# Instrumentation *The wrong dashboard* --- You've just joined as managing director, and on day three the monthly pack arrives. Forty pages. Revenue six ways, margin four ways, pipeline, utilisation, NPS, headcount, aged debt, a page of traffic lights. Each page was added by someone, after something: the debt page after a bad year for collections, the utilisation page after a consultant's review, the NPS page after a customer walked. You ask your finance director which numbers she'd keep if she could only keep five. She laughs, then goes quiet. The pack has ninety numbers and no opinion, and decisions are being made in the business anyway. --- An F1 car, a family SUV and a long-haul truck all have four wheels, a speedometer and a fuel gauge. Their dashboards are nothing alike, because the jobs aren't, and what's put in front of the driver is a decision about what the driver is supposed to manage. A monthly pack is the same object; most of them are inherited rather than chosen. --- Choosing your own starts from the processes, not the numbers. For any core process there are six things worth knowing: what's going in, what's coming out, how well the one converts into the other, how long it's taking, how much work is stuck inside, and whether what comes out is any good. A single metric will almost never tell you the whole story, because each of the six can look healthy while the process fails. Output jumps for a quarter because a backlog got flushed. Cycle time looks better because someone, under pressure, stopped doing the quality checks. Work-in-progress is the one almost nobody instruments, partly because the systems rarely capture it and fixing that is a project in itself, and it's the one that predicts the lead times everyone complains about ([[Flow and bottlenecks]] carries the law; [[Execution trap]] is a business discovering it). The six are a checklist for taking a process apart, not a quota for the monthly page: report the few that would change what you do, and sweep all six when something worries you. A page per core process, chosen that way, beats ninety numbers chosen by history. --- Then hold the line between metrics and analysis. Metrics are routine and cheap to produce, the same shape every month, read in minutes: the dashboard. When a needle moves outside its ordinary range, that's a question for analysis - bespoke, one-off, done when it's answered ([[Signal and noise]] is the discipline for deciding which movements deserve one). Forty-page packs are what happens when the two get conflated: every crisis's investigation becomes a permanent page, and nobody ever takes a page out. The rules I hold my own packs to are mostly about respecting the reader. Run charts wherever possible, because a trend is the fastest thing a human can read. Bar axes that start at zero. Comparisons against plan and prior year, where a plan exists. Definitions on the same page as the terms. Nothing presented that whoever presents it isn't willing to explain and defend. And commentary only where it adds interpretation, because most of it just rephrases the chart. --- The analytical rebuild is smaller than it looks: start from the two or three decisions you actually take each month, and work backwards to the numbers that would change them. Bergman & Beving ran a four-decade acquisition record with [[Bergman & Beving|one governing ratio]]. The organisational rebuild is slower, because every page has an owner - the FD who lived through the bad collections year will defend the debt page - and pages leave a pack more slowly than they join it. The test worth running either way: which numbers changed a decision in the last quarter? The honest answer is usually a handful, and most of the rest is old analysis nobody has taken out. ---