The balanced scorecard, on the second Tuesday

Build a process that produces the balanced scorecard for the board every month, on its own. It obtains its own data from the contributing departments and systems, computes each measure from a versioned definition, verifies its own accuracy and its own formatting, produces a version per audience, and lands before the papers go out on the second Tuesday cycle. Without an email chain, and without the last fortnight of two senior people.

The target is stated, because an instruction without one produces an improvement rather than an answer. Thirty minutes of one person time per cycle, and anybody who has read the instructions should be able to run it: no special knowledge of how the pieces fit, no relationship with the seven contributing departments, no memory of what happened last November. If it needs the director, it has failed, and if it needs the department manager, it has failed in a quieter way.

Note where that obligation sits. It is not a claim that anyone is capable of it, which would be a statement about people. It is a requirement on the instructions: having read them, a person should be able to produce the pack. Every question they still have to ask is a defect in the document, to be closed in the document, and the test below is a test of the writing rather than of the reader.

That target is what turns this from a reporting project into a process design. Twenty senior days becoming thirty minutes of anybody time is a factor of roughly three hundred, and nothing survives that except a process where the data arrives on its own, the checks run themselves, the format is fixed, and the only human act is clearing exceptions and writing a line against whatever was flagged.

The scorecard itself is part of the answer, because a process cannot be designed before the thing it produces is decided: every measure tied to a decision the board would actually take, every measure carrying a counter measure, and every measure reported at the freshest period it can honestly support.

That last clause is the second half of the instruction. The pack today reports the month before last, which means the board is looking at something five to ten weeks old, and the reason is production time rather than data availability. Operational data is complete within days of the month closing. If the process takes thirty minutes, the natural question is what can be pulled forward to the month just ended, and the answer is expected to say, measure by measure, what can and what genuinely cannot.

Two things are in scope that usually are not. The first is what to stop: the existing thirty eight page pack, and the recurring reports that will fund the capacity for this one. The second is the ownership question underneath it, because a scorecard is only as reliable as the function producing it, and that function currently reports performance and assures its own evidence.

A third fact shapes the whole answer, and it is the largest number in this brief. The pack is compiled personally by the director in charge of the department and the department manager, because it carries operational and financial information together and was judged too sensitive to delegate. It takes them the last two weeks of every month.

Price that honestly. Two senior people, half of their working time, twelve months a year: roughly two hundred and forty senior days annually, spent on assembly rather than on running the function or answering the questions the operation is waiting eleven days for. It is almost certainly the single most expensive recurring process in the corporate centre, it has never appeared as a cost anywhere, and it is the budget available to build the replacement.

It also explains the paralysis. The two people with the standing to redesign this are the two whose time it consumes, and they are consumed in exactly the fortnight when such work would have to be done.

One more absence, and it is the one that makes every improvement claim unverifiable: there is no history. The pack cannot be opened as it stood last month or as it stood the first time it was produced, so nobody can see whether a number was later restated, and no measure has a trajectory except the one drawn inside this month edition. The requirement is stronger than an archive: every prior edition has to be visible from the current one, back to the first, without anybody knowing where to look.

Then the part that consumes the second week. Every month the same thread argues the results and the formatting before anybody accepts the pack: three to five rounds of challenge, re-cut and re-order. Nobody has written down what accepted means, so in practice it means the objections have stopped.

One design principle is given rather than asked for, because it is the thing that ends the chain: a department owns its own data and does not send it. Either we agree where it parks it and how, or we agree where it already lives and the process takes what it needs automatically. Never by message.

And the mechanism underneath all of it: the whole thing runs on a single email chain with about fifteen people on it. Data goes round as attachments, drafts come back as attachments, the current version is whichever arrived most recently, and the distribution list is doing the job of an access control. Every other problem in this brief is made harder by that one fact, and it is also the cheapest to fix.

A fifth fact explains where the fortnight goes. The pack is collected rather than extracted: seven internal departments submit, each from its own system and its own timetable, and the assembly work is chasing, reconciling and adjudicating between them. Operations and finance disagree about outcomes, HR and operations disagree about headcount, and somebody senior decides in private which version the board will see.

And a fourth fact, which decides most of the design. The numbers come from four contracts commissioned by three different departments, each defining its own metrics. Much of what a board would like to see as one line does not aggregate at all, and the parts that appear to aggregate are the most misleading, because they move with the contract mix rather than with the work.

