Refund verification
Verified$42.00Refund confirmed
- Agent reported
$42 refund completed
- Billing confirmed
RF-59022 · $42.00 refunded
- CRM confirmed
CS-8831 · Case closed
Policy snapshot
Refund Policy v3.2
The business evidence layer for AI agents
AI agents act.
Pruvz proves
what happened.
AI agent outcome verification against your systems of record. Connect high-impact actions to decision-time policy, confirmed outcomes, and the evidence behind them.
Working product · Customer pilots open · Founding design-partner program
Non-blocking. Independent. Built for review.
Inspect evidence packet
BillingCRMPolicy
The accountability gap
When an AI agent acts,
what can your team prove?
A customer disputes a refund. A manager, auditor, or regulator asks what happened. Can your team show why the agent acted, which policy applied, and whether the result matches its report?
A successful technical trace alone cannot answer those questions. Pruvz connects the decision and policy to what your business systems actually recorded.
Agent report
“Refund completed.”
System of record
No matching refund.
The claim and the evidence. Side by side.Separate illustrative example.
Understanding the category
What is an AI agent evidence layer?
An AI agent evidence layer connects an agent's decision-time context and policy to the action taken, independent system-of-record observations, and the resulting verification and review state. Pruvz is the business evidence layer for AI agents: independent AI agent outcome verification against systems of record.
Where Pruvz fits
Observability, governance, and the evidence layer.
Agent observability explains technical execution. AI governance manages policy and risk. Pruvz independently verifies business outcomes in systems of record and keeps the evidence behind them. Governance owns the policy; Pruvz proves each action against it, so the same risk and legal teams read the evidence. These roles can work together in the same stack.
How it works
From agent activity to verified business evidence.
Good to know
Pruvz sits beside your agents and enterprise systems, with non-blocking integration through an authenticated API and asynchronous evidence collection. It captures the decision context, checks the outcome, and keeps the record ready for review.
- 01
Capture context
Record the user request, retrieved data, model decision, tool intent, and relevant business context.
- 02
Snapshot policy
Preserve the exact policy version, rules, thresholds, and exceptions used at decision time.
- 03
Verify outcome
Check trusted systems such as CRM, billing, ERP, ticketing, risk engines, and approval flows.
- 04
Review evidence
Route exceptions and missing evidence to human review with a complete, readable evidence packet attached.
03 · Verify outcome
How system-of-record verification works
Understand system-of-record verification- The report is a claim
- Pruvz treats the agent's report as a claim. Its verification service independently reads the business systems and assigns the result. Agents can submit claims, but cannot assign their own verified outcome.
- Verification window
- Read-back runs within a defined verification window, with retries. Unreadable or non-terminal observations stay pending; an unreadable source is not a business mismatch. The result becomes final on classification, and records what was observed and when.
- Human review
- Pruvz records human review decisions as new evidence. A human can trigger re-verification after a case is resolved externally; the original result remains on record. Bounded post-verification monitoring is planned.
Inside a Pruvz Evidence Packet
The full story behind each important agent action.
An evidence packet is an ordered, append-only business record. It connects the request and decision-time policy to the agent's action, independent observations, outcome classification, and review decisions, so business and technical teams can inspect the same evidence.
What the agent saw
User request, retrieved data, customer context, documents, previous cases, and relevant business state.
Which policy applied
The exact policy version, thresholds, rules, exceptions, and approval requirements used at decision time.
What decision was made
The agent's business decision, decision reason, allowed action, blocked action, or required escalation.
Which action was executed
Tool calls and business actions such as refunds, claim updates, plan changes, approvals, or CRM updates.
What actually happened
Verified outcomes from systems of record such as CRM, billing, ERP, ticketing, approval flows, and risk systems.
What needs review
Exceptions, missing evidence, threshold violations, suspicious actions, and items requiring human review.
Pruvz Evidence #PRVZ-2287 · Illustrative workflow. Five key fields are shown below. These values explain the record format; follow the evidence packet walkthrough for a real demo record.
