How do I calculate on-time-in-full delivery performance?

Calculate order-level on-time-in-full performance by counting the eligible orders that met both the agreed delivery timing and the complete-order requirement, then dividing by all orders in the defined group. Fix the promise, counting unit, period and evidence before comparing percentages.

An average of your on-time percentage and your complete-order percentage does not answer the question. Neither does a high percentage of units delivered. A customer who needed five items together can still be disappointed when four arrived promptly.

Write the promise before the formula

APQC's public OTIF measure describes orders delivered complete within the agreed time frame. Its online-retail definition likewise connects completeness to all order line items and timing to the customer's agreed window.

For your own measure, establish exactly what counts as an order and a successful delivery. Identify the accepted quantity, item specification, location and time window. Check how an actual customer contract treats split shipments, substitutions, early arrivals or other conditions.

A date for dispatching a parcel is different from a date for receiving it. Use the event the measure claims to describe. A printed label or a carrier pickup cannot establish customer receipt without the relevant evidence.

Keep the definition with the number. Two dashboards called OTIF can use different grouping, timing and quantity rules. Comparing their percentages before reconciling those rules can reward a change in measurement instead of an improvement in service.

Choose the orders that belong in the period

Define the cohort before examining results. One useful operating record follows orders due for delivery during a specified period, with a later fixed measurement cutoff. It keeps overdue and incomplete orders visible instead of looking only at parcels that happened to arrive.

Record your actual inclusion rules, time zone, cancellations and any permitted exclusions. Do not quietly omit an order because it became inconvenient, changed status or remained unshipped. If a contract requires a different cohort, preserve that definition rather than borrowing this one.

Keep the original accepted promise and any legitimate revision with its evidence. A current schedule overwritten with a later date can hide what the business originally undertook. A useful service review may need both original-promise performance and a separately defined revised-promise measure.

The delayed-order guide addresses a customer's actual delay and refund choices. Changing a reporting date does not itself provide the notice or agreement an affected order requires.

An original hundred-order record

Consider a wholly fictional business reviewing exactly 100 accepted orders due during April 2027, using UTC and fixed customer-agreed delivery windows. The review cutoff is May 7, 2027. Every order requests five identical specified units, so the original quantity totals 500 units.

The cohort contains every order due in that defined period. There are no cancellations, changed promises, substitutions, returns or disputed receipt records. Each order has one recorded delivery by the cutoff; some deliveries were late or incomplete. The case supplies the whole cohort, not a sample selected for investigation.

For the component timing check, a recorded delivery is timely when it falls inside that order's agreed window. Completeness is checked separately against five units. To pass OTIF, the order must have all five units delivered within its window. All received units otherwise meet the supplied item and destination requirements.

Seventy-five orders received all five units within the window. Five received only three units within the window. Fifteen received all five units after their window. Five received only three units after their window. The missing units remain undelivered at the May 7 cutoff.

Those inputs are original fiction, not warehouse results, a survey, an APQC benchmark or performance experienced by the author.

Count the intersection

Scroll across to see how timing and completeness combine. Each order belongs to exactly one row.

OTIF is 75 ÷ 100 = 75%. Each successful order passes both requirements. The fifteen complete but late orders do not qualify, and neither do the five timely partial deliveries.

The component timing count is eighty orders: seventy-five complete plus five partial deliveries inside the window. That is 80% under this explicitly stated component definition. It is not a claim that eighty complete orders arrived on time.

The complete-order count by the cutoff is ninety: seventy-five timely plus fifteen late, giving 90%. The late group still fails the timing requirement even though its final quantity is complete.

Use the component labels carefully in a real report. A customer's own on-time definition might already require a complete order, unlike this separate arrival-timing check. Do not compare those two percentages under the same label.

Do not average or assume independence

Averaging 80% timing and 90% completeness gives 85%. That is not the observed 75% joint success rate.

