Why the first question is not which tool to buy, but whose number the business should believe, and a five-step plan to get there.
In brief
- Most “data platform” requests are really a question: why do two reports about the same thing show two different numbers?
- A single source of truth is not one giant system. It is one agreed place where reconciled data can be seen.
- The plan has five steps: decision, map, layers, definitions, proof.
- Tools move data around. Only agreed definitions and owners make people trust it.
Most organisations that ask for a data platform are really asking a simpler question: why do two reports about the same thing show two different numbers?
The usual response is to compare tools: a warehouse, a lake, a lakehouse, a dashboard product. That is the wrong place to begin. Tools only move data around. They do not decide which version of a number is correct, who owns it, or what it means. An organisation that picks a platform before answering those questions tends to end up with the same disagreements, now stored in a more expensive place.
The real issue is not missing data. It is disconnected data.
In most mid-sized and large organisations, each system is correct about its own job. The sales system knows what was sold. The operations system knows what was shipped. The finance system knows what was invoiced. Each one is accurate when you look at it alone.
The trouble starts when someone needs an answer that crosses systems: How many units did we sell, deliver and bill this month, and where do those numbers disagree? Nobody owns that question, so people answer it by hand. They export three spreadsheets, line up the columns, argue about which one is right, and send the result by email.
That manual work is the hidden cost. It rarely appears in a budget line, but it shows up as:
- Slow answers. Management waits days for a number that should take minutes.
- Competing versions. Two teams quote different figures for the same measure, and meetings turn into reconciliation sessions.
- Fragile knowledge. The only person who knows how to produce the monthly report is one person, and that person takes leave.
- Stalled initiatives. Anything that depends on trusted data, including analytics, forecasting and AI, quietly waits behind this problem.
An illustrative example (not a client figure): if five people each spend three hours a week reconciling exports, that is about 780 hours a year, roughly 100 working days, spent producing numbers rather than using them.
A five-step plan for one trusted platform
1. Start from the decision that hurts, not from the data
List the five questions management asks most often and answers least reliably. Pick the one where the cost of a slow or wrong answer is highest. That single question defines your scope. If a data source does not help answer it, it can wait.
2. Map where the truth lives today
For each thing the business cares about (a customer, an order, a vehicle, a contract, a payment), write down which system is the system of record, and which other systems copy or change it. You will almost certainly find the same entity under different identifiers, status changes made by hand in spreadsheets, and two systems that each believe they are the authority.
3. Land it in one place, in layers
| Layer | What it holds | Who uses it |
|---|---|---|
| Raw | An untouched copy of what each source system sent | Data engineers, for tracing and re-processing |
| Cleaned | The same data with identifiers matched, duplicates handled and formats standardised | Analysts and integration teams |
| Business-ready | Reconciled, named, documented tables built around business questions | Management, finance, operations, dashboards |
The layering matters more than the brand of technology. Keeping the raw copy means you can always explain a number by tracing it back. Keeping the business-ready layer small and named in business language means people can actually use it.
4. Agree the definitions, in writing
Take each key measure (sold, delivered, active customer, margin) and agree what it means, how it is calculated, and who owns it. A one-page definition per measure, with a named owner who can settle disputes, is enough to begin. Treat privacy and access control as part of the design rather than an afterthought.
5. Prove it with one report, then widen
Deliver the answer to the question from step 1 to the people who asked it. Let them compare it with the old manual number. When the two match, or when the differences are explained, trust follows. Only then add the second question, the second source, the second team.
Four traps we see repeatedly
- Choosing the technology first. The platform should follow the questions, not lead them.
- Loading everything. More data does not create more trust. Reconciled data does.
- Skipping ownership. If nobody owns a definition, the argument returns the first time two numbers differ.
- Treating it as an IT project. If only the technology team is in the room, the definitions will be wrong.
Our experience
In our work with enterprise clients, we have built cloud data warehouses that bring records from several operational platforms into one place, so that cross-system reporting comes from a single source instead of separate extracts.
The lesson matches the plan above: the value came less from the technology than from agreeing first what the shared data should look like.
We can explore the same for your data, starting with one report. Where do two of your reports disagree today?
One place to look
An organisation does not need more reports. It needs one place to look, and the confidence that what it sees is the same thing everyone else sees. If your teams spend more time reconciling numbers than acting on them, the first step is usually not a new tool but a clearer map of where your truth lives today.
Talk to Balerion about mapping your data and planning the first release of a platform your teams will trust. Email us at talk@balerion.co.th.
