9 B2B SaaS dashboards that help people act, and what they do differently

Nine B2B SaaS dashboards that help people act, from Salesforce to Search Console. What each does differently, and where the pattern breaks.

  • Enterprise & Workflow UX
  • B2B SaaS and workflow software
  • B2B SaaS & Workflow Software
  • Data, Analytics & Decision Products
Insights

Nine B2B SaaS dashboards that help people act, from Salesforce to Search Console. What each does differently, and where the pattern breaks. A dashboard full of charts nobody acts on can have the right data and the wrong unit of attention.

Insights · Article

A dashboard full of charts nobody acts on can have the right data and the wrong unit of attention. It shows current values when the person needs to know what moved, against what, and which records to open. We read the public documentation for nine B2B products whose dashboards or reporting surfaces are built for someone to do something next, and set out what each one does for the person using it and what a product team can take from it.

We started with a hunch: the dashboards that help people act are built around a question or a queue rather than a set of metrics. The research mostly supported it, with two corrections. Few of these products tell you what to do; most narrow the problem and leave the judgement with the person. And a queue creates its own noise unless it has a way for items to leave.

What the useful ones do differently

Across the nine examples, four habits recur:

  1. They start from a change against a stated baseline. The reference point is visible: the start of the period, an expected range, a forecast.
  2. They turn exceptions into a list with states and exits. Items arrive for a reason, carry an owner or a deadline, and can be resolved, parked or closed.
  3. Every number opens the records behind it. Selecting a bar, a metric or a row shows the opportunities, issues or URLs it counts, in several of these products with the same filters applied.
  4. They say what the number means. Definitions, exclusions, the direction that counts as good, and the limits of the data are written down where the user can find them.

One thing was rarer than we expected. Explicit next steps appear where the decision is closed and the rules are external, such as responding to a card dispute or following a customer success playbook. Elsewhere, the product narrows the problem to these records, this cause or this owner, and stops there. We think that restraint is part of why they are trusted, though no source we found measures it.

ProductSurfaceWhat it helps someone do
SalesforcePipeline InspectionSee which deals changed since a chosen point and open them
TableauTableau PulseFollow a metric with a declared baseline and direction
AmplitudeAnomaly + Forecast, Root Cause AnalysisCheck whether a spike is unusual, then find the segments involved
SentryIssuesWork through new, escalating and regressed errors without losing parked ones
GainsightCockpitTurn a red health score into a task with an owner
StripeDisputesDecide whether to accept or counter a dispute before a deadline
LinearInsightsRead a chart beside the issue list and open the issues behind any bar
GoogleSearch Console Page indexing reportSee why pages are not indexed, which causes are yours to fix, and confirm the fix
GitHubSecurity overviewTrack detection, remediation and prevention with published metric definitions

How we assembled this. Each example is based on the vendor's own documentation, linked in the text, checked on 28 September 2026 and re-checked on 13 October 2026. We describe only behaviour the documentation states. We did not use these products with live data for this article, and none of the pages we read publishes a measured effect on users' decisions, so we make no outcome claims.


Start from what changed, and say compared with what

A current value asks the reader to remember what it was last week. These three products put the comparison in the product.

Salesforce Pipeline Inspection: each metric opens a list of deals

What it does for the person using it. Salesforce describes Pipeline Inspection as a consolidated view of pipeline metrics, opportunities and week-to-week changes in close dates, amounts, stages and forecast categories (Salesforce Help, checked 28 September 2026). A sales manager can click a pipeline metric such as Overdue or Increased to see the related opportunities, and highlights signal changes to amounts, stages or close dates. An insights panel draws on call recordings, emails and service cases (Trailhead, "Review deal health with Pipeline Inspection", checked 28 September 2026). A "Changes Since" quick filter sets the reference point, for example the start of the period, and reps can edit inline while activity is captured automatically (Trailhead, "Customize your Pipeline Inspection experience"; Trailhead, "Get to know Pipeline Inspection", both checked 28 September 2026).

What to take from it. The metric bar works as a set of doors rather than a scoreboard. "Overdue" is already a question, which deals slipped, and clicking it answers with the deals. The reference point for "changed" is visible and the user can move it. Our inference: this depends on the product knowing which records changed and when. If your data model stores only current values, a "what changed" view is a data project before it is a design project.

