Producing a number somebody will trust

The recurring failure in process reporting is not that data is unavailable. It is that the same question produces different answers depending on who asks and how, and that the difference is invisible in the resulting number.

A monthly production figure calculated by operations and one calculated by finance differ because they used different tags, different boundaries, different handling of bad data and different definitions of the month. Each is defensible. Neither is reproducible by the other.

Define the calculation once

The remedy is a documented definition for every reported figure: which tags, what aggregation, over what period boundary, with what handling of bad quality and downtime, and which conversion factors.

Labour-cost ratios can provide additional context when staffing effort is compared with output. Further details are available in this overview.

A useful external reference for timing and synchronisation is NIST time and frequency services.

This sounds bureaucratic until the first time two figures disagree and the definitions resolve it in five minutes rather than a week.

Calculate centrally, distribute widely

Where each department builds its own extraction, divergence is guaranteed. Where a defined calculation runs once and everyone consumes the result, it is impossible.

The practical implementation is a set of derived tags in the historian, calculated on a schedule, that constitute the official figures. Reports read those rather than recalculating from raw data.

This also makes the calculation auditable, because it exists in one place and its history is stored.

Period boundaries

A surprising share of reporting discrepancies come from boundaries. Does the production day run midnight to midnight or from the morning shift handover? Does the month end at midnight on the last day, or at the end of the shift that spans it?

Both conventions are used, frequently within the same organisation, and the difference on a boundary day can be material.

Handle downtime explicitly

An average that includes periods when the unit was not running is not an average of anything meaningful. A monthly average temperature for a reactor that was down for eight days describes a mixture of operation and ambient.

Reports should state the basis: whether the figure covers the whole period or only running time, and how much running time there was. That last number is frequently more informative than the figure it qualifies.

Show the working

A report that presents a figure alone invites the question of where it came from. A report that presents the figure with its inputs — the period, the running hours, the data quality percentage, the tags used — pre-empts it.

The additional space is small and it changes how the report is received, from something to be questioned to something to be discussed.

Automate it, then check it

Reports assembled by hand each month consume time and introduce transcription errors. Automating them is straightforward and introduces a different risk: a calculation that has silently broken continues to produce plausible numbers.

The defence is a periodic manual check — recalculating one figure independently, quarterly — and sanity bounds on the automated output that flag implausible results rather than reporting them.

Reports nobody reads

Process reporting accumulates. A monthly pack that has grown to forty pages over a decade contains a large proportion that nobody looks at, and the useful figures are buried.

An occasional review — asking recipients which pages they use — typically allows substantial reduction. The pages that survive are the ones worth automating properly.

Manual entries and their place

Most operational reports combine automatic data with manually entered figures: laboratory results, meter readings, production confirmations.

Manual entries need the same treatment as measured data — a timestamp, an identified source, and a distinction from measured values — or a report becomes an undifferentiated mixture whose reliability cannot be assessed.

They also need validation at entry. A transposed digit in a manually entered figure propagates into every derived number and is frequently caught only when someone notices an implausible result weeks later.

Reconciliation between systems

Where the same quantity is measured or recorded in more than one system — the historian, the metering system, the production accounting system — the figures will differ.

Small differences are normal and expected. What matters is that they are quantified, that a tolerance is agreed, and that differences beyond it are investigated rather than each system's users continuing to quote their own figure.

A periodic reconciliation, with the differences recorded and trended, converts a recurring argument into a monitored parameter.

Who owns the report

Reports without an owner decay: a calculation breaks, a tag is renamed, a source system changes, and the report continues to produce numbers that are quietly wrong.

Each report should have a named owner responsible for its accuracy, and a review date. Reports that cannot find an owner are usually reports nobody needs, which is itself a useful finding.

Self-service versus controlled reporting

There is a tension between allowing users to build their own queries and maintaining consistency of reported figures.

The workable resolution is to distinguish the two explicitly: a defined set of official figures, calculated centrally and used for anything reported outside the team, and open exploratory access for analysis and investigation.

Problems arise when an exploratory query becomes a reported figure without passing through the definition process, which is why the distinction has to be stated rather than assumed.

Report timing and data availability

Reports scheduled shortly after a period boundary may run before all data for the period has arrived: late-arriving laboratory results, a delayed interface, a communication outage that backfills.

The result is a report that is correct when produced and wrong an hour later, with two versions in circulation.

Either delay the report until data is complete, or state the cut-off explicitly and provide a defined revision mechanism. What causes difficulty is a report that is silently revised.

Retaining what was reported

A report regenerated from current data may not reproduce what was originally issued, because underlying data has been backfilled or corrected.

Retaining the issued version as a record, separate from the ability to regenerate, is what allows a figure quoted six months ago to be traced.