The 5 weak signals that a product should be liquidated (but your systems ignore)

In the life of a retail product, there's a pivot moment. A precise moment where it silently shifts from "active SKU" to "future problem." Weeks before the numbers clearly show underperformance. Months before the inevitable markdown. That moment is the point where a fast decision — early liquidation, transfer, supplier return, repositioning — would have preserved most of the margin. And it's precisely the moment most retail systems miss.

Not because the data doesn't exist. It all exists, somewhere in the information system. But because no traditional tool — ERP, BI, planning — is designed to capture weak signals: those small variations that seem insignificant in isolation, but which, cross-referenced, tell a clear story for anyone who knows how to read them.

The result is familiar to every retail leadership team. The liquidation decision almost always arrives too late. By the time the numbers become obvious, the damage is done: the stock has aged, attractiveness has dropped, competitors have already slashed prices, and now you need to chase -50% or -60% markdowns to recover what could have been saved with -20% six weeks earlier.

This article looks in detail at the five weak signals that, in a modern retail system, should automatically trigger an early liquidation alert — and why your current systems, with rare exceptions, don't see them.

Why weak signals slip through unnoticed

Before getting into the detail, a word on the general mechanics of the problem. Classic retail systems work on explicit thresholds. You define that an under-covered product triggers an alert, that a sell-through below X% by week Y becomes a case to handle, that dormant stock past Z weeks goes to review. These thresholds are useful, but they have a major flaw: they only trigger once the problem is already advanced.

In the real dynamic of a product, the trajectory toward underperformance starts long before the thresholds are hit. It reads in subtle combinations of variables: a gradual slowdown in certain stores, a drift in the basket the product appears in, price elasticity starting to flatten, quiet cannibalization by a new arrival in the assortment. None of these variations, taken in isolation, triggers an alert. It's their convergence that should — but which no tool reads automatically.

On top of this, a powerful cognitive bias. As long as a product keeps selling — even mediocrely — the urgency isn't there. Merch teams focus on the SKUs already in the red. The ones still in the gray zone, but slowly sliding, get pushed back. "We'll see at the next review." Except between two reviews, the product has kept sliding, and the window for early liquidation has closed.

Let's now look at the five signals that should, in an ideal world, automatically trigger a decision — and that your systems let through.

Signal #1: relative deceleration, invisible in absolute numbers

The first signal, and probably the most systematically missed, is relative deceleration. The product keeps selling. Its weekly volume stays acceptable. Its stock cover looks healthy. In absolute terms, nothing's wrong.

But when you compare its trajectory to its peers — products in the same category, the same price cluster, launched on comparable dates — you find its velocity diverges negatively. Its peers accelerate, it stagnates. Its peers stagnate, it decelerates. This divergence, almost imperceptible week to week, almost always signals a positioning, attractiveness, or product-market-fit problem that won't resolve on its own.

Why systems miss this signal: because they look at each product in isolation. The dashboard shows product X is selling, so all good. Nobody checks whether product X is selling slower than its category would suggest. That relative comparison, at the scale of tens of thousands of SKUs, is simply impossible by hand — and almost never automated.

Consequence: the problem is identified six to ten weeks after it started. By the time absolute volume finally collapses, it's already too late for a cost-controlled liquidation.

Signal #2: geographic fragmentation of sell-through

Second signal, particularly treacherous: a product holding an acceptable average sell-through, but with a store-by-store dispersion that's widening.

Concretely: across the network, the product shows for example 65% sell-through at mid-season — a number that triggers no alert. But in reality, that 65% average hides a fractured truth. In 20% of stores, the product is already at 90% and near stockout. In 40% of stores, it's at 60-70% in a neutral zone. In the remaining 40%, it's at 30-40% and heading for a catastrophic end of season.

That dispersion is an extremely informative signal. It says two things at once: the product works for some store clusters (the ones where it sells), and it fails for others (the ones where it accumulates). The right decision is neither to mark down uniformly, nor to let it run — it's to liquidate fast in the underperforming stores (via transfer to the overperforming ones, or targeted local markdown), while protecting the price in stores where it works.

Why systems miss this signal: because steering is mostly done on network averages or large clusters, never on fine SKU/store dispersion. And even when the data exists, no one has time to handle it at that granularity. Result: the product stays everywhere, you mark down everywhere at season's end, and you eat the cost of what could have been prevented.

