SURFAICE

Research Report · August 2026

Five Buckets, One Budget

What agentic AI changes about retail capital allocation — for VPs of Real Estate and Chief Development Officers who have to defend a capital plan in a room where every other function has a better-formed business case than facilities.

By Joe Valeri, MBA, MS — Founder and former CEO of Lucernex, co-founder of Surfaice, chairman and co-founder of Rendr. Twenty-nine pages, August 2026.

• 0.99 Mean correlation between one year's capital allocation and the last, across a third of business units
• $1.1bn Ross Stores' annual capital budget, publicly described in thirds and quarters with "maybe" twice
• ~3:1 Dollar General's Q1 2026 spend on the existing fleet against new stores — from a retailer opening 450 stores this year

Executive summary

Capital allocation in retail real estate is not one decision. It is a forced ranking across five competing uses of the same dollar: new stores, remodels, resets, closures, and capital projects. Each bucket is owned by a different function, measured on a different metric, and defended in a different meeting. None of them are independent of the others — and almost no retailer models them together.

The evidence that this fails is not anecdotal. McKinsey's study of capital allocation across roughly 1,600 companies over fifteen years found that for a third of business units, "the amount of capital received in a given year was almost exactly that received the year before — the mean correlation was 0.99," with an economy-wide mean correlation of 0.92. Last year's budget predicts this year's almost perfectly.

That is not a plan. It is inertia with a cover page.

And the companies that broke out of it — the top third of reallocators, shifting an average of 56% of capital across assets — earned 30% higher total returns to shareholders annually than the bottom third.

Retail's own disclosures show how coarse the allocation actually is. Ross Stores' CEO described the split of a $1.1 billion annual capital budget this way: "maybe think about a third each for DCs and new stores, and maybe 25% for store maintenance, and the balance for technology." That is one of the most specific public statements of capex mix in the sector — and it is delivered in thirds and quarters, with "maybe" twice.

Meanwhile the numbers that are disclosed point somewhere interesting. In Q1 fiscal 2026, Dollar General put $203 million into "improvements, upgrades, remodels and relocations of existing stores" against $73 million "related to store facilities, primarily for leasehold improvements, fixtures and equipment in new stores." Nearly three dollars into the existing fleet for every one into new units — and that from the retailer opening 450 stores this year. The center of gravity in retail capital has moved to the assets you already have, and the analytical tooling has not moved with it.

This paper examines what agentic AI changes about that forced ranking: not speed for its own sake, but the ability to consider equally all five buckets in one model with common units, so a roof replacement can be compared against a store opening on the same basis. It also sets out, in specific terms, what a system built to do this actually has to compute — the exit-cost formula, the constrained optimization, and the confidence gate that decides whether the answer is trustworthy enough to act on.

For CFOs and VPs of Real Estate managing $10M+ annual capital allocations, the difference between quarterly-cycle and continuous portfolio intelligence is becoming a competitive variable rather than a technology preference.

What is in the twenty-nine pages

1. Section I — The five buckets and the one budget
2. Section II — Why the five buckets are not independent
3. Section III — What agentic AI actually changes
4. Section IV — Before any number: the data quality gate
5. Section V — The worked example
6. Section VI — What you need to make this work
7. Section VII — Why this matters now
8. How retailers can deliver portfolio scenario modeling now
9. Conclusion — The one thing to change
10. Sources — every external figure cited to a named, dated, publicly reachable document

A note on the numbers, in the author's words: every external figure carries a citation to a named, dated, publicly reachable source. Figures derived from disclosed data are labeled derived. Nothing here is presented as a benchmark that isn't one.