Measurement and indicators

Process performance indicators

In French: Indicateurs de performance du processus (IPP)

Performance indicators measure a process from a few complementary angles — quality, lead time, cost — distinguishing those that record a result from those that announce a drift.

Recording or anticipating

The most useful distinction is not between families of indicators but between two natures. A result indicator (lagging) measures what has already happened: the month’s scrap rate, the quarter’s revenue. It is reliable and arrives too late for anyone to act.

A leading indicator measures something that precedes the result: rate of adherence to the standard, number of open corrective actions, regularity of team huddles. It is more debatable and makes it possible to act before the drift.

A dashboard made up only of result indicators produces meetings where the past is commented on. Both natures are needed, and one must know which one is being looked at.

The indicators most often encountered

RTY (rolled throughput yield) — the end-to-end yield, obtained by multiplying the yields of each step. It is always lower than the worst of them, which makes it uncomfortable and useful: four steps at 95 % do not make 95 %.

FTT / FTQ (first time through) — the share of units produced right first time, rework deducted. Not to be confused with RTY: FTT can be measured on a single step.

OTD (on-time delivery) — the share of deliveries made on the promised date. The trap lies in the definition of "the promised date": initial or renegotiated? The same process shows two very different results depending on the answer.

DPMO — the number of defects per million opportunities, which makes it possible to compare processes of different complexity and serves as the equivalence with the sigma level. It has its own entry, with the conversion table and the convention it embeds.

The dashboard

The rule that holds: few indicators, but balanced. The SQDCP format — safety, quality, delivery, cost, people — has the advantage of preventing one dimension from being optimised at the expense of the others. Five to ten indicators are enough; beyond that, nobody looks at all of them any more, so nobody looks at any.

Calculations

RTY = yield₁ × yield₂ × … × yieldₙ

FTT = (units produced − reworked − scrapped) / units produced

DPMO = defects / (units × opportunities per unit) × 1,000,000

The number of opportunities per unit is the most manipulable datum in DPMO: increasing it mechanically improves the result without anything changing in the process. It must be defined once, written down, and never move again.

The common mistake

Multiplying indicators to "cover everything". A thirty-line dashboard is not read: it is endured. Worse, it ends up measuring what is easy to count rather than what counts. I have more often gained by removing indicators than by adding them — and the best test remains asking someone what decision they will take if one of them turns red. With no answer, the indicator can go.

On the shop floor — Financial services

A back office tracked a file conformity rate of 98 % and was receiving a rising number of complaints. Both were true: the indicator covered the completeness of the documents, not the processing time, which was the sole reason for the complaints.

The other tools illustrated by this trade: banking, insurance and back office.

Frequently asked questions

What is the difference between a leading indicator and a result indicator?

The second measures what has already happened: it is indisputable and it arrives too late for anything to be changed. The first measures a condition that precedes the result, and therefore lends itself to action while there is still time. It is, however, more debatable, since it rests on an assumed causal link.

What is RTY?

The end-to-end yield of a sequence of steps, obtained by multiplying together the yields of each one. Its awkward property is useful: it always falls below the lowest of the individual yields, which reveals a cost of rework that step-by-step rates conceal.

How many indicators should a dashboard have?

Five to ten, covering different dimensions rather than ten variants of the same one. Beyond that, reading stops being exhaustive and becomes selective, with everyone looking at what suits them. The most effective test is to ask, for each line, what decision would be taken if it turned red: with no answer, the line can disappear.