Signal #3: drift in the purchase context (the basket signature)

Third signal, more subtle and almost never exploited: the drift of the basket the product appears in.

A product that's selling well appears, statistically, in certain types of baskets. It's bought with certain other products, at certain times of the week, by certain customer types. That purchase signature is stable as long as the product is working. And when it starts to drift — for example, the product only appears in solo baskets now, or in end-of-sale baskets, or with other products that are themselves struggling — it's generally an early signal that its commercial dynamic is degrading.

That drift precedes the volume drop by several weeks. When a product stops being a complement to other purchases and becomes an isolated opportunity buy, its forward trajectory is very often negative. You'll have to liquidate — the only question is whether you do it when the signal appears, or six weeks later when volume has confirmed.

Why systems miss this signal: because it requires transactional co-occurrence analysis that few chains industrialize. The data is in the POS receipt, but exploiting it at scale, continuously, across tens of thousands of SKUs, requires an analytical layer that classic BI doesn't carry.

Signal #4: the collapse of price elasticity

Fourth signal, particularly important for chains already running promotional or pricing actions: the collapse of price elasticity.

A commercially healthy product reacts to price variations. A -15% promo generates a measurable sales uplift. A moderate markdown re-energizes rotation. That's the normal behavior of a product in the attractiveness zone.

When elasticity collapses — when promotions or markdowns no longer generate the expected uplift — it's almost always a sign you've passed the point where price is still the lever. The problem isn't pricing anymore: it's desirability, perceived obsolescence, market saturation, or internal cannibalization. In that case, continuing to mark down gradually is pointless — worse, you're destroying margin for near-zero effect. The right decision is then a fast, deep liquidation, to recover what can be recovered and move on.

Why systems miss this signal: because measuring price elasticity requires comparing actual uplift from a promo to the expected uplift — which implies a predictive model, not just descriptive tracking. Very few retailers have industrialized that measurement at the SKU/store level continuously. Most just note, after the fact, that "the promo didn't work" — without turning that observation into an automatic liquidation signal.

Signal #5: cannibalization by a new entrant in your assortment

Fifth signal, and perhaps the hardest to accept humanly: cannibalization by a new product in your own assortment.

When a chain introduces a new SKU — capsule collection, seasonal launch, repositioning — it mechanically pulls part of its demand from existing products. That's expected. The problem is this cannibalization is almost never measured explicitly. You see the new product performing, you celebrate. You don't see that in its wake, two or three older SKUs are decelerating — not from a loss of intrinsic attractiveness, but simply because the new one is capturing their customer base.

Identifying these cannibalizations early enables a proactive decision: do you liquidate the cannibalized product fast (before its rotation collapses entirely), or reposition the new entrant to reduce overlap? In most cases, early liquidation is the better option — but it requires having seen the cannibalization while there was still time.

Why systems miss this signal: because identifying a cannibalization requires cross-referencing a new product's performance with the simultaneous drift of adjacent SKUs (in attributes, price, use), controlling for other explanatory factors (seasonality, weather, competitor moves), and isolating the net effect. It's a causal analysis, not a count. No traditional tool does it automatically at the scale of a full assortment.

What the five signals have in common: what they say about your stack

If you look closely at these five signals, you'll notice they share three characteristics.

First characteristic: they're relational, not absolute. None can be detected by looking at a product in isolation. Signal #1 requires comparing the product to its peers. #2 requires comparing stores to each other. #3 requires comparing the product to itself over time. #4 requires comparing actual uplift to expected uplift. #5 requires comparing a product to a new entrant. That relational dimension is exactly what classic tools, organized around absolute thresholds, systematically miss.

Second characteristic: they require continuous monitoring, not periodic review. Weak signals don't show up in a monthly or quarterly review. They form gradually, and the value of the information is almost entirely in the earliness of detection. Spotting a relative deceleration six weeks after it appeared isn't a weak signal anymore — it's a late observation. The same information, exploited three days after the drift starts, enables a near-zero-cost action.

