Time Series Isn’t Enough: How Graph Neural Networks Change Demand Forecasting

in supply-chain planning has historically been handled as a time-series downside.

• Every SKU is modeled independently.
• A rolling time window (say, final 14 days) is used to foretell tomorrow’s gross sales.
• Seasonality is captured, promotions are added, and forecasts are reconciled downstream.

And but, regardless of more and more subtle fashions, the standard issues persist:

• Persistent over-and under-stocking
• Emergency manufacturing adjustments
• Extra stock sitting within the improper place
• Excessive forecast accuracy on paper, however poor planning outcomes in follow

The difficulty is that demand in a provide chain is just not unbiased. It’s networked. For example, that is what simply 12 SKUs from a typical provide chain appear like once you map their shared crops, product teams, subgroups, and storage places.

So when demand shifts in a single nook of the community, the consequences are felt all through the community.

On this article, we step exterior the model-first pondering and have a look at the issue the way in which a provide chain really behaves — as a related operational system. Utilizing an actual FMCG dataset, we present why even a easy graph-based neural community(GNN) basically outperforms conventional approaches, and what which means for each enterprise leaders and information scientists.

An actual provide chain experiment

We examined this concept on an actual FMCG dataset (SupplyGraph) that mixes two views of the enterprise:

Static supply-chain relationships

The dataset has 40 energetic SKUs, 9 crops, 21 product teams, 36 sub-groups and 13 storage places. On common, every SKU has ~41 edge connections, implying a densely related graph the place most SKUs are linked to many others by means of shared crops or product teams..

From a planning standpoint, this community encodes institutional information that usually lives solely in planners’ heads:

“If this SKU spikes, these others will really feel it.”

Temporal operational indicators and gross sales outcomes

The dataset has temporal information for 221 days. For every SKU and every day, the dataset consists of:

• Gross sales orders (the demand sign)
• Deliveries to distributors
• Manufacturing facility items points
• Manufacturing volumes

Right here is an outline of the 4 temporal indicators driving the availability chain mannequin:

As may be noticed, virtually half of the SKU-Day mixtures have zero gross sales. The implication being a small fraction of SKUs drives many of the quantity. It is a traditional “Intermittent Demand” downside.

Additionally, manufacturing happens in rare, massive batches (lumpy manufacturing). Downstream supply is way smoother and extra frequent (low sparsity) implying the availability chain makes use of vital stock buffers.

To stabilize GNN studying and deal with excessive skew, all values are reworked utilizing log1p, an ordinary follow in intermittent demand forecasting.

Key Enterprise Metrics

What does a very good demand forecast appear like ? We consider the mannequin primarily based on two metrics; WAPE and Bias

WAPE — Weighted Absolute Share Error

WAPE measures how a lot of your complete demand quantity is being mis-allocated. As an alternative of asking “How improper is the forecast on common throughout all SKUs?“, WAPE asks the query supply-chain planners really care about within the situation of intermittent demand: “Of all SKU items that have been moved by means of the availability chain to satisfy demand, what fraction was mis-forecast?“

This issues as a result of errors on high-volume SKUs value way over errors on long-tail gadgets. A ten% miss on a high vendor is dearer than a 50% miss on a gradual mover. So WAPE weights the SKU-days by quantity offered, and aligns extra naturally with income impression, stock publicity, plant and logistics utilization (and may be additional weighted by worth/SKU if required).

That’s why WAPE is broadly most popular over MAPE for intermittent, high-skew demand.

[
text{WAPE} =
frac{sum_{s=1}^{S}sum_{t=1}^{T} left| text{Actual}_{s,t} – text{Forecast}_{s,t} right|}
{sum_{s=1}^{S}sum_{t=1}^{T} text{Actual}_{s,t}}
]

WAPE may be calculated at totally different ranges — product group, area or complete enterprise — and over totally different durations, comparable to weekly or month-to-month.

You will need to notice that right here, WAPE is computed on the hardest attainable degree — per-SKU, per-day, on intermittent demand — not after aggregating volumes throughout merchandise or time. In FMCG planning follow, micro-level SKU-daily WAPE of 60–70% is commonly thought of acceptable for intermittent demand, whereas <60% is taken into account production-grade forecasting.