An answer that presents a list of attractive metrics without the production calendar, the counter measures and the retirement list has designed a slide rather than a scorecard, and will be scored as such.

The second Tuesday falls between the 8th and the 14th, papers circulate five working days ahead, and the pack reports the month before last. Put those together and the board is discussing a period that ended between five and ten weeks earlier, having already lived through the month in between.

The important part is the fifth row. That lag is not caused by the data: operational figures are complete within days of the month closing. It is caused by the production taking a fortnight, which meant nobody could sensibly start on the day the month ended, so the cycle slipped a month and then stayed there. It is an artefact of the process, and a process that takes thirty minutes has no reason to carry it.

The honest design question is therefore three questions. What could be reported about the month just ended by the papers deadline. What is genuinely provisional at that point and should be marked as such. And what can never be current at all, which is claims validation, sustainment and audit findings, and should be dated on the page rather than quietly dropped for being old.

Automatic does not mean unattended, and it does not mean a dashboard. It means the process obtains its own inputs, computes every measure from a versioned definition, checks itself, renders itself into a fixed format, and presents a human with an exceptions list rather than a blank page and a fortnight.

The nine checks below are the ones the director and the department manager currently perform by reading. That is precisely why it takes two weeks, and why a check is occasionally missed in the last hour before the deadline. A rule does not get tired at half past six on the day the papers are due.

Note the third column. Nothing here blocks the pack. A missing input is published as a missing input, an anomaly is published as an anomaly with a line of explanation, and a failure to reconcile is published as a difference with its tolerance. The alternative, holding the pack until everything is perfect, is exactly how a cycle slips by a month and then never comes back.

The pack is not extracted from a system. It is collected from seven departments, each with its own source, its own timetable and its own account of the month, and then reconciled by hand by two senior people who are the only ones who know how the pieces are supposed to fit.

That is where the two weeks go. Very little of it is analysis. It is chasing submissions, waiting for the last one, noticing that operations and finance disagree about how many outcomes there were, deciding which version goes in, and rewriting a narrative around whichever number survived.

The fourth column is the important one. Every collision in it is a real disagreement between two departments about what happened last month, currently resolved in private by whoever is assembling the pack. That is a governance question disguised as a formatting problem, and no scorecard design survives leaving it unanswered.

The answer is required to apply one principle, stated here rather than left to be discovered, because it is the thing that actually ends the email chain. A contributing department owns its data and does not send it anywhere. Either we agree where you park it and exactly how, or we agree where it already lives and the process takes what it needs from there, automatically, on the cut off.

Both modes are acceptable and most companies need both. A system of record with a decent interface is read directly. A department whose numbers live in a spreadsheet publishes that spreadsheet to one agreed location, in one agreed shape, and keeps owning it. What is not acceptable is transfer by message, because that is what creates the second copy, the version confusion, the access exposure and the fortnight.

The shift this makes is one of accountability rather than technology. Today the person assembling the pack is implicitly responsible for whether every department numbers are right, which is impossible, so they resolve disagreements by judgement under deadline. Under the principle, correctness belongs to the department that produced the data, visibly, and the centre is responsible only for taking it faithfully and saying plainly when it did not arrive.

The data arrives from four contracts, commissioned by three different departments, each with its own definitions of the events it pays for. None of those commissioners will change a definition to make a board pack tidier, and none of them should.

So the group number is the problem. An outcome rate for the company as a whole is a weighted average of three different events, and it moves whenever the contract mix moves, whether or not anything about the work changed. A board watching that line is watching the shape of the order book and calling it performance.

There is a specific trap worth naming, because it is the one that catches able people. A group rate can fall while every single contract rate rises, simply because a contract with a harder cohort grew as a share of the total. Any scorecard that shows a group trend without showing the mix alongside it will eventually produce a confident conclusion that is exactly backwards.

Five perspectives, each with the question it answers for the board, the thing it must not be allowed to become, and the counter measure that keeps it honest. The general rule is in the third column of every row: a measure with no counter measure will be improved by damaging something the board is not looking at.

The participant perspective is the one most likely to be dropped for being hard to produce, and it is the one the values page commits to. If it is not on the page, the commitment is not real.

Six facts about the present state, how each came about and what it costs. Every figure is indicative and stated as a baseline to be measured, not as a finding. The first act of any answer is to replace this table with the real one, because a change without a measured starting point cannot be evaluated afterwards and will be claimed as a success by whoever is still in post.