Tableau Pulse: the direction that counts as good is part of the definition

What it does for the person using it. When someone defines a metric in Tableau Pulse, they choose a time dimension, set definition filters that viewers cannot adjust, and answer "Value going up is" with neutral, favourable or unfavourable (Tableau Help, "Create metrics with Tableau Pulse", checked 28 September 2026). Pulse then detects insight types such as Period Over Period Change, Unexpected Values (a value higher or lower than the expected range "based on historic observations"), Top Contributors and Record-level Outliers, which require a record identifier such as an order ID. When insight summaries are on, only Period Over Period Change and Unexpected Values feed the plain-language summary at the top of digests, and, where Tableau Agent in Pulse is turned on, question pages show "key insights, relevant visualizations, source references, and suggested follow-up questions" (Tableau Help, "Insight types in Tableau Pulse", checked 28 September 2026).

What to take from it. The most consequential design decisions happen before any chart is drawn: whether up is good, which filters are fixed, what the baseline is. Without them, the product cannot tell whether a rise is good or bad news, and neither can a summary written from it. Restricting the summary to two insight types is also a decision about attention. The record identifier requirement is the route back to the data: an outlier the product can name is an outlier someone can open.

Amplitude: "is this unusual?" comes before "why?"

What it does for the person using it. Amplitude's Anomaly + Forecast draws a light blue confidence band and a dashed expected value beside the actual line, with anomalies in orange outside the band. The Agile mode uses a 95% confidence interval and 120 days of training data; the Robust mode uses a further year of data and accounts better for seasonality (Amplitude Docs, "Anomaly + Forecast", checked 28 September 2026). From an anomalous point, a user can run Root Cause Analysis, which scans the event's properties against a baseline, surfaces the segments involved and adds external context such as country holidays and product releases. It supports Event Segmentation charts only and is limited to certain plans (Amplitude Docs, "Root Cause Analysis", checked 28 September 2026).

What to take from it. Two questions are separated: whether the movement is outside normal variation, and which part of the population moved. That order stops a team from explaining noise. Amplitude describes the output as the properties and segments that "explain" an anomalous point. What such an analysis can show is which segments moved with the anomaly; it cannot by itself establish cause. Our judgement is that the interface language should keep that distinction: "concentrated in" or "associated with" rather than "caused by".


Turn the exception into a queue with a way out

A queue answers "what needs attention?" directly. These three products show that the queue also needs rules for how items leave it, or it becomes another wall of red.

Sentry Issues: parked items come back only if they get worse

What it does for the person using it. Sentry's Issues feed can be filtered to For Review (new issues or regressions that have not been reviewed), Regressed, Escalating and Archived, and has a Recommended sort that puts "the issues most relevant to you" first (Sentry Docs, "Issues", checked 13 November 2026). An issue is New for its first seven days and becomes Escalating when it "exceeded its forecasted event volume". Archiving defaults to "until escalating", so a parked issue returns if events surge (Sentry Docs, "Issue states and triage"). Priority starts from log level and is raised automatically when an issue escalates, but "once an issue's priority has been changed manually, it will no longer be automatically adjusted" (Sentry Docs, "Issue priority", both checked 28 September 2026).

What to take from it. The queue has exits (resolve, archive) and a re-entry rule based on a forecast, so a person can clear the list without losing sight of what they cleared. The manual override is a small, telling decision: when a person has judged an item, the automation stops second-guessing them.

Gainsight Cockpit: a score becomes a task with an owner

What it does for the person using it. In Gainsight, a Call to Action (CTA) "is tied to a customer and represents an action that a CSM must take". CTAs can be created by hand or triggered by rules on signals such as a usage drop, support cases, a change in sponsor or survey results, and playbooks attach a predefined set of tasks (Gainsight Support, "CTAs, Tasks, and Playbooks Overview", checked 28 September 2026). If an admin sets it up, a health score turning red produces a CTA in Cockpit; high sentiment can prompt a CSM to consider the customer as a reference (Gainsight Support, "Scorecard FAQs"). For teams with too many CTAs, Gainsight's own advice is to cut less relevant ones, make playbooks leaner and auto-close CTAs after a set number of days (Gainsight Support, "Cockpit FAQs", both checked 28 September 2026).

What to take from it. The health score is not the end point; it is the trigger for a named piece of work with an owner, and good news routes to action too. The FAQ is the part worth copying: the vendor expects overload and documents how to prune. A rules-driven queue needs someone responsible for the rules.

