The dashboard is not the deliverable
A company spends somewhere between six and twenty weeks, and somewhere between $30,000 and $150,000, building an executive dashboard. It gets demoed on a Thursday. Three people say it looks great. Traffic is heavy for a week, weekly by the second month, and by the following quarter the only person opening it is the analyst who maintains it. We have been called into that room many times. The tool is almost never the reason. The dashboard failed because it answered questions nobody asks on a Monday morning, and because the answers waited in a place where nobody was standing.
It helps to be precise about what you are building. A report is a document you hand someone. A dashboard is a habit you are asking a busy person to form, and habits fail for reasons that have nothing to do with chart libraries. Four reasons come up over and over.
It arrived after the decision. The owner of a regional heating and cooling contractor decides on Thursday afternoon whether to pull two techs off a new-construction job and put them on service calls, because service is where the margin is and the backlog on the construction side is soft. A dashboard that refreshes Monday with last week's numbers is a history lesson. He needs Thursday's dispatch board and yesterday's booked-hours-versus-capacity, and he needs it before lunch.
It could not answer the second question. Gross margin is down 1.4 points. Which branch, which product class, which customers, and did we change a price or did the mix change? If the next click does not exist, the executive picks up the phone, and the analyst on the other end of that phone is the real dashboard.
Two numbers disagreed once. The dashboard said $1.84M and the sales report said $1.79M, and nobody could explain it inside ten minutes. Trust in a number is lost far faster than it is built. One unexplained disagreement sends an executive back to the spreadsheet they had before, usually permanently.
It measured things they do not control. A page full of market-share estimates and industry benchmarks is interesting, and interesting is not a reason to open something at seven in the morning.
You are probably here because
- You paid for a reporting project and the leadership team still runs on a spreadsheet
- There are now forty dashboards and nobody can say which one is right
- Two systems give two revenue numbers and both sides can defend theirs
- Every question ends with someone asking an analyst to pull something
The first three sections below are the fix for all four. None of them are about which reporting product you bought.
Watch how they get their numbers today
Before designing anything, spend two hours sitting with the person the dashboard is for while they get their numbers the way they get them now. Do not interview them. Watch them. What people say they need and what they actually open are different lists, and the second one is the specification.
You will usually find a spreadsheet. The owner of a beverage distributor opens the pick-exception report at 6:40 in the morning, scans for lines that shipped short, and calls the buyer about two of them. The operations director of a four-location dental group pastes yesterday's production and chair hours out of the practice management system into a workbook that computes production per chair hour by provider, and looks at one column. That workbook has been refined over four years by the person who needs the answer. It encodes maybe eight decisions and a dozen exclusions nobody wrote down.
Record two things for every number they look at: what it is, and what they do next. A metric with no next action is decoration, and decoration is what fills the dashboards nobody opens.
Decisions first, metrics second
The order matters and almost everyone gets it backwards. Do not start with a list of key performance indicators. Start with a list of the decisions this person makes on a repeating cycle, then work backwards to the smallest number of figures that would change one of those decisions.
For the heating and cooling contractor the recurring decisions are: do I need another truck next month, which open jobs are bleeding against estimate, who is going to miss a service commitment today, and am I going to be short on cash in three weeks. That is four decisions. They map to backlog in crew-weeks, job margin variance against estimate, at-risk work orders by promised time, and a rolling thirteen-week cash forecast. Four numbers, not forty.
| Business | The recurring decision | The number that answers it | Cadence |
|---|---|---|---|
| Beverage distributor | Which accounts do I put a route salesperson in front of this week | Volume trend by account against same period last year, plus lines cut for stockout | Monday morning |
| Four-site dental group | Do I move a hygienist between sites next week | Booked chair hours against available, per site, two weeks out | Thursday |
| Specialty contractor | Do I take the next job or stop bidding for a month | Backlog in crew-weeks, by trade | Weekly |
| Eleven-unit restaurant group | Which general manager do I call today | Labor as a percent of sales, by unit, against target and against the same weekday | Daily by 9am |
| Contract manufacturer | Do I quote a four-week lead time or a seven-week one | Committed machine hours against capacity, by work center | Daily |
Notice what is missing from that table: revenue. Revenue belongs on the page, because everyone wants to see it and it costs one tile. It is rarely the number that changes a decision this week.
Twelve numbers, and the discipline to stop there
The first screen should hold eight to fourteen figures and fit without scrolling on a laptop. That constraint is the whole design. Every tile you add makes the others harder to see, and the pressure to add one more comes from every direction — a VP wants their function represented, someone read an article about a metric, the analyst is proud of a calculation.
The test we use is blunt. If that number moved twenty percent and nobody would do anything differently this week, it does not belong on the first screen. It can live one level down, where the people who work in it can find it.