Multiplying 80% by 90% gives 72%. That also fails to reproduce this record. The supplied outcomes tell us exactly which orders satisfy both conditions; no independence assumption has been supplied or is needed.

Count the actual intersection. The conversion-denominator guide makes the wider counting distinction between events and eligible records. Here each order gets one decision against two stated delivery requirements.

The failed set has twenty-five orders. Twenty were late and ten were incomplete, with five in both groups. Adding twenty and ten without removing that five-order overlap would incorrectly claim thirty distinct failed orders.

The correct check is 20 + 10 − 5 = 25, and 100 less twenty-five gives the same seventy-five successful orders. The failure categories help explain the result without counting one order twice.

A high unit fill rate can hide order failures

The ninety complete orders supplied 450 units: ninety times five. The ten incomplete orders supplied thirty units: ten times three. Total units received by the cutoff are therefore 480.

Against 500 ordered units, the defined unit fill rate is 96%. Twenty units remain missing. That useful quantity measure does not establish that 96% of orders arrived complete or on time.

Of the 480 received units, 390 arrived within their relevant windows: seventy-five times five, plus five times three. The other ninety arrived late. The defined share of ordered units received within the window is 390 ÷ 500 = 78%.

The 75% order OTIF, 90% eventual order completeness, 96% unit fill and 78% timely-unit share all describe the same supplied record. None is an alternative name for another one. Choose the measure that answers the decision and keep its companions where they explain important differences.

Do not improve the score by losing bad orders

Suppose somebody removes the five late, incomplete orders from the cohort because they remain open. The numerator stays seventy-five while the denominator falls to ninety-five. The displayed rate becomes about 78.95%.

No delivery has improved. The higher figure came from excluding five failures from the original hundred-order question. It is an error under this case's inclusion rules.

If a legitimate reporting rule changes, explain the change and preserve the earlier definition. Recalculate comparable periods where appropriate instead of presenting a changed denominator as proof of operational progress.

Also retain evidence for missing or ambiguous receipts. A tracking status that cannot establish the actual delivery window should not automatically become a pass. Investigate the gap or label the unresolved status under the agreed measurement procedure.

Turn the failed set into work

Review the twenty-five failed orders against their actual facts. Which missed timing, quantity or both? What did the business promise, what happened, and which next action can still help the customer?

The category does not establish the cause. A late delivery may involve an unsupported promise, production delay, picking error, carrier issue or another circumstance. Identify the evidence before assigning responsibility or buying a remedy.

The backup-supplier guide concerns establishing a usable alternative before a disruption. A low OTIF result does not by itself show that another supplier is the right repair or that one is ready.

Keep customer resolution, process improvement and next-period measurement connected but distinct. A later delivery can complete an outstanding obligation while remaining a failure against the original window. Correcting a process can be worthwhile without rewriting last month's result.

Fund a review you can complete

For this finite fictional review, three available paid operations hours at a complete $30 cost $90. Two additional paid analysis hours at a complete $45 cost another $90. Total review work is $180 against a separately available $200 allocation, leaving $20.

All five hours are available without displacing accepted work. They cover the supplied record comparison, exception review and documented findings. No new tool, account or other payment is required in this bounded task.

These are original work allowances, not a measured processing speed or a claim that five hours can repair twenty-five orders. Actual customer remedies, additional delivery work and future process changes require their own appropriate scope and funding.

Make the next comparison meaningful

Keep the cohort rule, promise version, counting unit, delivery window, partial-order treatment, cutoff and evidence method with the result. Check them before comparing another month, site or provider.

Do not import a universal good-OTIF target from a different business. Your contractual commitments and service needs matter, as do the definitions behind any proposed benchmark. This fictional hundred-order record supplies no industry ranking or next-month forecast.

A useful summary would say: seventy-five of the hundred April-due orders met the complete-and-within-window requirement; twenty-five did not, including five that failed both components. Then identify the unresolved customer work and supported improvement actions.