The Hidden Cost of Retail Data Silos: When Your POS and Ecommerce Platform Don't Share a Cart

A customer adds a pair of trainers to their cart on your mobile app at lunch. They walk into your flagship store after work, intending to try the size before purchasing. The associate, holding a tablet running your POS, can't see the cart. They can't see the customer's loyalty status without a manual lookup. They can't redeem the promotional code the customer applied online. The customer, mildly irritated, says they'll "think about it" and leaves. The trainers go back on the shelf. The cart, abandoned online but never closed in-store, sits in a dashboard somewhere as a "high-intent abandonment."

This is the data silo problem in retail, told the way it actually happens — at the till, not in a board paper.

The standard prescription for this kind of fragmentation is well-rehearsed. Build a customer data platform. Pipe everything into a warehouse. Layer analytics on top. Train a model. Surface insights to the business. Most retail technology content treats the data silo problem as an analytics problem, and the solution as an analytics-layer solution.

We think that diagnosis is wrong. Or more precisely: we think it confuses a symptom with a cause. The data silo problem is not, fundamentally, that retailers can't see their data. The problem is that retailers can't transact across their channels. And until you fix transactions, no amount of analytics infrastructure will produce the unified customer experience that the warehouse was meant to enable.

‍

The conventional fix has a credibility problem

The case for an analytics-layer solution is intuitive. If your point-of-sale system, your ecommerce platform, your loyalty programme, your inventory database and your customer service tooling all hold separate fragments of the customer record, then the obvious fix is to consolidate those fragments somewhere. A customer data platform or warehouse becomes the unified ledger. Reports get built on top. Marketing optimises. Merchandising prioritises. Operations recognises new patterns.

The problem is that this is exactly what most enterprise retailers have been doing for the better part of a decade, and the operational reality has not meaningfully improved. Incisiv's State of Transformation in Retail and CPG report found that only 13% of retailers believe their technology will meet future customer expectations, and 89% fail to scale innovations across the organisation. These are not the numbers of an industry that has solved its data problem by buying more data infrastructure.

There is a reason for the gap between what the warehouse promises and what stores deliver. Analytics infrastructure unifies records. It does not unify transactions. A warehouse can tell you, three days after the fact, that a customer added a product to an online cart and then walked into a store and did not buy it. It cannot enable that customer to walk into the store and complete the purchase using their existing cart. The cart is the operational artefact. The dashboard is the report on what the operational artefact failed to do.

‍

Where data silos actually live

When retail operators say "data silos," they are usually pointing at architecture. Different vendors. Different schemas. Different update cadences. The POS pushes a nightly batch to the warehouse; the ecommerce platform streams events to a different pipeline; the loyalty system is a third-party SaaS with an export API. The diagnosis is: too many systems, not talking to each other.

But sit on the floor of a flagship store for an afternoon and watch what happens. The friction does not present itself as missing data. It presents itself as a customer at the counter with their phone open to a cart that the associate cannot pull up. It presents itself as an associate manually transcribing a promotional code. It presents itself as a refund that requires the customer to physically come back to the store where the original purchase was made, because the POS and the ecommerce platform recognise each other as foreign entities. It presents itself as a queue that forms because each transaction takes an extra ninety seconds to reconcile.

Seven in ten retailers report that customers forced to wait in a queue give up on making a purchase and leave the store within five minutes. Eighty-two per cent of shoppers actively avoid stores when they see a queue forming. These are not analytics failures. These are checkout failures, and they are happening because the systems that handle the actual transaction have never been wired to share state.

This is what we mean when we say the data silo problem is a checkout problem. The places where customer experience falls apart — the till, the returns desk, the kerbside collection point, the click-and-collect counter — are not analytics surfaces. They are transaction surfaces. And the transaction surfaces are the ones still running on legacy POS infrastructure that, for over 70% of retailers, is more than two years old, with 40% relying on systems more than five years old.