Second screens are cheap and first screens are not. Build a shallow tree: one executive page, then a page per function with the twenty or thirty figures the function actually manages, then the transaction detail. Most companies build the middle layer first and never build the top one.
What decides whether it is still open in six months — our weights
Weights sum to 100. Where we put design attention on a first build, from what we have seen survive and not survive. Judgment, not measurement.
A number with no comparison is a noun
"Revenue: $1.84M" tells a reader nothing. Against what? Three comparisons carry almost all the meaning: the prior comparable period, the plan or target, and a peer unit. Labor at 31.2 percent of sales in a restaurant is a fact. Labor at 31.2 percent against a 29 percent target, while the unit two exits down the highway ran 27.8 percent on similar covers, is a conversation with a general manager on Tuesday.
Pick the comparison that matches how the business actually moves. Retail and food service compare to the same weekday, because Tuesday is not Saturday and a month-to-date figure hides that a bad Friday cost more than three good Mondays. Distributors compare to the same week last year, because seasonality swamps everything else. A contractor compares to the estimate, because the estimate is the only number that was ever promised to anyone.
The second question is always "why"
Every number on the page needs a path down to the rows underneath it. Margin down 1.4 points, click, margin by branch, click, margin by product class in the branch that moved, click, the actual invoice lines. Two or three clicks from the tile to a list a person can read and act on.
This is the single most-skipped piece of a reporting build, usually because the summary was assembled in a way that cannot be decomposed — an aggregate loaded from somewhere with the detail dropped, or a spreadsheet extract that stops at a subtotal. It costs real work to keep the grain all the way down, and it is the difference between a tool and a picture.
Delivery beats login, almost every time
An executive who has to remember to go somewhere will stop remembering. The reporting that gets read is the reporting that arrives. A short email at six on Monday morning with nine numbers, each with its comparison and a direction, and a link into the detail, will outperform the same content behind a login by a wide margin. In the businesses we have worked in, a delivered weekly digest holds most of its readership indefinitely; the same content requiring a login typically settles somewhere well under a third of the leadership team, and lower for anyone who spends the day out of the office.
Exception alerts are the other half, and they need more discipline than anything else on this list. An alert that fires forty times a month is noise inside one quarter, and it also teaches people to ignore the next one. Budget two to four alerts per person per week, set thresholds from the last year of actual data rather than from a round number that felt right, and put an owner's name on each one. An alert nobody owns is a notification.
Freshness belongs on the face of it
Put two timestamps on the page: when the data was last loaded, and what period it covers. They are different, and confusing them causes the disagreements that kill trust. A distributor's invoices post overnight, so yesterday's revenue is genuinely incomplete until nine in the morning. Say so on the page — "orders through 11:59pm yesterday, invoices through 6am today" — rather than letting someone discover it during a board call.
The same goes for anything that restates. Payroll allocations, rebates, warranty accruals and credit memos all land after the fact. If last month's number can still change, the page should say by roughly how much it usually changes, and it should never quietly rewrite a figure someone screenshotted.
The definition fight is the actual project
This is where these builds really die, and it is not a technical problem. What is revenue? Booked, shipped, invoiced, or recognized? Net of credits? Does freight count? Does the intercompany transfer between two of your own locations count once, twice, or not at all? What is an active customer — bought in ninety days, or has an open agreement? Two systems will produce two answers and both will be defensible, which is why the argument never resolves on its own.
Somebody with authority has to decide, in writing, and the analyst is not that person. Produce a one-page entry per metric: the name, a plain-English sentence a new hire could read, the source system, the filters, the exclusions, and the name of the person allowed to change it. It is dull work, it takes about a week for a mid-size company, and it prevents the forty-minute meeting where the finance lead and the sales lead argue about a number while everyone else waits.
The definition document is the part clients undervalue and later reuse the most
It outlives the dashboard. When the reporting product gets replaced in four years, or the company buys another business and has to map its chart of accounts, the definitions are the asset that carries over. They also settle disputes with customers and lenders. Write them as prose a person can read, not as a copy of the SQL, and keep them where a non-technical person can find them.
It will be read on a phone in a parking lot
Assume the first read of the day happens on a phone: in a truck at 6:40, in a parking lot before a site visit, in the ten minutes before a meeting. That means the executive layout is a single column of large tiles, big type, and no horizontal scroll. Interactions built around hover, cross-filtering and multi-select do not survive a thumb. They are fine one level down where an analyst works at a desk.
Load time is part of adoption too, more than people expect. Under about three seconds and nobody thinks about it. Past ten and the habit does not form, no matter how good the content is. If a page is slow because it is recomputing a year of transactions on every open, pre-aggregate it — the executive layer is the one place where a nightly rollup is almost always the right answer.
