Fraud & Quality Shield + Rewards
Pay real people. Not bots.
The moment a public survey offers a gift card, it becomes a target. The Shield scores every response inside Salesforce before anyone is paid. Rewards then sends a Tremendous reward only to the responses that pass, inside the budgets and approval rules you set.
Built for research teams, nonprofits and programs that pay respondents for their time, on the surveys you already run with AutoSurvey.
Demo video
A short walk-through: a response is scored, a doubtful one is held for review, and an approved one is paid.
The problem
An incentive is an open invitation
Pay for responses and you pay twice for every fake one: once in rewards, and again in the hours spent cleaning the data before anyone can trust it.
Bots
Scripts fill out public survey links at a speed no person can match, and every completion looks like a claim.
Repeat submitters
One person answers again and again from the same browser or the same email address, and collects a reward each time.
Low-effort answers
Speeders, straight-line grids and keyboard-mash text skew your results even when a real person sent them.
Incentive fraud
When a reward goes out on completion, a bad response is paid before anyone has looked at it. That money does not come back.
How the Shield works
One score, one decision, every response
When a response is submitted, the Shield runs a set of deterministic rules on the server. Each rule that fires adds a named reason and a weight. The total is a risk score from 0 to 100, and the score lands in one of four decisions. Out of the box, Flag starts at 30, Hold at 50 and Block at 80. You can change all three.
Counts as normal.
The respondent sees the completion screen.
Counts, and is marked for attention.
A reward for a flagged response waits for a person.
Kept out of reports, exports and payouts until a reviewer decides.
The respondent sees the normal completion screen.
Not submitted, and never paid.
The respondent sees a neutral stop screen with no reason given.
Rules you can read
Duplicate email per form and per org, the same browser twice, identical answer patterns, copied open text, answers too fast for a person, straight-lining, garbage or keyboard-mash text, links in text, a hidden honeypot field, a failed attention check, a headless browser, and a surge on one form. A respondent with a verified email or a login scores lower. The scoring rules each have their own switch and weight.
A CAPTCHA only when you want one
Choose Cloudflare Turnstile, reCAPTCHA v2 or reCAPTCHA v3. Show the challenge always, never, or only when risk is raised, so most real respondents never see it. A verified challenge failure always blocks. So does a filled honeypot.
Einstein review of open text
Off until an admin turns it on. Einstein reads only the open-text answers and scores them for risk and quality, with a short rationale for the reviewer. You decide whether it may raise a decision to Flag or Hold. It never lowers a decision and never blocks. AI review uses Salesforce's Einstein Trust Layer and the customer's configured model, subject to the applicable service terms and data-handling settings.
A review queue with reasons
Every held response shows each rule that fired, its measured value and its weight. A reviewer can approve, reject or approve as a duplicate, one at a time or in bulk. Every action is written to an audit log.
Monitor mode first
A new policy starts in monitor mode: it scores everything and only flags, so nobody is held or blocked while you learn. Replay a draft policy against the last 30 days of events to see what it would have done, then switch it on.
A policy per form
Bind a policy to one form or to a whole suite. A form binding wins over a suite binding. A paid study can run strict while an internal pulse survey runs light.
Shield Rewards
Rewards that wait for the verdict
Rewards sends a Tremendous gift card or other reward from your own Tremendous campaign after the Shield accepts the response. Your Tremendous account pays. AutoSurvey never holds or moves money: it sends the order and records every step.
| Shield decision | What happens to the reward |
|---|---|
| Allow | Follows your approval policy: paid automatically, paid automatically up to an amount, or approved by a person. |
| Flag | Always waits for a person. |
| Hold | Always waits for a person. |
| Block | Never paid. |
Budgets with a hard cap
Give each study a budget with a cap, a start, an end and an alert percentage. A reward that does not fit the budget waits. One reward per email address, and an annual threshold per email address that sends the next reward to finance for review.
Approval rules you choose
There is no preset. Pick pay every Allow automatically, pay automatically up to an amount, or approve every payment. Separate permission sets keep setup, spend approval, corrections and the send worker in different hands.
Two ways to stop
An emergency stop halts every new send with one click. A safety stop turns it on by itself when the next send would pass 40 rewards or $1,000 in 60 minutes. You can raise those limits. You cannot switch the safety stop off, and only a person can resume.
A retried order reuses the same order key, so a retry cannot pay twice. Finance gets a CSV export by tax year. The first version pays US recipients in US dollars, one reward per response, through Tremendous only.
Privacy
Catch fraud without tracking people
Hashed, not stored
The Shield keeps emails and browser keys only as keyed hashes, made with a key that is unique to your org. A reviewer sees a six-character prefix: enough to spot the same person twice, not enough to identify anyone.
No IP, no location, no fingerprint
The Shield does not capture IP addresses, geolocation, device fingerprints, or mouse and keystroke data. It works from what the survey already sends.
Retention that follows the data
A Shield event is purged with the response it describes. Events from blocked attempts, which have no response, are kept for 365 days by default, and you can set a different period.
Salesforce-native
It runs in your org, on your records
Your data stays in your org
Scores, decisions, review actions, budgets and reward records are Salesforce records in your org, with packaged reports and in-app dashboards. To deliver a reward, Tremendous receives the order: the recipient's email, the amount and, if you choose, a recipient id. Your Tremendous key lives in a Salesforce Named Credential.
Permission sets, not workarounds
Shield Admin, Shield Reviewer, Rewards Manager, Rewards Approver, Rewards Corrector and Rewards Worker. Each role gets only what it needs. Both features are off when you install, and you switch them on in Setup.
Built for Flow
Two invocable actions, Shield Evaluate and Shield Is Eligible, put the decision in any Screen Flow. A platform event announces each decision, and a starter Flow ships with every branch wired.
What it does, and what it does not do
What it does
- Scores each response on the server at submit, with named reasons and weights you can change.
- Adds a Turnstile, reCAPTCHA v2 or reCAPTCHA v3 challenge: always, never, or when risk is raised.
- Holds doubtful responses in a review queue, out of reports, exports and payouts.
- Starts every new policy in monitor mode, and replays a draft policy against past events.
- Optionally asks Einstein to review open-text answers. Einstein reads the open text only.
- Pays Tremendous rewards only after the Shield accepts a response, inside budgets and approval rules.
- Stops sending by itself when the hourly safety limit is reached.
What it does not do
- Does not capture IP address, location, device fingerprint, or mouse and keystroke data.
- Does not pay a blocked response, and does not pay a flagged or held one without a person.
- Does not let Einstein lower a decision or block a response.
- Does not hold or move money. Your Tremendous account pays.
- Does not tell a blocked respondent why.
- Does not store a tax ID or a tax form, and does not file tax forms for you.
- Does not pay recipients outside the US, or in a currency other than US dollars, in the first version.
Requires AutoSurvey. Rewards requires your own Tremendous account. The Einstein review requires Einstein generative AI enabled in your org. Read more about the Fraud & Quality Shield.
See it on your own work.
Bring one program and one workflow. We will walk through Fraud & Quality Shield + Rewards with you on that work and answer your questions.