Frequently Asked Questions

No. Pyze measures how work is performed, not how individuals behave. It captures application interaction metadata: which application, which screen, the sequence of actions and their timing. The unit of analysis is the process. Where a customer wants individual identifiers removed entirely, they can be pseudonymised at the source before anything is transmitted, and most deployments run that way.

Page URLs limited to an explicitly whitelisted set of business applications, navigation and click events, and the timing and sequencing of those actions. No other data is captured.

Pyze never captures keystrokes, field-level values, free-text input or screen contents. It also captures nothing from an application or URL that is not on the whitelist. If a system is not explicitly enabled, no data is captured from it, and the whitelist takes precedence in every case.

A single user identifier, typically a username or an anonymous ID. It exists so interaction data can be joined to other datasets for end-to-end analysis. It can be anonymised or pseudonymised at source, and no customer or end-consumer personal data is collected.

Yes, and we encourage it. Deployments run in collaboration with the team produce better data and less friction. Several customers publish the whitelist internally before the collector is enabled.

The data scope is usually the main topic of discussion. It is limited to interaction metadata: no keystrokes, no content, no field values, pseudonymisable identifiers, and a fixed list of business applications. We will join those sessions and answer questions directly.

SOC 2 Type II, with a current independent third-party penetration test. The full policy set covering access control, encryption, incident response, business continuity and secure development is available on request.

Encrypted at the source in the browser, encrypted in transit over HTTPS, and encrypted at rest. Optional client-side anonymisation can be applied before anything is transmitted.

GDPR principles including data minimisation, purpose limitation and privacy by design, along with CCPA and CPRA and state-level US privacy requirements including Washington MHMDA.

Yes. Integration can be configured to match your preferred data architecture, including customer-managed environments, and processed insights can be streamed to a data lake or analytics environment you designate rather than accessed through ours.

Process mining reads the event logs your systems already write, so it sees work that reached a system of record. A great deal of operational work never does: it happens in email, spreadsheets, portals and second applications. Pyze measures execution at the desk and joins it to the process view, which is why Pyze is a Celonis platform application rather than an alternative to one.

Task mining typically records screen activity for a sampled group over a fixed study period. Pyze runs continuously across the full population, captures metadata rather than screen content, and joins the result to your procedure estate so execution can be compared against what was supposed to happen.

No. Pyze specifies which work should become an agent and measures what changed after one is deployed. Remaining independent of the build keeps the measurement objective.

This is common, and the drift is itself a finding. Comparing documented procedure against observed execution is how it becomes visible and quantified. In practice, defects in a procedure estate are identified before any telemetry is collected.

Around six weeks of continuous measurement, then roughly two weeks to produce the ranked and priced blueprint. The measurement period runs while the work happens normally.

A scoped list of applications, a deployment path for the collector, and access to your procedure documents if you want the conformance view. No process change, parallel time-recording exercise or workshop programme is required.

No. There is no change to how people do their jobs during the measurement window, so the measurements reflect normal behaviour.

Every figure traces back to observed execution, and the method is documented so it can be reviewed. Where something is modelled rather than measured we label it as modelled. Released capacity and redistributed capacity are reported separately and never summed, because summing them would double-count.