RFTRight First Time

A batch is RFT if it passes through every production step and every release test without any of the following: an unplanned deviation, an OOS or out-of-trend (OOT) result, a retest, a re-sample, a rework, a re-process, a re-package, a QA-disposition exception (use-as-is, concession, downgrade), or a planned deviation outside the validated envelope. Some companies also exclude batches with planned deviations even inside the envelope; the operational definition is documented in the metric's data dictionary and held constant so the trend is meaningful. The metric is calculated as (batches RFT in the period) ÷ (batches released in the period) and trended monthly, with the rolling 12-month figure being the headline value on the PQR.

Final yield says 'we shipped 99% of the kilograms we charged'. RFT says 'and 35% of the batches needed a deviation or a retest to get there'. A site can run at 99% final yield with RFT at 65% — every batch passing on the second attempt or after an investigation. The compliance and cost burden of the 35% is enormous: investigations, retest analyses, QA dispositions, reviewer overtime, batch-record errata, deviation-trend slides at every quality council. RFT is the visible cost of getting yield; final yield alone hides it.

• Process capability (Pp/Ppk) — broader IPC distributions mean more attribute results land near a spec limit and convert into OOS at the analytical noise level.
• Analytical method capability — methods near LoQ for a critical attribute generate retest noise that breaks RFT even when product is fine.
• Material consistency — variable raw materials propagate variability into the batch.
• Operator execution — incorrect dispense, mis-set parameter, missed step; reduced by EWI + standard work + poka-yoke.
• Cleaning carry-over — incomplete changeover producing IPC failures or OOS in trace-analysis tests.
• Equipment reliability — equipment failure mid-batch is a guaranteed deviation.
• Specification design — tight cosmetic or non-critical specs trigger RFT loss without product-safety reason; QbD design avoids this.

RFT drops 2-3 quarters before a Stage-3 Ppk excursion or an APR finding because deviations and OOS investigations accumulate before they trigger a capability signal. A site that watches RFT week-to-week catches drift before it becomes a Ppk story; a site that waits for the PQR finds out a quarter or two too late. Mature programmes pair the RFT trend with deviation root-cause categorisation — when the same category (e.g. 'material variability', 'operator dispense error') drives both a flat-but-low RFT and a rising deviation count, that's the prioritised improvement target. This is the operational expression of ICH Q10 §3.2.4 'continual improvement'.

• What counts as a deviation that breaks RFT — minor vs major, planned vs unplanned, in-envelope vs out-of-envelope?
• What counts as an OOS — invalidated assignable cause vs confirmed result?
• What counts as a retest — assignable lab error vs investigative retest?
• Time window — is a batch RFT 'at release' or 'at 12-month stability check'?
• Scope — by site, by product, by line, by shift? All four if you can.
• Treatment of batches still in investigation at month-end — excluded, or counted as failed pending resolution?

These choices are not technical — they are governance. Document them in a metric data dictionary, freeze them, and revisit at major changes (FDA Quality Metrics submission, new ICH Q10 self-assessment).

1. RFT calculated by hand from a deviation spreadsheet that's a quarter out of date.
2. Inconsistent rules — one site counts retests, another doesn't; cross-site comparison is meaningless.
3. Definition silently relaxed (some deviation types newly excluded) and RFT 'improves' by 10 points overnight with no underlying change.
4. RFT presented monthly without root-cause Pareto — the trend is visible but the next action is not.
5. Stability-related OOS not counted as breaking RFT — 12-month failures appear orphaned from the lifecycle metric.
6. RFT and product-quality complaint rate trended on different denominators — analyst can't reconcile.
7. Batches in long investigation excluded from denominator rather than counted as 'in-flight RFT pending' — gaming the metric.

• RFT computed from primary records — every deviation, OOS, OOT, retest, rework or QA exception attached to a batch automatically removes it from the RFT numerator.
• Configurable data dictionary — what counts as an RFT-breaker is set per tenant per product family, then frozen and version-controlled.
• Per-site, per-line, per-product, per-shift slices; rolling 12-month, calendar quarter, and FDA Quality Metrics fiscal year all available.
• Pareto by failure category live on the RFT dashboard — never a number without the breakdown.
• Leading-indicator drop alerts feed the management review and the Stage-3 CPV review work-items at the moment they happen.
• APR / PQR consumes the same primary records — the RFT in the PQR and the RFT on the live dashboard agree by construction.
• FDA Quality Metrics submission export with LAR / IOOSR / PQCR pre-formatted from the same data.