Digital transformation starts with data, not software

Plenty of digital transformation programmes begin with a list of software and end with a list of software — just more expensive.

The reason isn't picking the wrong tool. It's that tools arrive before the organisation has settled two things: what counts as the truth, and who is responsible for keeping it right.

Symptoms that are easy to recognise

You're in this situation if any of this sounds familiar:

• Three departments report three different revenue figures for the same month, and the meeting is spent arguing which is correct rather than deciding what to do.
• Month-end close is late because accounting waits on a spreadsheet from the warehouse, which someone types up by hand from another system.
• One person is "the only one who knows how to pull that report" — and the week they're on leave is the week nobody can decide anything.
• Adding software increases the number of times data gets keyed in, because the new system can't talk to the old one.

None of these are software problems. All of them are data with no owner and no definition.

Four steps, strictly in order

Step 1 — Declare the source of truth for each data type

For every important data type, be able to answer: which system is the system of record?

Customers in CRM or in accounting? Stock in the warehouse system or in the books? The employee list in HR or in the time clock?

The answer doesn't need to be architecturally perfect. It needs to be explicit and widely known. A mediocre system everyone agrees is authoritative beats two good systems where nobody decided which wins.

Then enforce one rule: other systems read from it; they don't re-enter it. Every re-entry is an opportunity for two records to diverge.

Step 2 — Standardise the master data

Master data is the reference set every transaction points at: customers, suppliers, items, GL accounts, units of measure, departments.

It's small, it changes rarely, and it determines the quality of everything downstream. An item catalogue with 200 duplicate rows corrupts every sales report, every inventory analysis, every costing calculation — permanently, until it's cleaned.

Cleaning master data is tedious and nobody volunteers for it. It also has the highest return on effort of anything in the programme.

Step 3 — Capture transactions at the moment they happen

This is what separates "we have software" from "we transformed".

Entering data after the fact — end of day, end of week, end of month — produces a system used only to reconcile the past. It helps nobody decide anything, because by the time data lands the decision is made.

Capturing at the point of occurrence — the order when the customer places it, the receipt when goods arrive, the hours as they're worked — produces a system that reflects the present. That's the kind an executive actually opens.

One operational detail decides this step: the person entering data must be closest to the work, and the action must be faster than what they do today. If entering into the system is slower than a notebook, they'll use the notebook — and you'll be back at step 3 in six months.

Step 4 — Only then, automate

Automation is safe only after the three steps above. Automating over bad data saves nothing — it multiplies the error faster.

This is also the point at which AI becomes usable. Not earlier.

Why the order can't be shuffled

People want to jump straight to step 4, because it's the only step visible from outside.

But each step is a precondition for the next:

• Without a source of truth (1), where exactly are you standardising master data (2)?
• With duplicate master data (2), what do the transactions (3) point at?
• With late transaction capture (3), automation (4) runs on last week's picture.

Skipping a step doesn't make you faster. It defers the cost, with interest.

How to measure it

Transformation is hard to measure because the benefit is diffuse. These four are concrete and hard to fake:

Metric	Before	After
Days to close the month	10–15	3–5
Times the same data is keyed in	3–4	1
Time to produce a report leadership asks for	days	immediate
Versions of "the truth" in one meeting	2–3	1

The last one matters most, and it's the only one everyone in the room feels immediately.

In closing

Software is a tool. Data is an asset.

Buying a tool takes weeks. Building the asset takes quarters — but it's what remains when you change tools, and the only thing that stops the next change from starting over.