S&OP Series (1/3): Baseline Accuracy Drives the Cycle: Issue 02

Assortment & Baseline Forecasts & their $ impact

What are the core phases of S&OP?

1. Assortment

2. Baseline Forecast

3. Consensus

4. Supply Review

5. Executive Review

This newsletter covers the first two:

Assortment – why launch and discontinue decisions only work when tied to forecasting.

Baseline Forecasting – how most teams bake in error, and how to fix it with outlier handling and the right methods.

We’ll walk through real examples, show what better forecasting actually looks like, and tie it directly to dollars saved.

Why Forecasting Starts With Assortment

Everyone agrees that managing your inventory is important. But what does that actually mean in detail?

It all starts with Assortment.

When do we launch new products? When do we discontinue old ones?

It sounds obvious, but if teams actually do this rigorously, it’s rare that they link those decisions actively to inventory and forecasting.

You want to stock up just enough before launch. You want to stock out exactly on the day you planned to discontinue. Not with pallets of deadweight left behind.

But assortment decisions alone aren’t enough. Once you’ve defined what to launch or discontinue, you still need to forecast it. Accurately.

That’s where the real challenge starts.

Where Forecasting Goes Wrong

Most teams still rely on one-methods-rule-them-all forecasting. You look at a time series and see growth, seasonality, maybe some outliers.

You apply a moving average. Guess what happens? You bake the outliers into your forecast.

Now do this across 200 SKUs and 5 channels. Suddenly, your error compounds everywhere.

Here’s what that looks like:

In this picture, two outliers are inside the 90 day moving average forecast. Can you guess which?

And here’s the thing nobody is talking about: forecast error doesn’t just hurt accuracy. It costs you real money.

The worse your forecast error (e.g., MAE: Mean Absolute Error, the average size of your forecast mistakes), the more safety stock you need.

And more safety stock means more cash tied up, lower inventory turns, and higher risk of write-offs.

You’re not just over-ordering. You’re locking in cash and burning margin.

How to Clean Up Forecast Error

1. Remove outliers first.

2. Run multiple forecast methods per SKU, per channel.

3. Measure error across a past period.

4. Choose the method with the lowest error.

Let’s walk through it.

Step 1: Remove outliers

Here’s what the cleaned forecast looks like when you strip out anomalies.

Now we get a better picture: the spikes in February and April were indeed anomalies and shouldn’t guide future inventory.

Take the moving average again, this time on the outlier-cleaned data.
Same method. Different input, different result.
June changes by 47% – that’s our forecast, since we’re still in May.

But as we all know, a moving average doesn’t understand seasonality.

Or weigh growth automatically from the past.

Step 2: Apply a better forecast model

That’s where other models come in.

We’ve got statistical, machine learning, deep learning, and even transformer-based models like the ones powering GPT (more on those later).

Let’s keep it simple for now. Use autoETS.

In one sentence: autoETS decomposes your time series into error, trend, and seasonal components, then auto-selects the best combo to predict future values.

What can we see above?

It captures seasonality for June better than the method before.

That’s great, but what does that actually mean in dollars?

Because at the end of the day, better forecasts only matter if they improve the business.

So let’s link it to real capital.

Imagine you’re running 200 SKUs, doing $100M in revenue, and holding $15M in average inventory.

And let’s be conservative and say, by removing outliers and picking better methods, you improve MAE by 15%.

The chart below shows exactly that:

The yellow line shows what just a 15% accuracy boost looks like.

Doesn’t look like a lot, but it can free up hundreds of thousands in working capital.

We’ll show it.

That improvement in forecast accuracy doesn’t just look good on a chart, it directly affects how much safety stock you need to hold.

What if you could cut safety stock, without risking stock outs? Let’s break it down:

How Better Forecasts Cut Safety Stock

• Z = 1.64 (this reflects a 95% service level, in other words, you’ll meet demand in 95% of cycles)

• Lead Time = 13 weeks (~3 months)

• MAE = Mean Absolute Error

This simplifies to:

SS ≈ 3.55 × MAE × Avg Monthly Inventory

• 3.55 comes from

  • Z = 1.64

  • 1.25 (conversion from MAE to standard deviation σ)

  • √(3 months) ≈ 1.732

Let’s run the numbers on our example to see what that means in practice:

Earlier, we said we’re holding $15M in inventory spread across 200 SKUs.

• Average Annual Inventory = $15M ÷ 200 = $75,000

• Average Monthly Inventory = $75,000 ÷ 12 = $6,250

• Let’s say - on average - 30% of that is safety stock → $1,875 per SKU/month

Want to sanity-check your own MAE?

Here’s how to reverse-engineer it based on how much safety stock you’re holding today.

1. Reverse-engineer today’s MAE

Let’s say your team tells you they’re holding 30% of inventory as safety stock, in our case, that’s $1,875 per SKU/month.

2. What if you cut MAE by 15%?

• New MAE = 0.85 × 8.5% = 7.2%

• New SS % = 3.55 × 7.2% ≈ 25.5% of inventory

• New SS $ = 0.255 × $6,250 ≈ $1,593.75

• Monthly reduced inventory per SKU = $1,875 − $1,593.75 = $281.25

• Annual reduced inventory per SKU = $281.25 × 12 = $3,375

The Inventory Impact

$281.25 × 12 × 200 SKUs = $675,000

…in working capital unlocked, just by reducing safety stock.

That’s before we even talk about fewer write-offs, fewer stock outs, or fewer emergency shipments.

There’s likely another $1M+ to be unlocked. More on that in a future post.

So what’s next?

Next time, we’ll look at what happens after the baseline: when sales and marketing override the forecast, and how to align everyone through a structured consensus forecast. Because planning doesn’t stop at the baseline.

See you next time!

Leon