‍

Counting the hidden cost

The reason silos read as a "hidden" cost is that no single line item on a P&L captures them. The cost is dispersed across abandoned carts that never get attributed back to a channel-switching event, returns processed at a loss because the originating order data could not be retrieved, loyalty redemptions written off as goodwill because the till could not validate them, and queue-driven walk-outs that show up only as flat conversion in a quiet quarter.

This dispersion is what makes silos politically difficult inside the business. The marketing team sees a soft acquisition month. The store ops team sees an inexplicable dip in conversion at the till. The finance team sees a refund rate creeping up. Each function diagnoses the problem within its own scope, and the underlying cause — a transaction layer that cannot see across channels — never makes it onto a single roadmap. Forty per cent of customers turn to competitors to complete a purchase when their experience falls apart. That number lands somewhere, but rarely in the budget line of the team that could fix it.

‍

Why analytics-layer solutions can't reach the till

There is a structural reason that warehousing and CDP investments have not closed the operational gap. Analytics layers are downstream. They consume events that have already happened. They cannot write back into the systems that generated those events fast enough to influence the next transaction.

Consider what it would take for a customer's online cart to be honoured at a physical till. The POS would need to query the ecommerce platform's cart state in real time. It would need to authenticate the customer against the same identity layer the ecommerce platform uses. It would need to apply the same promotions, recognise the same loyalty tier, and reconcile inventory against a shared source of truth. None of this is something a warehouse can do, because the warehouse is reading from these systems, not orchestrating between them.

The retailers who have made progress on omnichannel are the ones who have stopped treating the POS as a register and started treating it as a node in a transaction network. That is not an analytics decision. It is an architecture decision about where the cart lives.

‍

The cart is the unification layer

If we accept that the operational impact of data silos is felt at the moment of transaction, then the strategic implication is straightforward: the cart needs to be channel-agnostic. It needs to exist as a single object that any channel — web, mobile app, in-store POS, associate-assisted clienteling tool, customer service desk — can read from, write to, and complete.

This is a meaningful inversion of how most enterprise retail stacks are architected today. In the dominant pattern, each channel has its own cart. The web cart lives in the ecommerce platform. The store cart lives in the POS. The app cart, if there is one, often lives in a third place. The "data silo" terminology obscures the fact that these are not just data silos — they are transaction silos. Each channel can only complete its own transactions.

A unified cart layer changes what is operationally possible. An associate can resume a customer's web session at the counter. A customer can pay for an in-store selection on their phone and walk out. A click-and-collect order can be modified at the pickup desk without reissuing it. A return initiated online can be processed at any store without a paper trail. None of these require new analytics. They require the cart to be a shared object.

‍

What this means for retail ops teams

The reason this distinction matters for operators, rather than only for architects, is that it reframes the prioritisation conversation. If data silos are an analytics problem, the budget request goes to the data team and the timeline is measured in quarters of warehouse migration. If data silos are a checkout problem, the budget request goes to store operations and digital commerce jointly, and the timeline is measured in transaction-level wins that compound week over week.

It also reframes what success looks like. An analytics-led programme of work succeeds when dashboards consolidate. A checkout-led programme of work succeeds when associates can complete transactions they previously had to refuse, when customers stop abandoning carts at the threshold between channels, and when the queue at the till shortens because reconciliation no longer happens in the customer's presence.

Seventy-five per cent of customers say personalised service is a significant factor in where they choose to shop. Personalisation, in the operational sense, is not a marketing message — it is the ability of the person standing in front of the customer to act on what the customer has already done. That capability lives at the checkout layer or it does not exist.

‍

The pragmatic case

None of this is an argument against data warehouses, customer data platforms, or analytics investment. Those layers do important work for merchandising, marketing, and strategic planning. The argument is narrower: if your operational pain is that customers experience your channels as disconnected, the warehouse will not fix that. The cart will.

For enterprise retail operators reading the standard advice on data silos and wondering why two years of CDP rollout has not produced the unified experience the slide deck promised, this is worth sitting with. The technology you bought was solving a different problem than the one your customers were having. The customers were not asking for better reporting. They were asking to be recognised across the threshold of a store door, and the system that controls that threshold is the one running the till.