The Cost of Delay on a Data Platform, Priced Out

Every data platform business case is missing a number

The case gets built as cost versus benefit. Platform licenses, engineering time, integration work on one side. Efficiency gains and better decisions on the other.

Missing from both sides is the cost of the months themselves. Not the cost of building for nine months. The cost of not knowing for nine months, which continues whether or not you are building.

That number is usually larger than the project budget, and it is the strongest argument for a faster sequence. It is also the easiest number in the business case to defend, because it is built from decisions your operators already make with bad information.

How to price a month of not knowing

Do not model this top-down from a revenue percentage. Nobody believes those numbers and they should not.

Build it from four decision loops that mid-market retailers run continuously and currently run blind or late. For each one, estimate the decision frequency, the share of decisions made on stale or wrong information, and the average cost of getting it wrong.

Markdown timing. Sell-through visibility that arrives in week 8 of a 12-week season instead of week 3 means markdowns taken later and deeper. The gap between an early 20% markdown and a late 40% one, on the affected units, is the cost.

Replenishment on bad on-hand. Inventory records that are wrong enough to break auto-replenishment produce both stockouts and overstock on the same SKU list. The cost is lost margin on the stockouts plus carrying cost on the overstock.

Shrink detection lag. A shrink pattern found at the quarterly count instead of within days is a pattern that ran for an average of six weeks.

Labor against demand. Schedules built on last year's shape rather than current demand, corrected monthly instead of weekly.

A worked example, 120 stores

Take a 120-store specialty chain at $180M revenue, roughly $1.5M per store. Conservative assumptions, and every one of them is worth arguing with using your own numbers.

Markdown timing. 18% of revenue eventually goes on markdown. Acting three weeks earlier on the worst-performing third improves realized margin on those units by 4 to 6 points. On roughly $11M of affected retail, call it $450,000 a year, or $37,500 a month.

Replenishment. Stockouts on A-items running 3 to 5% of store-weeks. Halving that on the detectable half recovers roughly $600,000 of annual sales at a 38% margin, so $228,000, or $19,000 a month.

Shrink. Shrink at 1.4% of revenue, $2.5M. Cutting detection lag from a quarter to a week addresses the portion that is pattern-based rather than structural. At 8% of total shrink, $200,000 a year, or $16,600 a month.

Labor. Payroll at 11% of revenue, roughly $20M. A 1% efficiency improvement from weekly rather than monthly correction is $200,000 a year, or $16,600 a month.

Total: roughly $90,000 a month. Nine months of delay is $810,000 of decisions made without the information, against a platform project that probably budgets $400,000 to $700,000 all-in.

The delay costs more than the build. That sentence is the business case, and it survives being argued down by half.

Delay cost does not stop at go-live

The number above assumes value switches on the day the platform ships. It does not. Adoption ramps, and the ramp is usually three to six months, because trust is earned per metric and per person.

That has a sequencing consequence people miss. Starting the ramp in week three at 20% of eventual value beats starting it in month nine at 100%, on cumulative value captured, by a wide margin. Early partial answers are not just faster. They compound.

It also means a fast first answer buys something beyond the answer itself: it starts the trust clock, which is the slowest clock on the project and the one no engineering decision can accelerate.

What not to claim

Cost of delay arguments get discredited by overreach, so bound them honestly.

Do not claim the platform captures all of it. A realistic capture rate on decision-loop value is 30 to 60% in year one, because some decisions will still be made on instinct and some data will still be wrong.

Do not double-count. If the markdown improvement and the replenishment improvement both credit the same units, halve them.

Do not price loops nobody will act on. If store operations has no mechanism to respond to a daily on-shelf signal, the value of that signal is zero until the mechanism exists. Delivery and process are part of the capture rate, not an afterthought.

A defensible number that survives a CFO's questions is worth more than a large number that does not.

How to use it in the room

Do not open with it. Open with the timeline options, and use the delay number to price the difference between them.

Two plans on one page. Plan A: nine months, full scope, everything modeled, one go-live. Plan B: first reconciled answer in week three, one decision loop live by week six, remaining scope through month six.

The engineering cost of those two plans is similar. The difference is roughly six months of delay cost on the loops that Plan B lights up early, which in the example above is around $400,000.

That reframes the meeting. The question stops being whether to fund a data platform and becomes which sequence to fund, and the second question has a much cleaner answer.