- Agent decision
- Refund approved · $42Verified
- Policy version
- Refund Policy v3.2Snapshot
- Action executed
- $42 refunded in billingConfirmed
- System-of-record check
- Billing + CRM confirm the outcomeVerified
- Review status
- Auto-cleared by policyClear
Explore the remaining 8 fields
- Agent action ID
- act_8f31c0Captured
- Customer / account
- ACC-48120 · Tier 2Captured
- Policy match
- Matched decision-time thresholds and exception rulesMatched
- Decision reason
- Under $75, no prior abuse signalCaptured
- Checks performed
- Policy threshold + abuse signal + billing confirmationChecked
- Evidence sources
- CRM + Billing + PolicyComplete
- Verification confidence
- High, all required sources verifiedHigh confidence
- Integrity
- Ordered, append-only evidence trailRecorded
When evidence is incomplete or inconsistent
Outcome mismatches, missing evidence, and policy-sensitive cases need review. Human decisions are appended to the evidence trail and never rewrite the original verification result. Pruvz records and verifies; it does not approve or block the agent's business actions at runtime.
Report vs Reality · 85 seconds
The agent reports what it did. Pruvz provides the evidence of what happened.
An AI agent reports that a refund is complete. Pruvz captures the claim, independently checks the system of record without slowing the agent down, and builds a Business Evidence record: what the agent saw, the policy in force, what it decided, what it tried to do, and what the system of record shows. The same record applies to claims, cancellations, approvals, order changes, permission grants and account updates.
Read the film transcript
The thesis
The agent reports what it did. Pruvz provides the evidence of what happened, checked independently in the system of record. AI agents approve refunds, change orders and update systems, at machine speed.
The story everyone accepts
A customer asks for a refund. The agent calls the payment system, gets an accepted response, updates the CRM and tells the customer: your refund is complete. As far as everyone is concerned, the story is over.
The gap
But an agent's report does not always prove what actually happened. Sometimes the gap only shows up when the customer comes back to complain.
An independent check that does not slow the agent down
Pruvz captures the agent's claim, then independently checks the system of record without slowing the agent down.
The Business Evidence record
For every business action Pruvz verifies, it creates a Business Evidence record: what the agent saw, the policy in force at the time, what it decided and what it tried to do. Then it reads the system where the outcome should have been recorded, and shows every gap together with the evidence behind it.
The same record for other business actions
The same record can be built for other business actions, such as: an insurance claim, a subscription cancellation, a customer approval, an order change, a permission grant or an account update.
The thesis, again
The agent reports what it did. Pruvz provides the evidence of what happened.
Product availability
What works today. What we build together. What comes next.
Pruvz is a working product. Customer pilots and the founding design-partner program are open. Use the current capabilities below to scope a first workflow.
Running in the product, shown end to end in the demo
- Full verification flow: agent action captured, executed, and independently verified
- Decision-time policy snapshots, immutable and bound to each action
- Independent read-back of billing and CRM outcomes, running against live Stripe and HubSpot test environments
- Ordered, append-only evidence packets for every action, with cryptographic commitments and Merkle proofs
- Supports independent offline verification, with cryptographic assurance capabilities enabled per customer
- Outcome mismatch detection, human review decisions on record, and human-triggered re-verification of externally resolved cases
- Sign-in with enforced roles and tenant isolation, data minimization with redaction, and executed retention
- Business overview built from verified outcomes, with drill-down to evidence
What we build with founding design partners
- Applying the verification flow to a partner's real agent workflow
- Building connectors tailored to a partner's own systems of record
- Shaping the evidence model, review flow, and integration priorities
On the roadmap
- Additional system-of-record connectors: ERP, ticketing, approval flows, and risk systems
- Bounded post-verification monitoring of already-final results
- Narrative summaries and trend analytics on top of the verified metrics
Evidence-backed business intelligence
From verified actions to a trusted business view.
Pruvz aggregates verified outcomes into business metrics, with drill-down to the evidence behind each result. Product, operations, and business teams can review confirmed outcomes and exceptions across recorded agent activity.
Business outcome metrics
Review metrics built from verified terminal outcomes, including confirmed refunds and outcome mismatches, and inspect the underlying evidence.
Drill down to evidence
Move from business outcome metrics to the recorded actions and evidence packets behind each result.
Use cases and teams
Where AI agent outcomes matter to the business.
Start with a workflow whose outcome can be checked in a trusted business system. Billing and CRM read-back runs against live Stripe and HubSpot test environments. Other workflows below are candidates for a scoped design-partner engagement; connector coverage is agreed for each pilot.
Billing & refunds
Refund approvals, billing corrections, subscription changes, credits, and payment-related agent actions.
Claims operations
Claim intake, claim rejection, policy eligibility, approval thresholds, and exceptions that need human review.