Stripe Disputes: the screen is a decision with a deadline

What it does for the person using it. Disputes appear on the Disputes tab of the Stripe Dashboard, and each dispute's details page shows its category, which determines the evidence Stripe recommends. Stripe says the response window is usually 7 to 21 days depending on the card network, and missing it loses the dispute automatically. The person chooses Accept dispute or Counter dispute. Countering opens a form that asks why the dispute is in error and the product type of the original purchase, then shows the evidence sections relevant to that case, pre-filled where the integration supplied the data. Stripe flags when a fraud dispute is eligible for Visa Compelling Evidence 3.0, and warns that there is only one opportunity to submit (Stripe Docs, "Respond to disputes", checked 28 September 2026).

What to take from it. This surface can say what to do next because the decision is closed (accept or counter), the rules come from the card networks, and the deadline is fixed. Most product dashboards do not have those conditions, which is why prescribing an action there is riskier. Where they do hold, the Stripe pattern applies: show the deadline, the options, the consequence, and gather the evidence for the person rather than asking them to find it.


Keep the route from the number to the records

A chart that cannot be opened sends people to export, filter and rebuild the view elsewhere. These three keep the route inside the product.

Linear Insights: the chart sits beside the work

What it does for the person using it. Linear Insights opens in the right-hand sidebar of issue views, including team, project, cycle and custom views. Measures include issue count, effort, cycle time, lead time and triage time. Selecting a bar or segment filters the view to those issues, a table below the graph highlights the matching rows, and selecting a point on a scatterplot opens the issue. Data can be exported to CSV (Linear Docs, "Insights", checked 28 September 2026).

What to take from it. There is no separate reporting destination to remember. The chart is a way of looking at the list the person is already working in, and every mark is a set of issues that can be opened.

Google Search Console Page indexing report: grouped by cause, and by who can fix it

What it does for the person using it. The report shows indexed and non-indexed page counts over time, with a "Why pages aren't indexed" table listing each reason with its source and validation status. The source column matters: "In general, you can fix only issues where the source is listed as 'Website'." Opening a reason shows an issue-specific chart and up to 1,000 example URLs, each linked to the URL Inspection tool. After fixing an issue, the user can click Validate fix, which Google says usually takes up to about two weeks. The help page also warns: "Don't expect every URL on your site to be indexed" (Search Console Help, "Page indexing report", checked 28 September 2026).

What to take from it. The rows are causes, not metrics, which is closer to how people decide what to fix. The source column tells them which rows are theirs. Validation closes the loop, so the user finds out whether the fix worked without returning to watch the chart. The warning is interpretation built into the product: a large non-indexed count is not automatically bad news. The report is also candid about its limits, stating that the example list is capped.

GitHub security overview: organised by stage of response, with definitions published

What it does for the person using it. GitHub's security overview dashboard has Detection, Remediation and Prevention tabs, and every metric on the page responds to the date range and filters (GitHub Docs, "Viewing security insights", checked 28 September 2026). GitHub publishes a definition for each metric, with exclusions: mean time to remediate leaves out alerts closed as false positives, and reopened alerts are aged from their original creation date. An impact analysis table lists the ten repositories with the most open alerts, and a View details link opens the secret scanning report "with the same filters and time period selected" (GitHub Docs, "Security overview dashboard metrics", checked 28 September 2026). GitHub also warns that "Dashboard metrics can change for the same historical time period when viewed at different times" (GitHub Docs, "Security overview", checked 13 October 2026).

What to take from it. The tabs follow the stages of the work rather than the data source. The published definitions settle arguments before they start: two teams reading "mean time to remediate" can check they are counting the same alerts. Carrying the filters across when the user drills in means the detail matches the number they clicked. The warning about changing history helps the number stay trusted when it moves.


Where the pattern breaks

The evidence also limits the hunch in four ways.

Not every dashboard is for acting. Sarikaya, Correll, Bartram, Tory and Fisher analysed a corpus of 83 dashboards and the literature on their use. They found the purpose of dashboards has expanded beyond operational decisions to strategic and tactical ones, and that a dashboard "may not even support decision-making at all", serving awareness or communication instead (IEEE TVCG, 2019; preprint). Google's SRE book makes a related split in operations: a dashboard is "a summary view of a service's core metrics", while the thing that interrupts someone should be actionable, and "every page should be actionable" (Site Reliability Engineering, chapter 6, Google, 2016). Our reading: a monitoring overview and an action queue are different jobs. Forcing a queue onto an awareness view, or a wall of charts onto an action view, fails in both directions.

