Start with the attacks and processes you want to test
On August 6, 2026, the Google Threat Intelligence Group described a vishing campaign. Attackers posing as IT support called employees, directed them to fake login portals and captured credentials and MFA tokens. The report points to recent activity aimed at financial services, private equity and professional services (Google Threat Intelligence Group, August 6, 2026). For a financial institution, that justifies checking how calls and cross-channel sequences are trained, in addition to email phishing.
First, define two or three situations that matter to your organization. For example: an urgent call to reset access, a message asking to change a supplier's payment details or a transfer request that tries to skip the usual process. Check whether the platform can rehearse the expected behavior, such as verifying through an independent channel, following authorizations and reporting the suspicion.
The platform must complement technical and operational controls, not replace them. A simulation does not show by itself that real incidents have decreased.
Compare vendors by what you can test
Public pages describe capabilities, but they do not prove their effectiveness, the configuration available to your institution or the contractual fit. Use the same proof of concept and the same criteria with all vendors. The Zepo capabilities page describes vishing and real-time deepfake simulations, as well as combined attacks and real threat detection.
Use a common score and separate the mandatory requirements
Before the demos, agree on a scale from 0 to 5. Use 0 for no answer, 1 for a commercial claim, 2 for partial proof and 3 for a requirement demonstrated. Use 4 for a result demonstrated within your workflow and 5 for one that is also documented and repeatable. Score each criterion and apply the agreed weight:
Treat privacy, security, contractual requirements and campaign controls as pass or fail conditions, not as points that a high score in another category can offset. A test without enough documentation must stay pending, not approved.
Check that metrics describe behaviors, not labels
An isolated click rate does not show that risk has gone down. Ask for the definition and denominator of each indicator, together with the channel, the scenario difficulty, the date and the delivery rate. Check whether you can track reports of suspicious messages, time to report, verifications through an independent channel and repeated results in comparable exercises. The formulas for the rates and their denominators are in our guide to measuring phishing behavior.
Separate three things in reports: simulation results, reports of real threats and operational incidents. An increase in reports can reflect more threats or better escalation habits. None of these signals, on its own, attributes a reduction in losses or incidents to the platform.
If the product creates profiles or scores associated with employees, ask which data generate them, who views them, how long they are kept and how errors are corrected. Avoid turning an isolated interaction into a permanent label or a public ranking.
Review privacy, DORA and the institution's obligations separately
Results linked to a person can be personal data. The GDPR requires, among other principles, a defined purpose, minimization, limited storage, security and demonstrable accountability (European Commission, principles of the GDPR). The institution must review the legal basis and the applicable safeguards with its privacy, legal and HR teams. Check whether your national data protection authority has published guidance on employee data.
During vendor evaluation, ask for a list of data and subprocessors, processing locations, possible international transfers, retention and deletion periods, role-based controls, audit logs, incident response and processing terms. Ask also whether the score influences decisions about people and what human review and correction mechanisms exist.
DORA, Regulation (EU) 2022/2554, applicable since January 17, 2025, sets digital operational resilience requirements for the financial entities within its scope. They include ICT and third-party risk management (EIOPA; official text, EUR-Lex). Its application depends on the type of entity and the specific requirement. A training or human risk platform does not certify DORA compliance or replace the institution's responsibility for its program, its controls and its vendors. Review the applicable scope and the contract with specialist advice. For an overview of the five areas, read our article DORA covers five areas of risk.
Run a safe, comparable proof of concept
Use the same script, authorized test group and acceptance criteria for each option. Do not collect real passwords or use real payments, accounts or customer data.
- Simulate a call or message that requests an urgent access reset. Check whether the person can verify through an independent route and report the call.
- Rehearse a request to change a bank account or make a transfer. Check that the exercise reinforces the institution's approval process and does not invite people to skip it.
- Review the follow-up response: feedback, related training, reporting channel and escalation to the security team.
- Request a report that shows metrics, denominators, limits and aggregated results, with access permissions applied.
- Verify one priority integration and review the data it exchanges.
- Compare the demonstration with the documentation on privacy, security, support and cost.
Record what was demonstrated, what depends on configuration and what remains unverified. If a function cannot be tested in the demonstration, ask for current documentation and written acceptance criteria before scoring it.
Frequently asked questions
What is the best platform for a financial institution?
There is no universal answer. The choice depends on the risk channels, the processes, the privacy requirements, the integrations and the evidence the institution needs to review.
Does a human risk platform guarantee DORA compliance?
No. DORA covers the institution's digital resilience and its applicable obligations. A tool can support specific activities, but it does not replace the institution's assessment, controls, governance or responsibility.
Do simulations show that incidents have gone down?
Not on their own. They indicate how a population responded to specific scenarios. To assess results, compare them with equivalent exercises and keep simulation data separate from real incidents.