In the last week of August, a carrier quietly pulled a sailing out of its Asia to North Europe rotation. Nothing on the shipper's scorecard moved that day. Drewry counted 49 cancelled sailings across the main East-West trade lanes between week 35 and week 39, covering August 24 to September 27, roughly 6% of scheduled departures in that window. For most of the cargo booked on those services, the first KPI to register the decision will be the on-time number for late September, five or six weeks after the fact, reported in October.
That lag is not a data problem. The event was public the week it was announced. It is a reporting design problem: the metrics on the page were chosen because the ERP could already count them, not because anyone decided what they were for.
Why the standard scorecard reads like a post-mortem
Open almost any end-to-end supply chain report and you find the same eight numbers: OTIF, fill rate, freight spend per unit, inventory turns, days of supply, perfect order rate, cost to serve, and a carrier scorecard nobody has re-baselined in two years. Every one of them is a verdict on a period that has already closed.
Verdicts are useful for governance. They are close to useless for operations, because by the time a monthly OTIF number turns red, the shipment that caused it was delivered three weeks ago and the customer has already called. This is the same failure mode that makes most supply chain dashboards misleading about the business: the refresh cycle of the report and the decision cycle of the operator are not the same clock.
An end-to-end scorecard that actually earns its place has four layers, not one. Outcome, leading, response, and data trust. Most teams report only the first.
Layer 1: outcome KPIs, kept but demoted
Keep OTIF, perfect order rate, landed cost per unit, and inventory turns. They are the language your CFO and your customers speak, and you cannot manage a network without them. Just stop opening the review with them.
What matters far more than the number is definition discipline. OTIF against the customer's requested date and OTIF against your own confirmed date are different metrics, and teams routinely report one while the customer measures the other. Two plants running slightly different rules will produce two scores that cannot be added together. If you have never audited the formula, start with the two definition traps that quietly skew an OTIF calculation before you argue about the target.
One outcome KPI per business question is enough. Four is a scorecard. Nine is a filing cabinet.
Layer 2: leading KPIs, the ones that move first
This is the layer that would have caught the blanked sailing in week 35. Leading indicators measure the conditions that produce next month's outcome, so they change while there is still something you can do.
The four worth building first:
Schedule integrity by lane. Planned versus actual departure and arrival, tracked as a rolling percentage per trade lane rather than per shipment. A lane whose schedule integrity drops from 78% to 61% over three weeks is telling you about capacity behaviour before a single delivery is late.
Dwell at each node. Time between arrival and onward movement at port, terminal, warehouse, and customs. Dwell is the earliest honest signal of congestion, and it is measurable the same day.
Booking to gate-in lead time. How long it takes your own organisation to convert a confirmed booking into cargo at the gate. When this stretches, you are absorbing risk internally and calling it normal.
Forward cover on at-risk lanes. Days of supply, but filtered to the lanes currently showing degraded schedule integrity, rather than averaged across the network where the exposure disappears.
The reason these rarely reach a scorecard is not that operators do not want them. It is that the underlying events live in carrier feeds, forwarder portals, and terminal systems that were never joined to the order record, so nobody can compute a lane-level number without a week of spreadsheet work. Getting past that means normalizing carrier and system events into one operational record first, which is what a command center like Orkestra does with 200+ pre-built integrations across carriers and ERP, TMS, and WMS platforms. The KPI is a by-product of having the data in one shape. Teams that have already done the work of consolidating freight tracking across carriers usually find these four metrics are a query away.
Layer 3: response KPIs, how fast the organisation actually reacts
Detection is not resolution, and most scorecards measure neither. Three numbers close that gap:
Time to detect. From the moment an event occurs to the moment a human or a rule flags it. Measured in hours, not days.
Time to first action. From flag to the first outbound decision: a re-book, a customer call, an expedite approval.
Time to resolve. From flag to case closed, tracked by exception type so you can see which categories are structurally slow.
Add one percentage alongside them: the share of exceptions caught before the customer feels them. Across the industry, 62% of supply chain exceptions are detected after customer impact, which means the majority of the work happens with the credibility damage already done. Response KPIs are what make that number move, because they put a clock on the part of the process nobody currently owns. Orkestra's exception management works on exactly this loop, auto-detecting exceptions, prioritising them by business impact, and routing the resolution workflow to a named owner. OIA Global cut exception resolution time by 60% in a single quarter on that basis.
Layer 4: data trust KPIs, the layer nobody reports
Every metric above inherits the quality of the events underneath it. Three checks belong on the page, next to the operational numbers rather than buried in an IT report:
Event coverage. The percentage of shipments carrying a complete milestone set. If 30% of ocean shipments have no gate-out event, your dwell metric is measuring a self-selected sample.
ETA accuracy and drift. Predicted versus actual arrival, plus how often the prediction changes in the final 48 hours. A stable wrong ETA is more dangerous than a volatile one, because people plan against it. Scoring it properly is a discipline in its own right.
Master data completeness. The share of orders with a valid incoterm, HTS code, and consignee address at the moment of booking. Almost every downstream exception traces back to a field that was empty at the start.
Publishing these alongside the operational KPIs changes the conversation in the monthly review, because it removes the argument about whether the number is real. It also tends to explain why the BI investment underdelivered: enterprise reporting usually fails on the data model rather than the analytics tooling sitting on top of it.
The test before a KPI goes on the page
For each candidate metric, answer three questions in one sentence each. What decision changes when this number moves? Who owns the threshold? What is the first action when it breaches?
If any answer is missing, the metric is context, not a KPI. Move it to an appendix. A scorecard of ten numbers with owners beats a scorecard of forty without them, every quarter.
What to do with this on Monday
Take your current report and sort every metric into the four layers. Most teams find eight outcome metrics, one or two leading indicators, no response metrics, and nothing at all on data trust. Rebalance to roughly three outcome, four leading, three response, and three data trust. Put a name and a threshold against each one, then delete the rest. If the scorecard covers more than one region, settle how those regions get compared before anyone reads a ranking off it.
The harder half is making the leading and response layers computable, because they depend on events your core systems do not hold in one place. That is the gap Orkestra's analytics is built to close: one normalized record across orders, shipments, inventory, and partners, so schedule integrity, dwell, and time to resolve are standing metrics rather than a quarterly data project. Get the layers right and the October review stops being an explanation of what went wrong in September.
Book a demo to see the scorecard running against your own lanes.
