Why Customer Data Is the Most Underutilised Asset in Business

There is a checklist that organisations use, mostly unconsciously, to reassure themselves about data maturity. None of the items on it are evidence of maturity.

• Owning a modern cloud data platform. This is infrastructure, not capability.
• Producing executive dashboards. Reporting is a description of the past.
• Employing data scientists. Talent without decision ownership produces analysis, not outcomes.
• Collecting transaction data. Almost every business in payments, retail and banking has this. It is table stakes.
• Reporting more KPIs. Metric proliferation usually indicates the absence of agreement about what matters.
• Deploying predictive models. A model in production that nobody acts on is a well-engineered opinion.

IDC research commissioned by Seagate estimated that around 68% of the data available to enterprises went unleveraged. It is important to be transparent that this was vendor-commissioned research and that measuring unused data is inherently imprecise. Even allowing generously for that, the scale of the gap has been consistent across studies for a decade, and it matches what I see in operating businesses.

It helps to lay out the full chain, because most organisations can point to exactly where they stop.

Most organisations stop at stage three and call it success. Insight is genuinely satisfying to produce, it presents well, and it requires no organisational change to deliver. Stages four to six require someone to alter their behaviour, which is why they are skipped.

An insight that nobody is accountable for acting on is a cost centre with good presentation skills.

First, fragmented data. The signals required for one decision typically sit in several systems, with inconsistent definitions and no shared customer identifier. The technical work to resolve this is well understood. The reason it does not happen is that no single business owner is accountable for the decision that would justify it.

Second, poor usability. Analytical outputs frequently arrive in tools that the decision maker does not use during the work in which the decision is made. A recommendation that requires a banker or a store manager to open a separate portal will not be adopted, and the failure will be attributed to culture rather than to design.

Third, questions defined too broadly. 'Understand our customers better' is not a brief. 'Predict which merchant relationships will lapse in the next ninety days, in time for the relationship manager's monthly planning cycle' is a brief, and it implies its own success measure.

Fourth, no decision owner. If the analysis has no counterpart who is accountable for the outcome, it becomes reading material. This is the single most reliable predictor of whether an analytics investment will produce value.

Fifth, no feedback mechanism. Without measurement of what happened after the action, the system cannot improve, and the organisation is condemned to relitigate the same judgement every year.

The most useful exercise I run with commercial teams takes about two hours. List the recurring decisions the business makes. Not the strategic once-a-year choices, the operational decisions taken weekly or monthly: which accounts to prioritise, which customers to contact, what to price, what to stock, what to approve, where to deploy capacity.

For each, estimate three things: how often it is made, how much value rides on it annually, and how much better it could plausibly be with better information. The output is almost always surprising. The decisions with the largest aggregate value are rarely the ones receiving the most analytical attention, because they tend to be unglamorous, distributed and made by people well below the executive team.

Then build backwards from those decisions. What information would change the choice? How quickly must it arrive? Who acts on it? This inverts the normal sequence, in which a platform is built and use cases are sought afterwards.

Before approving any data asset, dashboard or model, I ask the sponsoring team to answer six questions in writing. If any answer is missing, the initiative is not ready, regardless of how strong the technical case is.

The pattern across all four is the same. The valuable decisions are frequent, distributed and operational. The analytical challenge is rarely sophistication. It is delivery into the workflow at the moment of the decision.

Privacy and consent set the outer boundary of what should be done, which is narrower than what can be done. Regulation is the floor rather than the standard, and customer expectation moves faster than legislation. A use of data that would embarrass the company if described plainly to the customer should not proceed.

Poor data quality is a genuine constraint and not merely an excuse. Acting decisively on unreliable signals is worse than acting slowly on reliable ones, particularly in credit, pricing and risk. The correct response is to scope the first decisions around the data that is trustworthy, and let demonstrated value fund the remediation of the rest.

Historical data encodes historical behaviour, including the effects of past commercial choices. A model trained on who was previously contacted will faithfully reproduce who was previously contacted. Treating those patterns as objective truth about customer preference is one of the most common analytical errors in commercial organisations.

There is also a reasonable objection that decision-led data strategy underinvests in foundations. Some platform work has no immediate decision attached and is still necessary. The answer is proportion rather than purity: fund foundations explicitly as foundations, with a defined set of decisions they are intended to unlock, rather than disguising them as value initiatives.