What a business dashboard should actually show

The tell is simple. Watch someone open their dashboard. If they look at it for four seconds and close it, the dashboard is decoration. If they open a second tab to check something the dashboard should have answered, it is incomplete in the one place that mattered.

Both failures come from the same source. The dashboard was built around what the tools could export rather than around a question somebody actually had.

What belongs on the screen and what does not?

Start from decisions, not from data sources.

Write down the three decisions you make repeatedly. Whether to spend more on a channel. Whether to add a shift. Whether to chase a slow month. Each decision needs specific inputs, and those inputs are your tiles. Everything else comes off.

A working tile has four properties. It has an owner. It has a comparison period, because a number alone means nothing. It has a threshold that would trigger action. And it has a known refresh rate, so you know whether you are looking at yesterday or last month.

A tile without a threshold is a tile nobody will ever act on. "Sessions: 4,820" produces no response. "Sessions: 4,820, down eleven percent against the four-week average, alert below 4,000" produces one.

Cut anything that measures activity rather than outcome. Impressions, followers, posts published, emails sent. Those belong in a vendor's work log, not on your screen. The difference between a metric that predicts revenue and one that only describes effort is the single largest quality gap between dashboards.

How should a dashboard show uncertainty?

This is the part almost nobody builds, and it is the part that separates an honest dashboard from a persuasive one.

Every number on your screen belongs to one of three categories, and the dashboard should say which.

Measured. First-party, deterministic, countable. Revenue from your point of sale. Form submissions in your database. Calls that connected, with duration. Server log entries verified by IP. You can defend these to an accountant.

Estimated. Modelled, sampled, or attributed. Ad platform conversions using modelled data. Analytics sessions after consent loss and blockers. Any keyword ranking, since rankings vary by location and personalization. These are useful for direction and worthless for arithmetic.

Unknowable. Things vendors will happily sell you a number for anyway. Your share of AI assistant answers. Word of mouth. Brand awareness in your city. The right treatment is a label saying you do not know, not a confident figure with a decimal point.

The line between those categories is not a matter of taste, and for search you can draw it from the platform operators' own documentation. Google Search Central's June 2026 announcement of Search Generative AI performance reports in Search Console lists the dimensions it publishes: impressions, pages, countries, devices and dates. Clicks, click-through rate, position and the user's prompt are not available, and AI Mode is not separated from AI Overviews. On the local side, Google's Business Profile performance report publishes views, searches, calls, website clicks, direction requests, messages and bookings, and publishes no rank or position dimension at all.

So a tile reading "AI Overview clicks" or "average local pack position, weighted" is not a hard tile to build. It is a tile reporting a quantity the platform does not emit. Colour code them, footnote them, or group them into separate rows. What matters is that nobody in a Monday meeting argues about a number that was never precise enough to argue about.

Should any tile compare me to an industry benchmark?

Almost never, and the local search category is a clean demonstration of why.

The only large study of Google Business Profile performance with a disclosed sample is BrightLocal's Google My Business Insights Study, published 15 July 2019, covering 45,264 local businesses across 36 industries in the US, UK, Canada and Australia. Its headline figure is a view to action conversion rate of 4.68%. It is still the number that circulates, seven years later.

It cannot be reproduced on a modern account, and the reason is worth understanding because it will happen to your dashboard too. Per Sterling Sky's documentation of the change, Google removed Photo Insights, the direct versus discovery breakdown, and direction request location data when Google My Business Insights became Google Business Profile Performance. Views is now a deduplicated unique visitor count, with Google stating a person can only be counted once a day. The 2019 denominator and the 2026 denominator are different quantities with the same label. Comparing your rate to 4.68% is comparing two things that were never the same measurement.

The alternatives are worse. Birdeye publishes a Google Business Profile action mix, and states no sample size and no methodology anywhere on the page while gating the report. WebFX discloses only "proprietary industry benchmarks" from in-house performance data. PlePer, SOCi and Yext publish no open benchmark with a disclosed sample. There is no current published average for an independent local business that is worth putting on your screen.

