Salesforce fraud decision sync is the part of a fraud stack that support actually feels. Your policy engine answers allow, step-up or block in milliseconds, and then that answer lands in a webhook log nobody opens while a ticket is live. The customer writes in, the agent sees an account with no notes, and the first reply is a request for a screenshot.
The failure is quiet and it costs twice. Support spends time reconstructing a decision your system already made, and the customer repeats the action that triggered it.
This post covers the plumbing: which fields to write on the account and lead objects, what belongs in each one, and how to push the payload without building a sync service yourself.
Start with the three values support reads first
An agent working a ticket has seconds, not minutes. Three values carry almost all the context.
Decision and the suggested step
Prynt Control answers use-case decisions — signup, free_trial, bonus, login, checkout, extra_seat, sms, paywall — with allow, step-up or block, plus a suggested step: captcha, verify_email or card_or_passkey. Write them into two fields. “Step-up” alone is a riddle; “step-up, verify_email” is an instruction the agent can repeat to the customer.
Reason code as a stable token
Keep the reason code short and stable, and never splice user input into it. Analysts group by reason code, and a value that changes shape as soon as an email or a clicked URL enters it breaks every report built on top. Long detail goes in its own field; the code stays a code.
Trust level and risk score
deviceTrust on login returns trusted, new or risky — a clean picklist. Pair it with riskScore in a number field so the team can sort and filter list views. Support reads the trio of decision, reason code and trust level; the number is for whoever asks how close the case was to the line.
Plan the field map before you touch the API
Decide the shape once, then reuse it on both objects so a lead that converts keeps its history:

| Field | Object | Type | Value |
|---|---|---|---|
| Decision | Lead, Account | Picklist | allow, step-up, block |
| Suggested step | Lead, Account | Picklist | captcha, verify_email, card_or_passkey |
| Reason code | Lead, Account | Text | the code from the decision |
| Trust level | Lead, Account | Picklist | trusted, new, risky |
| Risk score | Lead, Account | Number | riskScore |
| Last decision at | Lead, Account | Date/Time | event time |
| Request ID | Lead, Account | Text, external ID | requestId |
Keep one long-text field per object for the reasons array — {{trigger.reasons | json}} renders straight into it — and resist filling descriptions with prose. Fields that support cannot filter on are fields support will ignore.
Push it with a workflow, not bespoke glue
Prynt Automate triggers on fraud decisions and events and writes to Salesforce, HubSpot, Pipedrive, Slack, Teams, signed webhooks, Sheets, n8n and Zapier, with retries and a run log. The variables you need are fixed: {{trigger.decision}}, {{trigger.riskScore}}, {{trigger.visitorId}}, {{trigger.requestId}}, {{trigger.url}} and {{trigger.reasons | json}}.
Two details separate a sync from a mess. First, map {{trigger.requestId}} to an external ID field: when a retry replays a run, it updates the same record instead of creating a twin. Second, send visitorId and requestId even though no human reads them — they are what lets an engineer join a Salesforce record to the original event when someone asks a harder question a month later. The reference lives at our docs.
Run it in mirror mode before you enforce anything
Control starts with a 14-day mirror period: decisions are recorded and nothing is blocked. Use that window to validate the field mapping rather than to debate thresholds. If a picklist receives a value nobody planned for, or a step-up arrives with no suggested step, you want to find it while the answer is still allow.

Keep Salesforce a mirror, not the source of truth
Policy belongs to your server. Prynt never blocks by itself: your code reads the decision and decides what to do with it. Salesforce is a read model for humans, so an agent editing the decision field should open a request in your own review queue instead of quietly rewriting history. Write only what support needs — the decision, the code, the trust level, the score and the identifiers. Control stores phone numbers and cards only as keyed hashes, and card data has no reason to travel to CRM. The platform overview explains how the products share one agent and one bill, and payment fraud detection covers the checkout decisions that benefit most from this pattern.
FAQ

Should the decision live on the lead or the account?
Both, with identical fields. Signup and trial decisions arrive before a lead converts; login and checkout decisions usually arrive after. Writing the same values to both objects means the history survives conversion without a migration.
Can support override a block from Salesforce?
Not with a field edit. Treat any override as a request into your own system, where it can be reviewed, logged and replayed. A picklist changed by hand looks like a decision and is not one.
Do we have to send personal data to Salesforce?
No. The decision, reason code, trust level, risk score, requestId and visitorId are enough for support to work a ticket. Keep raw card data and unmasked phone numbers out of CRM entirely.
If you want to see the exact payload your CRM would receive before you map a single field, start at pryntid.com.
Try it free
Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.