Most retailers are about to make a seven-year AI decision and call it a hardware project.
The point-of-sale refresh is on the calendar. It gets scoped the way it has always been scoped: uptime, transaction fees, EMV compliance, footprint, maybe a nod to mobile checkout. Procurement runs the RFP. Finance approves it as capex. IT deploys it. Done for another cycle.
That framing was fine in 2015. In 2026 it is a quiet act of self-sabotage.
The terminal is the data layer now
Here is the thing nobody puts in the RFP. The POS is not where you take payment. It is where you capture the richest, highest-frequency, first-party behavioral data your enterprise owns. Every basket composition, every abandoned cart at the till, every loyalty swipe, every manual price override, every return, every substitution. That event stream is the ground truth your agents will reason over.
And agents are coming to the store floor whether your data is ready or not. Retailers are using agentic AI to drive inventory optimization, shelf management, checkout automation and predictive customer experiences. The use cases are not speculative. They are being piloted now.
The problem is the foundation underneath them. An AI agent does not work like a dashboard. In retail, that environment includes your POS data, inventory feeds, customer profiles, competitor pricing, order management systems, supplier communications, and support queues. An AI agent does not just read from these systems. It reasons across them simultaneously and acts, updating a stock allocation, triggering a replenishment order, responding to a customer inquiry, or flagging a pricing anomaly, without a human initiating each step.
That reasoning is only as good as the data it is grounded in. And the POS is the single biggest, freshest input to that grounding. If your terminal emits thin, batched, end-of-day summary data, your agents inherit thin, batched, stale intelligence. No model budget fixes a weak capture layer.
The gap is bigger than your board thinks
The honest read on the market is not that retailers are a little behind. It is that the store floor is the weakest link. IDC’s Ananda Chakravarty said it without hedging: “Unfortunately, retailers are woefully unprepared for the shift to autonomous agentic AI, especially at the retail store level.”
Why? Because the data was never built for this. There is a lack of clarity and lack of mature standards concerning how agentic AI will interface with such systems, as many retail data sets are not yet modernized. The POS you buy this year is either the thing that closes that gap or the thing that cements it for seven more years.
This is where the Cost of Doing Nothing (CODN) stops being a slide and starts being a balance-sheet reality. CODN on a POS decision is not the price difference between a basic terminal and a modern platform. It is every quarter of event data you capture in a format you cannot replay, cannot stream, and cannot hand to an agent. You do not get that data back. You are not deferring spend. You are financing a forced migration later at a far worse exchange rate, while your competitors compound a usable data asset the whole time.
And the smart money already sees it. While retailers typically try to get the longest lifespan possible out of POS hardware to maximize ROI, 67% of high-performing businesses say they have prioritized upgrades in 2026, a sure sign of how AI requirements are shaping investment. High performers are not refreshing because the old units broke. They are refreshing because the old units cannot feed what comes next.
Reframe the buy before you sign
The fix is not more budget. It is a different scorecard.
Stop grading POS vendors on transaction reliability and fees alone. Those are table stakes. Grade them on four things your current RFP almost certainly ignores. Event granularity: does it emit structured, item-level, real-time events or batched summaries? Streaming: can it push to your data platform live, not overnight? Edge compute: can it run inference locally where latency and data residency matter, which is exactly why retailers keep applications like point-of-sale transactions and loss prevention on edge while retaining unified control across cloud and edge environments. And an open, readable, writable API surface an agent can actually act against.
If a terminal cannot clear those four bars, it is not a cost saving. It is a ceiling.
The retailers who win the agentic era will not be the ones with the best models. Everyone rents the same models. They will be the ones whose stores emit clean, rich, real-time data because somebody in procurement understood that the POS line item was the AI line item all along.
Your next POS cycle is your next AI cycle. Scope it like one, or spend the next seven years paying for the one you did not.