Dashboard commentary that writes itself:
what moved, and why it might matter
A dashboard shows the numbers but rarely says what they mean, so people stare at a chart trying to decide if a dip is a problem or ordinary noise. An agent adds a short written note next to each metric that moved: what changed, how it compares to normal, and what might be worth checking.
The process today
A dashboard is good at showing a number and bad at saying whether that number is a problem. A line dips, and the person looking at it has to decide, on the spot and usually without the full history memorized, whether that dip is within the range this metric normally wanders in or something that actually needs attention. Different people looking at the same chart make different calls, and the same person on a Monday morning reads a chart more carefully than the same person checking it quickly between meetings on a Thursday.
This gets worse as dashboards accumulate metrics. A team that started with five key numbers on one screen ends up with thirty across several tabs after a year of adding “just one more chart,” and nobody has the bandwidth to actually look closely at all thirty every day. The metrics that would benefit most from a careful read, the ones with a genuine anomaly worth investigating, get the same thirty seconds of attention as the ones sitting exactly where they always do, because there is no signal distinguishing “normal” from “worth a second look” until a person manually checks the history.
The deeper cost is that a dashboard full of numbers without context trains people to stop really looking at it. After a few months of glancing at charts without time to properly interpret them, checking the dashboard becomes a habit rather than an analysis, and a real problem can sit visible on the screen for days before anyone connects the dots, not because the data was hidden, but because nobody had the moment of attention that turning a number into a sentence would have forced.
What the agent does
The agent reads the same dashboard your team already looks at and, for each metric, calculates a baseline from that metric’s own history: what range it normally moves in, accounting for known patterns like weekday cycles or seasonality where relevant. Every time the dashboard refreshes, the agent writes a short note next to each metric: whether the current value sits inside or outside that normal range, and by how much, in a sentence rather than forcing the reader to eyeball a chart and estimate.
Where a metric move correlates with another metric that also moved, a spend change lining up with a conversion shift, a traffic spike lining up with a campaign launch, the commentary names the correlation as a possible explanation, with the supporting numbers shown, rather than stating it as settled fact. This is deliberately a hypothesis a person can accept or dismiss in seconds, not a conclusion the dashboard asserts on its own authority.
The commentary lives wherever your dashboard already lives, as an annotation directly on the chart if the tool supports it, or as a companion summary that sits alongside the dashboard and updates on the same refresh schedule. Nothing about the dashboard’s actual data changes; the commentary is a layer on top that turns “here is a number” into “here is a number, and here is whether it is normal,” which is the part a person was previously doing manually and inconsistently.
For teams juggling several dashboards across different parts of the business, the same approach runs on each one, so the handful of numbers that actually moved outside their normal range this week surface consistently, instead of depending on which dashboard tab someone happened to open first.
What stays with humans
Deciding whether a flagged anomaly warrants action, judging a correlation the commentary surfaces against context the data itself does not contain (a known one-off event, a planned change), and any decision that follows from the dashboard stay entirely with the people who own that part of the business. The agent explains what the numbers show; it does not decide what to do about it.
Guards
Every explanation shown is tied to specific, visible numbers, not a bare assertion, so a person can check the reasoning in seconds. The normal range used for each metric is calculated from that metric’s own real history and gets reviewed and retuned after the first couple of weeks of use, so the commentary does not either cry wolf on ordinary noise or stay quiet on something that genuinely moved. Access to the underlying data stays read-only and scoped to the dashboard being commented on.
Price and timeline
| Option | Price | What it covers | Timeline |
|---|---|---|---|
| Single automation | from $500 | One dashboard, written commentary on each key metric, refreshed with the dashboard | 3 to 8 days |
| Department package | from $2,500 | Commentary across several dashboards plus weekly plain-language reports for the team | 2 to 4 weeks |
Running cost is usually $10 to $40 a month in model usage depending on dashboard refresh frequency and metric count, with a budget cap set before launch.
Related
This is the always-on companion to weekly reports in plain language, which rolls the same kind of analysis into a scheduled narrative, and it depends on the same clean, joined data that data cleaning and deduplication is meant to guarantee underneath it. See the automation-everything overview and the AI agents service page for the full catalogue. The two-brand analytics warehouse with an AI analyst in Telegram answers exactly this kind of “is this normal” question on demand, and the trading app funnel audit shows how much a correctly joined baseline changes what a metric move actually means.
If your dashboard has thirty metrics and nobody has time to properly read all of them every day, get in touch and we will show you what commentary on top of it would actually look like.
Tired of doing this by hand? We can take the whole routine off your team, not just this step: Routine takeover, from $400 →
FAQ
How much does dashboard commentary automation cost?
From $500 to add written commentary to an existing dashboard with a defined set of metrics. Covering several dashboards or adding cross-metric explanations across a wider data model is $1,200 and up.
How long does setup take?
3 to 8 days, most of it spent establishing what a normal range looks like for each metric from its own history, so the commentary does not flag ordinary variation as something noteworthy.
Which dashboard tools does it connect to?
Looker, Metabase, Power BI, Tableau or a custom dashboard built on your own warehouse, through whichever connection method the tool supports: a direct database read, an API, or an export the dashboard already generates.
What if the AI's explanation for a metric move is wrong?
Every explanation is phrased as a hypothesis tied to the specific correlated data that supports it, not a confident claim, and the underlying numbers are always visible next to the commentary so a person can judge whether the explanation actually holds up.
Does this expose our dashboard data anywhere else?
The commentary reads from your dashboard or warehouse with a scoped, read-only connection and writes back only into the dashboard or the companion summary you choose. No separate copy of the underlying data is kept outside that pipeline.