Let people write on it
Add a comment field to each period. Why was the third week of August bad? "Two techs on vacation, one truck in the shop four days, and we lost the hospital account's spring order to a lead time." Six months later, nobody remembers, and someone builds a whole analysis to rediscover it. Annotation is the cheapest feature on this list and it is left out of nearly every build we inherit.
When you should not build anything
There are real cases where the honest answer is that software is not the problem.
You are small enough that a person is faster. One location, under roughly $10M in revenue, one controller who already sends a Monday spreadsheet that everyone reads. A reporting project here often replaces something that works with something that requires maintenance. Improve the spreadsheet, lock the formulas, put it on a schedule, and spend the money somewhere else.
The underlying data is wrong. If half your work orders are closed without labor hours, or the item master has three spellings of the same customer, a dashboard will publish that to the whole leadership team and everyone will conclude the dashboard is broken. Fix the capture at the source first. That is usually a process and training problem with a small software component, not the reverse.
The real problem is a person, not a number. Sometimes an executive wants a dashboard because they do not trust a manager's reports. Software will not settle that, and building it makes the tension worse because now there are two versions of the truth and a fight about which is right.
Send us the spreadsheet you actually use.
Email the workbook your leadership team runs on, plus a sentence about the decisions it feeds, to contact@precisionfederal.com. You get back a short written note: which numbers we would put on the first screen, which we would cut, and where the definitions are going to cause an argument. One business day. No charge and no meeting.
contact@precisionfederal.comThe mistakes we get called in to fix
- Forty dashboards and no owner, so nobody can say which one is current
- A figure that changes depending on the hour you open it, with no timestamp explaining why
- Averages hiding the distribution — average days to pay is 41, and a third of the balance is past 60
- Metrics copied from an industry article that no one in the building can act on
- A first screen that takes 90 seconds to load because it recomputes three years on every open
- Filters that reset, so the executive re-selects their region every single time
- No drill path, which quietly turns the analyst back into the reporting system
- Charts where a table would be clearer — eleven units and one number each is a list, not a bar chart
A four-week build that leaves a habit behind
Executive Reporting: First Cut
The parallel run is the step teams cut when they are late, and it is the step that earns the trust. Every disagreement between the new page and the spreadsheet has a cause, and the cause is nearly always a definition rather than a bug. Write down each one and what you decided. When the executive asks in November why the number differs from what they remember, that document is the answer.
Before you call it done
- Every tile traces to a decision somebody makes on a cycle
- Every number carries a comparison — prior period, target, or peer unit
- The drill path reaches actual transactions in three clicks or fewer
- Load and coverage timestamps are visible on the page
- Each metric has a written definition and a named owner
- The executive layout is readable on a phone with one hand
- A digest is delivered on a schedule, not waiting for a login
- Alerts have thresholds derived from history and an owner per alert
- Users can leave a note on a period, and the notes persist
- A parallel run against the old spreadsheet closed with every difference explained
Bottom line
Executive reporting fails as an adoption problem far more often than as a technical one. Start from the decisions, not the metrics. Keep the first screen small enough that everything on it matters. Put a comparison next to every number and a path underneath it to the rows that produced it. Deliver it rather than waiting for someone to log in. And settle the definitions in writing with a person who is allowed to settle them, because the argument you do not have in week two is the argument that ends the project in month five. Those five decisions matter more than which reporting product you buy, and none of them are expensive.
Frequently asked questions
Eight to fourteen on the first screen, fitting without scrolling. The test for inclusion is whether a twenty percent move in that number would change something anyone does this week. Everything else belongs one level down on a functional page, where the people who manage it will look for it.
Usually one of four things: it answers questions nobody asks on the cycle they work in, it cannot answer the follow-up question without an analyst, it disagreed with another source once and lost trust, or it requires them to remember to go somewhere. The last one is fixed by delivering a short digest on a schedule.
Rarely, at the executive layer. Match the refresh to the decision cycle. A restaurant group deciding who to call needs yesterday complete by nine in the morning. A contractor reviewing backlog weekly does not need anything faster than nightly. Real-time data costs several times as much to build and run, and it adds noise to decisions made weekly.
Both, usually. They are measuring different things — booked versus invoiced, gross versus net of credits, with or without freight and intercompany. The fix is not a reconciliation script. It is a written definition, owned by a person with the authority to set it, and a note on the page saying which definition is being shown.
Often, yes, and it is worth trying before buying anything. Built-in reporting is weakest when you need to combine the ERP with something else — a payroll system, a CRM, a scheduling tool, a spreadsheet of targets. If everything you need lives in one system and the drill path works, the cheapest good answer is the one already in the license you pay for.