The replacement is your own trailing ninety days, and it is better anyway, because it controls for your market, your category and your competitors without you having to model any of them.

What does the honest version of that tile look like?

Here is the arithmetic. Pull it from your own Google Business Profile performance report.

This quarter. 4,200 profile views, 168 actions, where an action is a call, a website click, a direction request, a message or a booking. That is a 4%derived action rate.

Against the published benchmark. 4.68% from a 2019 study, so you appear to be underperforming by roughly two thirds of a point, and somebody in the meeting will now ask why. That question has no answer, because the 2019 views figure counted views and yours counts unique daily visitors.

Against yourself. The prior quarter was 3,900 views and 150 actions, a 3.8%derived action rate. Views are up 8%derived and actions are up 12%derived, and the rate improved.

What the tile should say. Actions 168, prior quarter 150, normal band 138 to 175, alert below 130. Source: GBP performance report. Category: measured. No benchmark row at all.

Run it with your own numbers. Take four quarters of views and actions, calculate the rate for each, and use the spread as your band. That takes about fifteen minutes and it retires the benchmark conversation permanently.

Which sources should feed it?

Fewest possible, and every one of them first party where you can manage it.

Your transaction system is the anchor. Whatever holds money changing hands is the number everything else gets compared against. Marketing tools should be judged against it, never the reverse.

Your call data belongs here too, because for most local businesses the phone is the highest intent channel and the least measured one. It is also the one currently under the most pressure. Sterling Sky's State of Local SEO in 2026 reports that clicks to call from Google Business Profiles are declining even for profiles with stable rankings, using Jepto data compiled from 179 profiles across 34 US law firms over two years, and that website clicks did not fall the same way because call icons dominate mobile while website icons appear on desktop. If your dashboard carries one call tile and no context, a mobile-side decline will read as a marketing failure. Getting calls counted as conversions properly usually moves more than any other single dashboard improvement.

Then one traffic source and one ad source, tagged consistently. Not five.

And the final page in your conversion path, which is a source of truth most sites throw away. A thank you page that only says thank you is a missed measurement point as well as a missed sales opportunity.

Why do dashboards get abandoned?

Because they were built for a moment rather than for a habit.

Somebody spends three weeks connecting sources, the dashboard looks impressive, it gets shown at a meeting, and by week six nobody opens it. The build was the project. The routine was never designed.

The fix is the same one that keeps a kitchen clean on a Friday night. The behaviour has to be attached to something that already happens, with an owner and a visible consequence when it does not. Operating controls hold when they survive the worst night of the week, and a dashboard is an operating control, not a piece of software.

The second reason is drift, and it has two forms. The first is technical: a tracking change, a website rebuild, or a new booking tool breaks a feed, and the tile keeps rendering a number that no longer means what it did. The second is definitional, and it is the one that caught the 2019 benchmark above. A platform quietly redefines the metric, the tile keeps working, and your year over year comparison silently becomes a comparison of two different things. Every dashboard needs a last updated timestamp on each tile, a note of the metric definition, and someone whose job includes looking at both.

What to do this week

Print your current dashboard, or screenshot it. Cross out every tile that has not changed a decision since it was built. If more than half survive, you are unusual.

Add a comparison period to every surviving tile. Same period last year where seasonality exists, previous four-week average where it does not.

Label each surviving tile measured, estimated, or unknown. This takes twenty minutes and it will change how your next vendor conversation goes, because most reporting disputes are really disagreements about which category a number belongs to. It also makes reading a monthly marketing report critically a much shorter job.

Delete any tile whose comparison is an industry average you cannot trace to a sample size.

Then set one alert. One. A threshold on the tile that matters most, delivered by email. A dashboard that contacts you beats a dashboard you have to remember to open.

Questions about your dashboard?

Email me at eric@seod.com with the single decision you most want your dashboard to help you make. Not your data sources, just the decision. I will send back the three tiles that decision actually needs and the ones you can drop.

I answer these myself, usually the same day. No pitch, and no signup required to get an answer.

More on measurement sits in the analytics library.