Third characteristic: they require a prescriptive layer, not just a descriptive one. Detecting the signal isn't enough — you have to translate it into action. And the action depends on the signal: a relative deceleration might justify a transfer, a geographic fragmentation justifies a targeted local markdown, a cannibalization justifies a repositioning or a liquidation. That signal-to-decision translation is, everywhere, the missing layer in current retail stacks.

Why your current tools weren't built for this

Let's say it clearly: this isn't a flaw in your tools. It's their very design that makes them blind to these signals.

Your ERP is a bookkeeping tool. It records flows, it guarantees accounting integrity. It doesn't do comparative analysis, doesn't track elasticity, doesn't detect cannibalization.

Your BI is a visualization tool. It answers the questions you ask it, but it doesn't ask questions on its own. It can display the five signals, provided an analyst spends hours modeling them one by one — but it doesn't monitor them automatically, continuously, across tens of thousands of SKUs.

Your planning tool reasons at aggregate meshes and long horizons. It's not built to capture micro-drifts in real time.

Your ML forecasts, even excellent ones, predict volumes — they don't formulate action recommendations based on complex relational signals.

None of these tools is bad. They're simply insufficient, in isolation, to carry continuous prescriptive detection logic. That logic requires one more layer: a system that continuously scans the network, cross-references signals, formulates actionable recommendations, and executes them without breakage.

The cost of blindness: what retailers leave on the table

How much does this blind zone cost? The order of magnitude is known to every chain that's done the exercise of quantifying. On a typical retail assortment, 5 to 10% of SKUs are on a liquidation trajectory at any moment in their cycle — and the earliness of the decision massively determines the final cost.

An early liquidation, triggered on time, typically costs -20% to -30% in markdown depth, on stock still reasonably valued. A late liquidation, triggered by standard thresholds, costs -50% to -70%, on already-depreciated stock, plus the hidden costs of extended carry, shelf space, and missed commercial opportunities.

The gap between these two scenarios, at the scale of a retailer generating €500M in revenue, regularly comes out to €3 to €8 million per year of unrecovered gross margin. That's the bill, never clearly presented in any reporting, for the inability to capture weak signals on time.

The Solya approach: detect, recommend, execute

That's precisely the layer Solya brings to your existing stack. Not to replace your ERP, your BI, or your models — but to add on top the continuous prescriptive detection logic that turns weak signals into executed decisions.

Concretely, Solya connects to your data sources — POS, ERP, e-commerce, supply chain — and rebuilds a unified view of the network at the SKU/store level, continuously updated. The engine permanently scans that view looking for cross-signals: relative decelerations, geographic fragmentations, basket-context drifts, elasticity collapses, latent cannibalizations. For each detected signal, it formulates an actionable recommendation — early liquidation, priority transfer, store-cluster targeted markdown, supplier return — integrating your business rules and operational constraints.

The teams keep their hand on every structural trade-off. They define thresholds, validate sensitive liquidations, adjust strategy based on commercial context. Solya takes care of the mechanics: continuously scanning tens of thousands of SKU/store pairs, cross-referencing signals, identifying opportunities, executing validated actions.

It's not another visualization tool. It's the layer that turns your data stack, however sophisticated, into measurable operational performance — by recovering the margin that's lost today in the silence of ignored weak signals.

The real question to ask

How many products, in your current assortment, are slowly sliding toward a late liquidation — without any of your systems flagging it?

If you can't answer that question precisely, you're probably letting a significant share of recoverable margin slip every season. Not because the data is missing — it's all there, somewhere in your systems. But because no layer in your current stack is designed to cross-reference it and turn it into timely action.

That reservoir, immense and structurally invisible, is what leading retailers have started addressing. Not by piling on dashboards. Not by perfecting their forecasts. But by building the prescriptive detection layer that turns weak signals into recovered margin — and which makes, in the end, the difference between an assortment you endure and an assortment you steer.

---

What weak signals are slipping through in your network?

At Solya, we offer retail leaders a personalized 30-minute diagnostic to identify, on your own perimeter, the SKUs currently showing the most advanced signals of latent underperformance — and to quantify the margin potential recoverable through early detection.

👉 [Book your Solya diagnostic] — 30 minutes, by video, with one of our retail experts.

You'll walk away with:

• A map of the weak signals currently ignored by your systems
• A quantified estimate of margin potential recoverable through anticipated detection
• The first high-ROI use cases to industrialize this detection across your assortment