The Agentforce launch checklist: T-90 days to go-live

Why a countdown beats a someday-list

Most Agentforce launches slip for the same reason: data cleanup is treated as one big task instead of a sequence with deadlines. Working backwards from your go-live date forces the right order — measure before you fix, fix before you validate, validate before you launch. The checks below follow the four-phase readiness playbook and the six data quality dimensions it is built on.

T-90 days · Scope & baseline

You can’t fix what you haven’t measured. Before any cleanup, establish what the agent will read and how that data scores today.

• List every object and field your Agentforce agents will read — including Description, Notes, and Comments fields retrieval can pull from.
• Run a baseline scan across all six data quality metrics on those objects: Completeness, Conformance, Validity, Timeliness, Duplicate, and PII Exposure rates.
• Record the baseline scores. Every later decision — and your go-live sign-off — is measured against them.
• Flag PII hotspots: high-volume free-text fields where staff paste customer messages are where SSNs and card numbers hide.
• Name a compliance owner who will sign off on the PII numbers before launch.

T-60 days · Remediation

Work findings in risk order, not convenience order. PII first — it’s the only category that can turn into a regulatory incident.

• Clear every SSN and credit-card finding from agent-readable text fields.
• Fill empty key fields the agent depends on — an empty Description forces a vague answer.
• Standardize picklist and reference values so “US”, “USA”, and “United States” stop being three different facts.
• Fix invalid formats in structured fields (emails, phone numbers, IDs) — agents quote them verbatim.
• Merge duplicates on agent-read objects so one customer has one history.
• Update or archive stale records that would feed the agent confident, outdated answers.

T-30 days · Validation

Rerun the same scans and compare against the baseline. Sign-off should rest on evidence, not assurances.

• Rerun the full scan suite on the same objects and fields as the baseline.
• Verify you meet the launch thresholds (table below) on every agent-read object.
• Get written compliance sign-off backed by the PII Exposure numbers.
• Document the before/after scores — your proof the org was launch-ready.

Launch week & after

Data quality decays the moment users keep entering records. Readiness is a cadence, not a milestone.

• Run a final PII sweep on high-volume text fields in launch week.
• Freeze large imports and migrations until after go-live.
• Schedule recurring scans: PII weekly, completeness and consistency monthly, a full six-metric sweep quarterly.
• Review early agent conversations for vague, contradictory, or outdated answers — each one points at a data problem with a known fix.

Go-live thresholds at a glance

Pass criteria for the validation phase. The full reasoning behind each number is on the Salesforce data quality benchmarks page.

What if you fail half the checklist?

That’s the normal starting point — first scans almost always surface problems nobody expected. A failing checklist at T-90 is exactly why the countdown starts at T-90 and not launch week. Prioritize PII findings first, then completeness and consistency, and rerun the scan after each pass. If your date is closer than 90 days, compress the phases rather than skipping the baseline: an unmeasured launch is the one failure mode you can’t recover from quickly.