A queue can become the new wall of charts. The closest research we found comes from a neighbouring field. In a retrospective study of 112 primary care clinicians' electronic health record data, Ancker and colleagues found that the likelihood of accepting a clinical reminder dropped by 30% for each additional reminder per encounter, and by 10% for each five percentage point increase in the share of repeated reminders (BMC Medical Informatics and Decision Making, 2017). That is clinical decision support, not SaaS dashboards, so treat it as a warning rather than a measurement of your product. It points the same way as the design choices above: Sentry's archive-until-escalating, Gainsight's auto-close advice and Search Console's reminder that not everything needs fixing are all mechanisms for keeping a queue short.

People need more than a view. In an interview study of data workers who use dashboards as their main interface to data, Tory, Bartram, Fiore-Gartland and Crisan found that their needs were far more diverse than consumption, and that their engagement with data "was often intermittent, dependent on experts, and involved an awkward assembly of tools" (IEEE Computer Graphics and Applications, January 2023). That is an argument for the route to the records, and for export where the product cannot answer the follow-up question. Linear and GitHub both offer CSV export, which serves that need.

Nobody has shown these designs improve decisions. None of the vendor pages we read publishes an outcome measure. The research gap is broader: Dimara and Stasko found that decision tasks rarely appear in visualisation task taxonomies and that "the majority of visualizations addressing decision making were not evaluated based on their ability to assist decision tasks" (IEEE TVCG, 2022; author version). Bach and colleagues, after reviewing 144 dashboards, noted "surprisingly little design guidance for dashboards" (IEEE TVCG, 2023; arXiv). The patterns in this article are reasoned from how the products are built, not proven by measurement. Test them against the decisions your users actually make.


Seven questions to ask of your own dashboard

Use these on the view your customers complain about. Each question comes from one or more of the examples above.

QuestionWhat it exposesWhere we saw it answered
Compared with what?Whether the view shows change against a visible baseline, or only current valuesSalesforce "Changes Since"; Pulse expected range; Amplitude forecast band
Is up good?Whether the product can tell good news from bad without the reader's memoryPulse "Value going up is"
What does this number open?Whether each metric leads to the records it counts, with the same filtersSalesforce metrics; Linear bars; GitHub View details
Whose is it to fix?Whether the view separates what the user can act on from what they cannotSearch Console source column; Gainsight CTA owner
How does an item leave the list?Whether exceptions can be resolved, parked or closed, and come back if they worsenSentry archive until escalating; Search Console Validate fix
What is the definition, and what is excluded?Whether two teams reading the same label count the same thingsGitHub metric definitions
What happens when the list gets long?Whether anyone owns the rules that create the queueGainsight Cockpit FAQ

If most answers are "we don't know", the problem is unlikely to be chart choice. It is more likely the decision, the definitions and the route to the records, which is where we would start.


Where this work sits

At Tcules, we design data surfaces around definition, provenance, comparison and consequence, so a person can notice what changed, interpret it and act. The questions above are the ones data visualisation and decision products work settles before choosing a chart, including keeping source, freshness, uncertainty and missing data visible.

When the underlying issue is how objects, records and views relate, that is information architecture. When the dashboard's tables and filters behave differently from the rest of the product, the cause may sit upstream in the design system, and how to audit a design system shows how to find where it stops being shared.

About the authors

  • Rishabh Choudhary

    Product Designer

    Rishabh Choudhary is a Product Designer at Tcules who writes about design systems and about designing for people the average interface leaves out. Rishabh's articles look at ten age-inclusive products, inclusive design for older and disabled people, eLearning platforms from the side of both learners and administrators, and software for lab and life-science teams. On design systems, the pieces cover components that several brands can reuse and B2B SaaS design systems worth studying. Rishabh has also co-written on fintech UX, healthcare UX and choosing a fintech design agency.

    More from Rishabh
  • Rushali Bhatt

    Product Designer

    Rushali Bhatt is a Product Designer at Tcules. Rushali's articles start from the people inside complex software: clinicians and patients in healthcare products, users who cannot take part in research sessions, and the reviewers who check an AI feature's output before anyone acts on it. Topics include usability studies for enterprise software, human-in-the-loop patterns, a readiness checklist for AI product UX, what to fix first when adding AI to a live SaaS product, and what a heuristic evaluation should deliver. Useful reading for product teams whose software people rely on to make decisions.

    More from Rushali

Turn your dashboard into something people act on

Share the decisions your users need it to support.