Note the board pack row. Thirty eight pages, four analyst days a month, read to page three. That is the true cost of the thing being replaced, and it is also the budget available to build the replacement.

The scorecard is the ask. These seventeen are the conditions it has to be produced under, and an answer that designs a beautiful page the function cannot assemble by the deadline has answered a different instruction.

Several are fixed by tooling and discipline whatever the org chart says. Two, data quality at source and the questions nobody owns, are not the data function to fix alone. An answer should say which of the seventeen it addresses, which it leaves standing, and which it refers elsewhere by name.

A function cannot be designed before somebody says what it is for, and a scorecard is an expression of a purpose whether or not the purpose was stated. These five goals are proposed here so the design has something to serve.

Each states how it is measured, and three of the five measures cannot currently be produced at all. That is the most useful finding on this page: a department that cannot measure its own goals is in exactly the position of the operation it reports on.

An answer may argue with any of these goals. What it may not do is propose a scorecard without saying which purpose it serves.

Five production models will be scored, and they were named before any weighting was set. One: keep the present arrangement, email chain included, and simply shorten the pack. Two: structured submission, where every department files to a template in one controlled location by a published cut off and the centre assembles by hand from what is filed, with no email in the loop. Three: automated pull, where a reconciled data layer takes what it can from source systems and contributors supply only what genuinely cannot be pulled, which is commentary. Four: full automation with exception handling, where everything computable is computed, the nine checks run themselves, and humans only clear exceptions and write one line each against flagged measures. Five: buy a reporting platform and rebuild the pack inside it.

The set contains carrying on, because a brief that excludes the status quo has already decided, and it contains at least one option somebody credible would argue for, which is the fifth. Two options were considered and excluded, and are named rather than omitted: outsourcing production, excluded because assembling it is where the function learns what is true; and asking the board to move its date, excluded because the cadence is fixed by the group calendar and the instruction is to work inside it.

A note on scope, since it decides the scoring. Options three and four require the reconciled data layer, the versioned definitions and the access layering to exist. Those are not free and any answer proposing them must include the cost of building them rather than assuming them into place.

Five criteria, weighted, with the method of scoring named for each. The threshold is sixty out of a hundred: below that an option is not recommended regardless of comparison, because continuing honestly with the current pack is always available.

Criterion three carries a floor in addition to its weight. A design that leaves the operational measures reporting the month before last scores below half on it and is not recommended, whatever it earns elsewhere. Automating a fortnight of assembly and keeping the five week lag would be an expensive way to preserve the problem.

The board date does not move. Second Tuesday, papers five working days ahead, including in the months where that lands on the 1st.

Headcount is flat this financial year. The replacement is funded by retiring existing outputs, not by hiring, and the answer names what is retired.

Commissioner reporting is unaffected. Contractual returns keep going out in their required formats, on time, throughout any transition.

No identifiable participant data goes into board papers. Aggregate only, with a minimum cell size stated in the design.

Every number on the page carries its basis: sourced, derived, provisional or indicative. A figure with no basis does not go on the page, whoever asks for it.

The written instructions are the deliverable, not a by product. They must be sufficient on their own: a person who has read them should be able to run the cycle without asking anybody, and they are kept with the process rather than in somebody head or somebody inbox. Where a reader still has to ask, the instructions are amended, not the person.

The format is frozen for twelve months. Presentation is agreed once, in the design, and is not reopened in a production cycle. A request to change the layout is a change request for next year edition, not a reason to re-cut this month.

A dispute about whether a number is correct is resolved by the definition and the reconciliation rule, logged, and it does not delay the pack. A dispute about what a number means belongs in the board meeting, which is the forum that exists for it.

Accepted has a written meaning: named signatory, stated checks, and a time. The pack is accepted when those are satisfied, not when the thread goes quiet.

Every edition is immutable and addressable, and every prior edition is reachable from the current one. The version index sits in the document itself, listing each earlier edition back to the first, including any reissue inside a month, so a reader never has to know where an archive lives. The measure level series is held outside the document, so a redesign of the page never breaks a trajectory.

The document stops travelling as an attachment on a standing email chain. It lives in one controlled location, access is by named role against the sensitivity layer, versions are numbered, and people are sent a link rather than a file. Participant level data is never attached to anything.