Forecast Bias — Directional Error

Bias measures whether or not your forecasts systematically push stock up or down. Whereas WAPE tells you how improper the forecast is, Bias tells you how operationally costly it’s. It solutions a easy however crucial query: “Can we constantly over-forecast or under-forecast?”. As we’ll see within the subsequent part, it’s attainable to have zero bias whereas being improper more often than not. In follow, constructive bias ends in extra stock, greater holding prices and write-offs whereas unfavourable bias results in stock-outs, misplaced gross sales and repair penalties. In follow, a bit constructive bias (2-5%) is taken into account production-safe.

[ text{Bias} = frac{1}{S} sum_{s=1}^{S} (text{Forecast}_s – text{Actual}_s) ]

Collectively, WAPE and Bias decide whether or not a mannequin is not only correct, its forecasts are operationally and financially usable.

The Baseline: Forecasting With out Construction

To determine a floor ground, we begin with a naïve baseline, which is “tomorrow’s gross sales equal immediately’s gross sales”.

[ hat{y}_{t+1} = y_t ]

This method has:

• Zero bias
• No community consciousness
• No understanding of operational context

Regardless of its simplicity, it’s a robust benchmark, particularly over the brief time period. If a mannequin can not beat this baseline, it’s not studying something significant.

In our experiments, the naïve method produces a WAPE of 0.86, that means practically 86% of complete quantity is misallocated.

The bias of zero is just not a very good indicator on this case, since errors cancel out statistically whereas creating chaos operationally.

This results in:

• Firefighting
• Emergency manufacturing adjustments
• Expediting prices

This aligns with what many practitioners expertise: Easy forecasts are steady — however improper the place it issues.

Including the Community: Spatio-Temporal GraphSAGE

We use GraphSAGE, a graph neural community that permits every SKU to combination info from its neighbors.

Key traits:

• All relationships are handled uniformly.
• Info is shared throughout related SKUs.
• Temporal dynamics are captured utilizing a time collection encoder.

This mannequin doesn’t but distinguish between crops, product teams, or storage places. It merely solutions the important thing query:

“What occurs when SKUs cease forecasting in isolation?”

Implementation

Whereas I’ll dive deeper into the info science behind the function engineering, coaching, and analysis of GraphSAGE in a subsequent article, listed below are a number of the key ideas to know:

• The graph with its nodes and edges types the static spatial options.
• The spatial encoder part of GraphSAGE, with its convolutional layers, generates spatial embeddings of the graph.
• The temporal encoder (LSTM) processes the sequence of spatial embeddings, capturing the evolution of the graph during the last 14 days (utilizing a sliding window method).
• Lastly, a regressor predicts the log1p-transformed gross sales for the subsequent day.

An intuitive analogy

Think about you’re making an attempt to foretell the worth of your home subsequent month. The value isn’t simply influenced by the historical past of your personal home — like its age, upkeep, or possession data. It’s additionally influenced by what’s occurring in your neighborhood.

For instance:

• The situation and costs of homes much like yours (comparable development high quality),
• How well-maintained different homes in your space are,
• The supply and high quality of shared companies like colleges, parks, or native legislation enforcement.

On this analogy:

• Your own home’s historical past is just like the temporal options of a selected SKU (e.g., gross sales, manufacturing, supply historical past).
• Your neighborhood represents the graph construction (the perimeters connecting SKUs with shared attributes, like crops, product teams, and many others.).
• The historical past of close by homes is just like the neighboring SKUs’ options — it’s how the habits of different comparable homes/SKUs influences yours.

The aim of coaching the GraphSAGE mannequin is for it to study the operate f that may be utilized to every SKU primarily based on its personal historic options (like gross sales, manufacturing, manufacturing facility points, and many others.) and the historic habits of its related SKUs, as decided by the sting relationships (e.g., shared plant, product group, and many others.). To depict it extra exactly:

embedding_i(t) =
  f( own_features_i(t),
     neighbors’ options(t),
     relationships )

the place these options come from the SKU’s personal operational historical past and the historical past of its related neighbors.