Customer support
Escalations, account updates, compensation, plan changes, retention offers, and SLA-driven decisions.
Risk & governance
Policy exceptions, audit requests, compliance reviews, approval proofs, and internal investigations.
Who Pruvz is for
Product, business, operations, compliance, risk, and legal teams.
Engineering teams connect the workflow and systems of record. Operations teams review exceptions. Product and business leaders follow outcomes. Compliance, risk, and legal teams open the same record when an action is disputed or audited: the policy in force, the decision, and what the system of record confirmed.
Meet the teams Pruvz is built forSecurity and trust
Independent verification. Evidence built for review.
Understand how Pruvz separates execution from verification, limits data access, and preserves the evidence behind each result.
Execution and verification are separate trust domains: your agent acts with its own credentials, Pruvz verifies with read-only, least-privilege access, and in the production deployment model the two are never the same identity. Every claim, observation, and review decision is appended to an ordered, append-only record with cryptographic commitments that support independent tamper-evident verification. Sensitive fields are redacted at the connector edge, sign-in runs through Microsoft Entra with tenant-scoped access, and the production deployment model is designed to provide encryption in transit and at rest and to support enterprise security reviews and SOC 2 / ISO 27001-aligned environments. Deployment-specific security documentation, including a security one-pager and DPA readiness pack, is available during design-partner and enterprise evaluations.
Read the six security pillars and the full architectureResources
Explore the evidence. Understand the category.
Go deeper into the product, the verification model, and the questions your team needs to answer.
Product demo
Watch the verification flow, end to end.
Evidence packet
Inspect a real captured record, field by field.
Security & architecture
Understand the trust model and current availability.
Business view
See verified outcomes roll up into business metrics.
Where Pruvz fits
Compare approaches to business outcome verification.
Outcome verification
Connect agent actions to business results.
Refund verification
See how billing read-back confirms a refund.
Systems of record
Learn what makes an independent observation.
From the Pruvz blog
Guides to evidence, policy, and human review.
Questions teams ask first.
Is Pruvz available today?
Yes. Pruvz is a working product: an AI agent acts, Pruvz captures the decision-time context and policy, independently verifies the outcome in the systems of record, and routes mismatches to human review. The same flow runs end to end against live Stripe and HubSpot test environments, and the recorded product demo shows it working. Customer pilots are open, and the founding design-partner program is open for teams that want to apply the verification flow to their own workflows through a scoped engagement and help shape connector priorities.
Can I see Pruvz working?
Yes. The product demo shows the full flow end to end: agent action, policy snapshot, independent system-of-record verification, outcome classification, and human review of mismatches. Visit pruvz.ai/demo to see what it covers and book a live walkthrough.
How is Pruvz different from agent observability and tracing?
Observability and tracing tools show technical traces: model calls, tool calls, and execution. Pruvz creates business evidence: policy, decision, action, system outcome, and review state. Signed audit trails and traces only prove what was recorded; business evidence proves what actually happened, confirmed in your systems of record.
Does Pruvz approve or block actions?
No. Pruvz does not need to be the approval system. It records and verifies what happened, and can route exceptions or missing evidence for human review.
What happens when evidence is missing?
The action is marked as missing evidence, outcome mismatch, policy exception, or needs review, depending on the case, and routed to the right team.
What does a compliance or risk team get when an action is disputed or audited?
The evidence packet for that action: the decision-time policy snapshot, the agent's decision and the action it took, the independent read-back from the system of record, the verification result, and every human review decision appended afterwards. Records carry cryptographic commitments and Merkle proofs that support independent tamper-evident verification, offline.
Make your AI agents provable.
Bring one high-impact workflow. Let’s build the evidence around it.
What you get
- The Pruvz verification flow applied to one of your high-impact agent workflows
- Evidence packets, mismatch detection, and human review for that workflow
- Direct influence on the evidence model, connectors, and roadmap
- Priority access as Pruvz moves toward general availability
What we need from you
- One high-impact agent workflow, such as refunds, claims, or support actions
- Access to the relevant systems of record in a staging or test environment
- A short feedback loop with the team that owns the workflow
How the engagement runs
- A structured proof of concept, scoped in weeks rather than quarters
- Success criteria agreed up front: which actions get verified, against which systems
- Weekly working sessions with Pruvz
- Next step: a 30-minute intro call to map your workflow to the evidence model
Become a design partner
Tell us about your team and the workflow you want to verify.