How to Know If Your Data Is Reliable (Without Being Technical)

Five signs any manager can spot without opening a database. If two or more sound familiar, your decisions are at risk.

At some point in your company, this happened: someone presents a number, someone else says “that doesn’t look right,” and suddenly the meeting stops being about the decision and becomes a debate about the data.

That’s not a technical problem. It’s a business problem. And it has clear signs that anyone can spot — no SQL knowledge required.

Why this matters more than it seems

When data isn’t reliable, decisions still get made — but on gut feeling dressed up as analysis. The report is there, the number is there, but deep down nobody fully trusts it.

The cost isn’t just the time wasted arguing about data instead of deciding. It’s that the important decisions — which product to push, where to cut costs, which client to prioritize — get made on information that might be wrong. And that error isn’t always visible. Sometimes it surfaces months later, when the damage is already done.

In companies operating in fast-moving markets, the cost of making a wrong pricing or inventory decision because of bad data can easily exceed the cost of building the infrastructure that would have prevented it.

The five signs

Sign 1: Two teams have the same metric and get different numbers

Finance says the month’s margin was X. Sales says it was Y. Both exported from the system, both ran their calculations, and arrived at different results.

This happens when there’s no single source of truth. Each team built its own version of the data — with its own filters, formulas, and criteria for what to include. The result is two “official versions” of the same number, and neither is definitively correct.

The cultural impact: when numbers aren’t reliable, executives stop trusting reports and start developing their own intuitions. The data team stops being an enabler of decisions and becomes a source of debate.

Sign 2: Nobody knows exactly where the report everyone uses comes from

The Monday sales report. The monthly close. The dashboard the board reviews every week. Does anyone know exactly what data it uses, which system it pulls from, what calculations it applies?

If the answer is “someone set it up two years ago” or “it’s a spreadsheet that downloads from the ERP and then has some formulas applied,” that report is a black box. When something changes in the source system, the report can become outdated without anyone noticing.

The warning sign: when you discover that the “sales” report doesn’t include returns, or that the product “margin” doesn’t account for logistics costs — because those rules are buried in Excel formulas nobody documented.

Sign 3: Before an important decision, someone asks to “verify the number”

There’s a board meeting. A decision needs to be made — open a new location, adjust pricing, renew a contract. And someone says: “before we present this, can we verify the number is right?”

That manual verification — calling someone, cross-referencing another spreadsheet, “double-checking” — is the clearest sign that the system doesn’t generate trust on its own. Reliable data doesn’t need verification before every important use. It arrives with the validation process already built in.

In practical terms: if your team spends time “verifying” data instead of analyzing it, reliability lives in people rather than in the system. And that’s fragile.

Sign 4: Closings change after they’re closed

The March close came in at $X. Two weeks later, someone finds an adjustment, a missing invoice, a return that wasn’t captured. The close becomes $Y.

A closing that changes is a closing that didn’t have complete data when it was closed. This isn’t necessarily anyone’s fault — data from some systems might arrive late. But if it happens regularly, there’s a structural problem in how information gets consolidated.

The real impact: the decisions made based on the March close were made on a wrong number. If marketing budgets were adjusted, if a hire was approved, if a supplier agreement was signed — all of that happened on incorrect information.

Sign 5: There’s one person who “knows how the spreadsheet works”

Almost every mid-sized company has one. The person who built the model, who knows which tab touches which formula, who has to be present when the monthly report gets updated because otherwise “it breaks.”

When knowledge about how data works lives in a person rather than in the system, the reliability of the data depends on that person being available, not leaving the company, not changing anything without warning.

This is the most dangerous sign in the long run. The risk isn’t obvious until that person goes on vacation or resigns. At that point, the entire process grinds to a halt.

What each sign actually implies

None of these signs reflect a moral failing or indicate that anyone did their job poorly. They’re natural consequences of companies that grew without their data infrastructure growing with them.

What they do imply is that the decisions made with that data carry an unquantified margin of error. You don’t know how much might be wrong — you just know it could be. And that uncertainty, even if nobody verbalizes it, is present in every management meeting.

What reliable data infrastructure actually looks like

A system where data is reliable has three observable characteristics — no technical knowledge required:

Traceability: any number in the dashboard can be traced back to its source. If the report says the month’s margin was $X, there’s a documented chain of steps that explains how that number was calculated.

Uniqueness: there’s one version of each metric, with one definition. “Revenue” means the same thing in the finance report as in the sales report.

Automation: data updates on its own, on a predictable schedule, without someone having to manually run a process. If the pipeline fails, the system alerts — it doesn’t silently show stale numbers.

Building this doesn’t require a large team or an enterprise budget. It requires doing things in the right order, with the right tools. You can read how it works technically in Medallion Architecture Explained.

Where to start

The first step isn’t implementing anything. It’s understanding exactly what data you have, where it comes from, what state it’s in, and where the failure points are.

With that clarity, you can build infrastructure where data arrives validated, where there’s one version of every number, and where every change is tracked. The process is called a Data Audit — it takes about two weeks, and the output is an honest picture of where you actually stand.

Frequently asked questions

How long does it take to build reliable data infrastructure?

For a mid-size company with 3–7 data sources, 4–10 weeks to have the first reliable reports. The first week is typically the audit, the following weeks are building the ingestion and transformation pipeline, and the final weeks are modeling the reports the business needs. Results don’t all arrive at once: the first working dashboards usually appear around week 5–6.

What if my data lives in Google Sheets and Excel, not just formal systems?

Google Sheets and Excel are valid sources. They can be connected automatically to the pipeline and treated like any other source: data arrives in Bronze, Silver cleans and normalizes it, Gold combines it with everything else. The key is that at some point in the chain, the manual “update the spreadsheet” step becomes a structured input rather than a consolidation task someone has to do by hand.

Do I need to replace my current systems (ERP, CRM)?

No. Reliable data infrastructure is built on top of the systems you already have, without modifying them. The ERP, the CRM, and the spreadsheets remain the sources of operational truth. What changes is that there’s a centralized layer connecting them, cleaning them, and making them available for analysis — without touching how those systems operate.

---

If two or more of these signs sound familiar, schedule a call. In 30 minutes we’ll tell you how deep the problem goes and what the first concrete step would be.