The Interest-Only Trap: A Ninety-Day Plan to Clear Data Debt Without Halting the Business
Photo: Foreign and Commonwealth Office, OGL v1.0, via Wikimedia Commons
Data debt rarely arrives as a decision. It accrues. A quarterly report that nobody opened after the reorganisation, a customer table carrying three versions of the same postcode field, a manual reconciliation that began as a two-week fix and is now a permanent fixture in someone's calendar. None of these items is dramatic in isolation. Together they set a recurring interest rate that the business pays every month in analyst hours, delayed decisions and audit friction.
The temptation, once a finance director puts a number on that cost, is to launch a remediation programme. More often than not it stalls in month four, having consumed a budget and produced a catalogue nobody reads. The alternative set out here is narrower and more disciplined: a ninety-day repayment plan, run alongside business as usual, aimed squarely at the debt that carries the highest commercial interest.
Defining the balance sheet
Data debt is the gap between the data estate an organisation depends on and the state that estate is actually in. For planning purposes it helps to separate it into four kinds:
- Structural debt — duplicate fields, missing keys, inconsistent reference data, tables copied rather than joined.
- Reporting debt — dashboards nobody reads, overlapping reports, unofficial versions circulated by email after the official one.
- Documentation debt — no lineage, no agreed definitions, no record of why a particular filter exists.
- Behavioural debt — manual workarounds, personal spreadsheets, adjustments applied directly in a live system.
Only the first is genuinely an engineering problem. The other three are governance problems, and in most mid-market firms they are where the interest is actually being paid.
Pricing the interest before touching the principal
Clearing data debt entirely is not a realistic objective for a firm without a dedicated data function, and it is not the point. The point is to lower the monthly cost of carrying it. Before prioritising a single fix, put a rough figure on that cost:
- List every recurring manual reconciliation and record the hours it consumes each month.
- List every report known to be duplicated or distrusted, noting who rebuilds it and how often.
- Multiply total hours by a loaded day rate that includes overheads and employer costs.
- Add the cost of one slow decision per quarter — a pricing change, an acquisition, a regulatory submission — delayed by a week for want of a trusted number.
The output is not a precise figure, but it is a defensible one, and it moves the conversation from a technical argument about tidiness into a commercial argument about capacity.
Month one: inventory and triage
Do not begin with tools. Begin with a register. Three lists will do: reports, reconciliations and recurring data requests. For each report, pull the last-accessed date from the business intelligence platform, name the current owner (or record that there is none) and note which decisions it informs.
Score every item on three dimensions, each from one to five:
- Commercial risk — what happens if this number is wrong or unavailable.
- Frequency of use — daily reliance scores higher than annual curiosity.
- Effort to remediate — hours and dependencies, not technical elegance.
| Tier | Typical contents | Treatment |
|---|---|---|
| Tier 1 | Statutory and regulatory reporting, pricing, payroll, customer commitments | Fix in the first thirty days; assign a permanent owner |
| Tier 2 | Commercial management information, sales and margin reporting | Consolidate and document within the quarter |
| Tier 3 | Historical dashboards, exploratory reports, one-off analyses | Retire or archive by default |
Tier 3 is where the fastest progress sits, and it is usually where resistance is weakest once people realise that archiving is not deletion.
Month two: retire, consolidate, document
Apply a simple rule: a report that has not been accessed in ninety days and has no named owner is archived, with a route to restore it on request. Expect this single measure to remove a meaningful share of the estate at most mid-market firms.
Consolidation follows the same logic. Where two reports answer the same question, keep one. Where new report requests arrive during the quarter, accept them only if they replace at least one existing output — a rule that costs nothing to adopt and changes behaviour quickly.
Documentation during this month should be deliberately minimal: a short register of core metrics, each with one definition, one owner and one source of record. Fifteen metrics is a sensible ceiling for a first pass. A definition that cannot be stated in two sentences is not yet a definition.
Month three: controls and cadence
Repayment without controls refills the account. Three light mechanisms prevent that:
- A change log. Any alteration to a core metric definition is recorded with a date, a name and a reason. It can be a shared document; it does not need to be a platform.
- A quarterly debt review. Ninety minutes, standing agenda, the same five indicators each time.
- A debt allowance. Ring-fence a modest share of analytics capacity — ten per cent is a workable starting point — for repayment work only, protected from ad hoc requests.
Five indicators of a healthier estate
Progress is easier to sustain when it is visible. Track the same five measures each quarter:
- The proportion of reports accessed in the last ninety days.
- The number of manual reconciliations still performed monthly.
- The number of core metrics with a single agreed definition and a named owner.
- The average time taken to answer the question: where does this number come from?
- The rate of duplicate records in customer and supplier master data.
None of these requires a new platform. All of them can be produced from existing systems and a modest amount of honesty.
Failure modes worth avoiding
Three patterns reliably derail remediation work. The first is the big-bang clean-up, which treats debt as a project with an end date rather than a standing cost. The second is buying a catalogue or lineage tool before definitions have been agreed, which automates the documentation of confusion. The third is treating the exercise as IT-owned; most of the debt lives in business processes, and the fixes have to live there too.
A ninety-day plan will not leave a mid-market firm with a pristine data estate. It should, however, leave it with fewer reports, clearer definitions, a named owner for each critical number and a way of measuring whether next quarter is better than this one. That is a realistic exchange for a quarter's work — and considerably cheaper than continuing to service the interest.