Slack, Microsoft and Google Make Dashboards Easy. Here's How to Trust Their Numbers

Slack users no longer need a data team to build a dashboard. They describe what they need in plain language, and Slackbot builds a dashboard, report, executive deck or interactive page, which Slack calls a Surface.

Slack added Surfaces in a July 29 update. In September, Salesforce renamed it Slackforce Surfaces and opened it to every plan, including free teams.

Slackbot builds each Surface from the conversation and connected tools, such as Salesforce. Colleagues can then filter, comment on and pin the result. Salesforce says no business intelligence (BI) team is needed. Work that once went through data analysts can now be done by anyone in minutes.

Microsoft and Google are moving in the same direction. Microsoft said in late September that Copilot users can describe a dashboard and have Copilot build it. Copilot in SharePoint can also turn a list into a dashboard that refreshes when opened.

Google's Gemini, meanwhile, builds spreadsheets in Sheets from a description. These auto-generated dashboards raises three problems.

1. Two people can build dashboards showing the same figure, calculated differently.

2. Each dashboard includes only what its builder sees, because a Surface is built under the user's own permissions. Two employees asking the same question could get answers drawn from different parts of the company's data.

3. Figures may be out of date. Salesforce says Surfaces update automatically, but its launch page says that feature only starts rolling out in October. Until then, a Surface may show stale numbers.

So what happens when every team builds its own dashboard from different data?

The Label's the Same, The Numbers Are Different

Data kept in separate silos leads to different definitions for the same metric, said Sameer Kadoo, director of tech solutions at the Royal Bank of Canada Capital Markets. Without a single source of truth, companies end up with duplicated calculations and conflicting figures, he said. The trouble is that the labels match, so differences are easy to miss.

Headcount is a good example. In one dashboard, it might mean employees active on a particular day, and in another an average across the month, said Nadia Bifolchi, director of analytics at HR research and advisory firm McLean & Company.

Turnover can be just as slippery, counting every departure in one view and only voluntary ones in the next. "The charts may both look polished, but they're not describing the same thing," Bifolchi said.

Across a company, this becomes what Jason Beres, COO of software company Infragistics, calls metric sprawl: "five versions of 'revenue,' three definitions of 'active user,' and no way to tell which is accurate."

What Use Is a Dashboard With Partial and Outdated Data?

Even when definitions match, the data underneath may be incomplete. Each dashboard reflects only the region, business unit or system its builder reaches. A view like that "may be accurate within that boundary and still be incomplete for an organization-wide decision," Bifolchi said.

When data sources are partial, a dashboard "will fail to show the true or complete picture of an organization,” Kadoo agreed.

For example, take a sales manager who can see the pipeline but not the finance team's adjustments for refunds and lost customers, Beres said. Nothing on the screen warns the manager that anything is missing.

Out-of-date data adds another blind spot. A sudden dip in a metric might look like a problem when the real cause is a data source that stopped refreshing.

When Decisions Go Wrong

These problems are hard to spot because polished charts look certain. The visual presentation "can create more confidence than the underlying data deserves," Bifolchi warned. People also trust a chart more than a raw table, so a flawed view gets acted on faster and with more conviction, Beres said, which he calls "optimizing what you can see."

For example, a team that sees sales rising, but not the returns or cancellations that follow, may pour more money into a channel that is actually losing it.

The reverse can happen, too. "An enterprise risk may be critical, but if the data feeding into its assessment is missing, it may show as low impact, leading the organization to deprioritize it in its risk mitigation plan," Kadoo said.

Dashboards that disagree cause a different kind of damage. One team is convinced retention is the urgent problem, another is sure it is acquisition and each has a dashboard to prove it. To Beres, that is "misalignment that looks like a people problem but is really a data problem."

Meetings then turn into reconciliation sessions, with teams spending "more time reconciling reports than making decisions,” Bifolchi added.

Eventually, trust in the numbers erodes. Once that happens, people "revert to gut instinct and the loudest voice in the room, which is the exact outcome dashboards were supposed to prevent,” Beres said.

Standardizing the Official Measures 

The fix starts at the source, by deciding what the key numbers mean before anyone builds a dashboard around them.

That means defining each important metric once, in one place, and making it the default. Each official measure needs a shared definition, a calculation method, a named owner, a stated source system, a refresh schedule and an intended use, Bifolchi said.

Dashboards built on those definitions can then be certified and labeled, so the official number is easy to find. "Self-service should build on top of trusted definitions, not around them," Beres said.

Underneath those definitions, companies need consistent data sources and strong governance, Kadoo said. He also stressed identity and access management, the rules that control what each employee's role allows them to do with company data.

Here the concern is who can change the data, not just who can see it. The goal is to stop mistaken or isolated updates from reaching sources everyone relies on.

What IT Can Do

For IT and data teams, the first job is to take stock. The starting point is an inventory of existing dashboards and the data sources behind them, to find where figures conflict, Kadoo said.

From there, companies agree on standard metrics, assign data owners, review who has access to what and train employees on the approved sources. Kadoo also suggested a formal process for employees to request changes to data sources, plus regular data reviews.

None of that means shutting self-service dashboard creation down. Blocking it only pushes people toward workarounds, according to Beres, who said IT should "shift from gatekeeping to guardrails."

In practice, that means making the right option the easy one, Beres said. Employees should have a well-documented set of approved data sources, with access rules attached to the data itself so they apply regardless of who builds the dashboard. Every dashboard should also show where its data came from, when it was last refreshed and who owns it.

Freshness deserves particular attention. "A dashboard that doesn't update automatically is a snapshot pretending to be a live view," Beres said. Tracking which dashboards people use also helps, so popular ones can be promoted and stale ones retired.

Exploratory dashboards should be kept separate from those approved for decisions, Beres and Bifolchi said. Experimenting is fine, but in Bifolchi's view, a dashboard that becomes widely shared, regularly used or tied to a high-impact decision such as executive reporting needs extra review. She also recommended a way to retire duplicates.

How Employees Can Check a Colleague's Dashboard

Employees can protect themselves, too. Before relying on a dashboard they didn't build, Bifolchi suggested checking five things: "the source, the scope, the metric definition, the refresh date and the owner." If any are missing, it should be treated as a starting point rather than a final answer, she said.

Compare one figure on the dashboard with a number you already trust, and if they don't match, find out why before acting on anything else it shows, Beres recommended.

Tools such as Surfaces make building a dashboard almost effortless. The harder work is agreeing what the numbers mean, where they come from and how old they are. "A clean visual alone does not mean the underlying data is complete,” Bifolchi said.