The Outcome: A Structural Step-Change

The impression is kind of exceptional:

In sensible phrases:

• The naïve method misallocates practically 86% of complete demand quantity
• GraphSAGE reduces this error by ~27%

The next chart exhibits the precise vs predicted gross sales on the log1p scale. The diagonal pink line depicts excellent forecast, the place predicted = precise. As may be seen, many of the excessive quantity SKUs are clustered across the diagonal which depicts good accuracy.

From a enterprise perspective, this interprets into:

• Fewer emergency manufacturing adjustments
• Higher plant-level stability
• Much less guide firefighting
• Extra predictable stock positioning

Importantly, this enchancment comes with none further enterprise guidelines — solely by permitting info to circulate throughout the community.

And the bias comparability is as follows:

At underneath 5%, the gentle forecasting bias GraphSAGE introduces is effectively inside production-grade limits. The next chart depicts the error within the predictions.

It may be noticed that:

• Error is negligible for many of the forecasts. Recall from the temporal evaluation that sparsity in gross sales is 46%. This exhibits that the mannequin has realized this, and is appropriately predicting “Zero” (or very near it) for these SKU-days, creating the height on the middle.
• The form of the bell curve is tall and slim, which signifies excessive precision. Most errors are tiny and clustered round zero.
• There’s little skew of the bell curve from the middle line, confirming the low bias of 4.5% we calculated.

In follow, many organizations already bias forecasts intentionally to guard service ranges, relatively than threat stock-outs.

Let’s have a look at the impression on the SKU degree. The next chart exhibits the forecasts for the highest 4 SKUs by quantity, denoted by pink dotted strains, towards the actuals.

A number of observations:

• The forecast is reactive in nature. As marked in inexperienced circles within the first chart, the forecast follows the precise on the way in which up, and likewise down with out anticipating the subsequent peak effectively. It is because GraphSAGE considers all relations to be homogeneous (equally vital), which isn’t true in actuality.
• The mannequin under-predicts excessive spikes and compresses the higher tail aggressively. GraphSAGE prefers stability and smoothing.

Here’s a chart displaying the efficiency throughout SKUs with non-zero volumes. Two threshold strains are marked at WAPE of 60% and 75%. 3 of the 4 highest quantity SKUs have a WAPE < 60% with the fourth one simply above. From a planning perspective, this can be a strong and balanced forecast.

Takeaway

Graph neural networks do greater than enhance forecasts — they modify how demand is known. Whereas not excellent, GraphSAGE demonstrates that construction issues greater than mannequin complexity.

As an alternative of treating every SKU as an unbiased downside, it permits planners to cause over the availability chain as a related system.

In manufacturing, that shift — from remoted accuracy to network-aware decision-making — is the place forecasting begins to create actual financial worth.

What’s subsequent? From Connections to That means

GraphSAGE confirmed us one thing highly effective: SKUs don’t reside in isolation — they reside in networks.
However in our present mannequin, each relationship is handled as equal.

In actuality, that’s not how provide chains work.

A shared plant creates very totally different dynamics than a shared product group. A shared warehouse issues in a different way from a shared model household. Some relationships propagate demand shocks. Others dampen them.

GraphSAGE can see that SKUs are related — however it can not study how or why they’re related.

That’s the place Heterogeneous Graph Transformers (HGT) are available.

HGT permits the mannequin to study totally different behaviors for several types of relationships — letting it weigh, for instance, whether or not plant capability, product substitution, or logistics constraints ought to matter extra for a given forecast.

Within the subsequent article, I’ll present how shifting from “all edges are equal” to relationship-aware studying unlocks the subsequent degree of forecasting accuracy — and improves the standard of forecast by including that means to the community.

That’s the place graph-based demand forecasting turns into actually operational.

Join with me and share your feedback at www.linkedin.com/in/partha-sarkar-lets-talk-AI

Reference

SupplyGraph: A Benchmark Dataset for Supply Chain Planning using Graph Neural Networks : Authors: Azmine Toushik Wasi, MD Shafikul Islam, Adipto Raihan Akib

_{Photos used on this article are synthetically generated. Charts and underlying code created by me.}