible. A pair matching on a fuzzy name alone queues for a person. The reverse - auto-merging the ambiguous pair, queuing the obvious one - is the mistake that loses a sales team's trust in one afternoon, because the first bad auto-merge a rep discovers is the last time they believe the CRM.
The standing hygiene missions
| Mission | Cadence | Deliverable |
|---|---|---|
| Dedupe sweep | Weekly | High-confidence merges executed and logged; ambiguous pairs queued |
| Enrichment pass | Weekly on new records, monthly on active segments | Empty fields filled with provenance; conflicts flagged, observations never overwritten |
| Verification refresh | Monthly on active segments, before any send on older lists | Verdicts updated; undeliverables suppressed; replacements found |
| Standardization | Weekly | Countries, titles, industries and phones normalized; the integrations that write drift patched |
| Staleness audit | Monthly | Records untouched past thresholds surfaced with a recommended action: re-enrich, re-verify, archive |
Each mission is deliberately narrow - one job, one small diff, computed exactly - because narrowness is what makes it trustworthy enough to run unattended. A mission that only normalizes country fields earns autonomy fast; a broad "clean everything" automation stays gated forever because nobody can reason about its blast radius. The mission set in daily operation is the subject of CRM hygiene with agents, and the ninety-day arc on a neglected instance is walked through in the CRM cleanup quarter.
Inspectable trust: provenance on every field
The property that ties the whole system together is that trust is inspectable at field granularity. Every value carries its source, its date and its confidence, so a rep reading "employee count: 240" can see it came from an observed source in August rather than an import in 2023, and a mission deciding whether to refresh a field can read the date instead of guessing. Three rules follow. Observed beats inferred: a value a person entered from a real conversation outranks any vendor's estimate, and enrichment fills blanks and flags conflicts rather than overwriting. Writes are idempotent: a re-run of any mission produces the same state rather than a second copy. Destructive operations gate: merges, archives and deletions park for approval until that class of action has earned autonomy on its track record. The same three properties are what make the CRM enrichment automation safe to schedule against a live instance.
Routing and the handoff
Scoring exists to route. A verified, high-fit, high-momentum record should reach the right owner within minutes, with the context that produced the score travelling with it: which signal fired, which fields qualified it, what the recommended first touch is. Latency is the metric - the response-time research that has shaped this field for over a decade still holds - and the routing log is the audit trail, because "why did this lead go to X" carries organizational heat and deserves a record rather than a recollection. The routing rules that survive contact with a real team are in the routing and scoring guide.
How AstroFabric does it
The verification agent owns scoring and lead verification: fit scores computed against your executable ICP from enriched fields and recomputed on a schedule, bulk email and phone verification with the verdicts above, dedupe on confidence tiers with both prior states logged, and suppression of customers, competitors and the records you name. The enrichment agent runs the enrichment pass with provenance on every value and never overwrites an observed field with an inferred one. Together they run the five hygiene missions as standing missions against any connected CRM, each producing a small diff and a digest where your team already talks.
Every CRM write passes an approval gate, merges and archives stay gated until that class of action has earned autonomy, every run is bounded by a credit ceiling, and the audit log records the source and date behind every field the agents touched. The AI agents for outreach data page maps the verification motion, and agentic AI for targeting maps the scoring side.
Go deeper in this cluster
- CRM enrichment automation: the complete guide - How to keep a CRM enriched without a person running exports: the three triggers, the field plan, the write rules that make it safe against a live instance, provenance inside the CRM, and the numbers that say it is working.
- CRM hygiene with agents: the maintenance nobody budgets - Dedupe, enrichment, standardization and staleness - as standing missions instead of quarterly projects: the mission set, the confidence thresholds, and the write-safety underneath.
- Lead routing and scoring that reflect reality - Routing is latency engineering, scoring is a model that needs validation: the minutes-not-hours routing bar, live-signal score recomputation, and the quarterly honesty ritual most teams skip.
- Use case: the CRM cleanup quarter - A neglected CRM goes from liability to asset in ninety days: the damage audit, the backlog-clearing sweeps, the threshold tuning, and the handover to standing missions that keep it clean.
Frequently asked questions
What is lead verification?
Checking that a contact can actually be reached before anyone acts on it: email verification at the mailbox level, phone connectivity where the source supports it, and the dedupe and suppression checks that confirm the record should be worked at all. Each check returns a verdict with a defined action.
How should a lead score be built?
From an executable ICP derived from closed-won, closed-lost and churned history, computed on enriched fields rather than typed ones, with a fit component and a decayed momentum component kept separate from marketing engagement. Recompute it on a schedule and validate it quarterly against what closed.
What should happen with catch-all email addresses?
Send under a per-domain cap or after a secondary confirmation, never at full volume. A catch-all domain accepts everything, so the mailbox cannot be confirmed, and including the whole tier to make a list look bigger is how a clean sending domain inherits a bounce rate.
How do automated merges avoid destroying records?
Confidence tiers: high-confidence pairs (domain plus normalized name plus a verified email) merge automatically with both prior states logged and reversible; ambiguous pairs queue for a person. Thresholds tune against the queue’s verdicts over time, and the reverse ordering is never allowed.
What is a suppression list?
A maintained list of records that must never be worked or paid to reach - customers, open opportunities, competitors, employees, opt-outs - applied before scoring, before sequences and before ad audiences. It is the cheapest hygiene mission and the most visible when it is missing.
What does provenance on a field mean?
The source, date and confidence recorded on every value, so trust is inspectable per field: a rep can judge a number at a glance, a mission can decide whether to refresh it, and observed values can be protected from being overwritten by inferred ones.
Sources
- Google - Email sender guidelines (the bounce and complaint thresholds verification protects)
- Harvard Business Review - The Short Life of Online Sales Leads (the response-latency evidence)
- Gartner - data quality research
Every playbook on this blog ships as a runnable mission.
Open a workspace and the playbook library is waiting - describe the outcome and the agents carry it end to end, on your plan's monthly credits.