Zephior publishes the baseline, arithmetic, assumptions and exclusions behind the Ziva performance reference so buyers can evaluate it without relying on an unexplained headline.
Methodology version 1.0 · Last reviewed 22 August 2026
STATUS
The 91% figure is a reference-workflow planning model. It is not an audited portfolio average, a universal customer result or a contractual guarantee. Actual outcomes must be measured in the customer’s own process.
CALCULATION
(30.0 hours − 2.7 hours) ÷ 30.0 hours = 91%
REFERENCE MODEL
The model represents one structured, document-heavy response with a prepared knowledge base and an accountable reviewer. It compares active human work, not elapsed calendar time.
| Work category | Manual baseline | Ziva-assisted oversight | What changes |
|---|---|---|---|
| Intake and requirement setup | 2.0 h | 0.4 h | Import and structure requirements; a human confirms scope and instructions. |
| Source discovery and evidence | 8.0 h | 0.5 h | Retrieve from approved knowledge; a human verifies source fit and currency. |
| Drafting and assembly | 12.0 h | 1.2 h | Generate source-cited drafts; a human corrects, completes and owns commitments. |
| Review and approval coordination | 5.0 h | 0.4 h | Route structured review; accountable approvers retain decision authority. |
| Formatting, export and final QA | 3.0 h | 0.2 h | Produce the target file; a human validates buyer format and release readiness. |
| Total active human time | 30.0 h | 2.7 h | 27.3 hours of modeled human time saved. |
ASSUMPTIONS
MEASUREMENT PROTOCOL
A valid customer result requires a before-and-after design that counts all human work and checks output quality. Draft-generation time alone is not enough.
Select response families with similar size, formats, risk, languages, contributors and approval depth. Freeze the start and end events.
Capture active time across intake, search, drafting, subject-matter input, review, correction, approval, formatting and submission QA.
Use ordinary requests and current source material. Include exceptions, late changes and access-restricted questions rather than a curated demo only.
Compare accepted final outputs. Track unsupported claims, substantive rewrites, missed requirements, export defects and approval exceptions.
Publish the median, distribution, sample size and response mix. Separate active human time from calendar duration and software processing time.
BOUNDARIES
REWORK
Rework depends on source readiness, question ambiguity, required tailoring, reviewer standards and the definition of a correction. Zephior previously displayed a 4% figure without a sufficiently documented public measurement basis. It has been removed. We will publish a rework benchmark only with a defined sample, denominator, collection method and review date.
VERSIONING
Methodology changes receive a new version and review date on this page. Material changes to the baseline, assisted time, scope or evidence status will also update the Ziva product page.
Use a bounded pilot with equivalent work, complete time capture and explicit quality gates.