No data arrives as a message or an attachment. Each contributor either exposes its source for direct reading or publishes to the agreed location in the agreed shape, and the centre never holds a private copy that the owner cannot correct.

Every contributor can see what was taken from it, when, under which definition version, and how it appeared. Accountability for correctness sits with the owner, and it is only fair if the owner can see the consequence.

One cut off for every contributing department, published in advance, the same every month. What has not arrived by the cut off is shown on the page as not arrived, with the department named, rather than chased into the evening by a director.

Commissioner definitions are fixed. The group view is built by mapping contractual measures onto an internal model, never by asking a commissioner to count differently or by quietly redefining a contractual term internally.

Sensitivity is handled by layering, not by restricting the whole document. Operational aggregates, provisional financials and contract commercials have different audiences, and the design states which layer each measure sits in and who may see it.

The sections that judge the data function are reviewed by somebody outside its line before they go to the board. The director may write the pack; they may not be the only person who has checked the part that scores their own department.

And no definition may be changed in the run up to a board to make a measure land better. That is a constraint rather than a criterion, so it is not tradeable against anything.

Five mechanisms, each with how it works, who carries it and what it would take to measure. The first is the one that matters most: the scorecard becomes the operating model, so whatever is left off the page is left out of the company, and the thing most likely to be left off is the thing hardest to produce by the 1st.

The actual scorecard, drawn, at the size it would be read. Not a description of one. For every measure: its definition, denominator, exclusions, source, basis, owner, the decision it informs, the threshold at which the board would act, and the counter measure sitting beside it.

The process that produces it, end to end: what is pulled and from where, what is filed and by whom, the cut offs, the nine checks and what each does on failure, how the exceptions list reaches a human, and what that human is actually required to do. If the answer is more than an hour of work a month for anybody, say why.

A production calendar walked against a real month, showing for each measure the period it reports, whether it is final or provisional at the papers deadline, and what appears on the page to say so. Including the months where the papers are due on the 1st, which is where any pull forward will be hardest.

For every measure, a ruling on aggregation: does it combine across the four contracts, and if it does, what the combined figure means and how mix is separated from performance. Measures that do not aggregate are shown per contract or not at all, and that decision is stated rather than implied by the layout.

A retirement list. Which of the thirty eight pages and which recurring reports stop, who is told, and what capacity that releases.

An acceptance rule, in writing: what has to be true for the pack to be accepted, who signs, by when, and what happens to a late objection. Plus the dispute path that separates a correction from a disagreement about what the result means, so the second stops delaying the first.

A history model, navigable from the current edition rather than filed somewhere else: the version index in the document, every prior edition openable exactly as issued back to the first, the measure trajectory held independently of the page, and restatements shown when a published figure later changes. The first edition produced under the new process is also the baseline, and the answer should say so explicitly.

A transport and version model: where the document lives, how contributions arrive without an email chain, who may open which layer, how a version is identified, and what the board is sent. If it is still an attachment to fifteen people at the end of the design, the design has not finished.

A data contract for each of the seven contributors, written out: read directly or published, the exact location and shape, fields, grain, period, definition version, quality rules applied on arrival, the cut off, the named owner, and what the process does when the contract is not met. This is the artefact that replaces the email chain, and an answer without it has described the principle rather than applied it.

A collection model: for each measure, the contributing department, the named owner in it, the source system, the cut off, and the rule for what happens when two departments disagree, including who adjudicates and where that is recorded.

A production and access model: which layer each measure belongs to, who may see it, who assembles it, and what genuinely still requires the director and the department manager. State the senior days returned, because at twenty a month that number is the business case and it is larger than anything else in this instruction.

A rule for change: how a measure gets added, what it displaces, and who is allowed to refuse a director request for a permanent new page.

A position on each of the seventeen issues in the catalogue, and a statement of which of the five departmental goals the design serves, with any it cannot serve identified rather than quietly dropped.

The written instructions, at the level of detail somebody unfamiliar can follow, plus the timed proof: a named person who has never produced this pack, running one cycle from the instructions alone, with the elapsed time recorded and every question they had to ask listed. Each question is a defect in the instructions, and the answer says how each was closed.

A dry run: the scorecard produced for the month just gone, with real numbers, before anybody is asked to approve the design. Including the parts that will embarrass somebody, because a scorecard that is comfortable in its first month was designed to be.

And a named owner at a level that holds both sight and control. On the actor model that is group, and this is one of the few decisions where group can genuinely see what it is deciding about.