MAPE, Bias, and Fill Rate: The Three Metrics That Tell You If Your Demand Plan Is Working

Most brands that have a demand plan don't measure whether it's working.

They track revenue. They watch inventory levels. They feel the stockouts when they happen and process the markdowns when they have to. But they don't have a number that tells them, objectively, whether their forecasting process is getting better or worse over time.

That's a problem, because a demand plan you can't measure is a demand plan you can't improve. You're flying without instruments.

Three metrics give you the instrument panel: MAPE, bias, and fill rate. Each one measures something different. Together they tell you whether your plan is accurate, whether it's directionally honest, and whether it's translating into the inventory outcomes your customers actually experience.

Here's what each one means, what good looks like at different stages of growth, and why you need all three.

MAPE: How Big Are Your Forecast Errors?

MAPE stands for Mean Absolute Percentage Error. It measures the average size of your forecast errors, expressed as a percentage of actual demand.

The formula:

MAPE = Average of (|Actual − Forecast| ÷ Actual) × 100

In plain language: for each period you're measuring, calculate how far off your forecast was as a percentage of what actually sold. Average those percentages. That's your MAPE.

A simple example:

MAPE = (22 + 13 + 32) ÷ 3 = 22.3%

Lower is better. A MAPE of 22% means your forecast is off by an average of 22% each month — in either direction.

What good looks like:

For a brand doing around $10M in revenue, a MAPE of 25–35% is a normal and acceptable starting point. It reflects real-world conditions: limited history on some SKUs, promotional noise, channel transitions, and a planning process that's still maturing. World-class demand planning at enterprise CPG companies runs 10–15% — but those organizations have years of clean data, dedicated planning teams, and statistical infrastructure most small brands don't have yet.

Don't benchmark yourself against enterprise. Benchmark yourself against where you were last quarter.

One important limitation: MAPE tells you how big your errors are. It doesn't tell you which direction they go. That's what bias measures.

Bias: Are You Consistently Wrong in the Same Direction?

This is the metric most small brands don't track — and it's the one that catches the most expensive planning problems.

Forecast bias measures the directional error in your forecast: are you consistently over-forecasting, consistently under-forecasting, or roughly balanced?

The formula:

Bias = (Sum of Forecasts − Sum of Actuals) ÷ Sum of Actuals × 100

A positive bias means you're over-forecasting — you're predicting more demand than actually materializes. A negative bias means you're under-forecasting — demand keeps coming in above your plan.

Why this matters:

You can have an acceptable MAPE and a terrible bias. If your forecast is off by 20% every month but the errors are random — sometimes high, sometimes low — the plan is imprecise but not systematically broken. Your inventory position will be roughly right on average.

But if your forecast is off by 20% every month and always in the same direction — always over, always under — you have a process problem that the MAPE number alone won't surface. Consistent over-forecasting means you're producing too much every cycle, tying up cash in inventory that sits. Consistent under-forecasting means you're stocking out regularly, losing sales and damaging retailer relationships.

The direction of the bias tells you what the error is costing you:

• Positive bias (over-forecasting): Cash tied up in excess inventory. Storage costs. Markdown risk. Write-off exposure on products with shelf lives.
• Negative bias (under-forecasting): Stockouts. Lost sales. Retailer chargebacks for short shipments. Customers who found a competitor and didn't come back.

Both are problems. They just sit in different places on your P&L.

What good looks like: A bias within +/- 10% is healthy. It means your errors are roughly balanced — you're not systematically skewed in one direction. Beyond +/- 10%, you have a directional problem worth diagnosing.

What bias usually means when it's high:

A consistent positive bias (always over-forecasting) almost always traces back to one of two things: the forecast is being inflated by promotional assumptions that don't materialize at the expected lift, or the forecast is being set to match a revenue target rather than a demand signal. You're forecasting what you want to happen rather than what the data suggests will happen.

A consistent negative bias (always under-forecasting) usually means the opposite: the baseline forecast is too conservative, seasonal peaks are being underestimated, or promotional lift isn't being modeled at all.

In both cases, the bias is telling you something structural about how the forecast is being built. It's not bad luck. It's a process pattern.

Fill Rate: Is the Plan Working for Your Customer?

MAPE and bias measure the quality of your forecast. Fill rate measures the outcome your customer experiences.

Fill rate is the percentage of demand you were able to fulfill on time and in full. It's the metric that sits at the intersection of demand planning and inventory management — you can have a good forecast and still have a low fill rate if your lead times are wrong, your safety stock is too thin, or your supplier execution is unreliable.

There are a few versions of fill rate worth knowing:

• Order fill rate: Percentage of customer orders shipped complete. An order for 100 units that ships with 85 units is an 85% order fill rate.
• Line fill rate: Percentage of order line items fulfilled completely. If a customer orders 10 different SKUs and 8 ship complete, that's an 80% line fill rate.
• Case fill rate: Common in retail and wholesale — percentage of cases (units of a standard shipper) fulfilled on time and in full. This is the number most major retailers track and will hold you accountable to.

What good looks like:

For brands selling to major retailers, a case fill rate below 95% will typically result in chargebacks. 98%+ is the standard most large retailers expect from their suppliers. Below 95% consistently is a compliance problem, not just a planning problem.

For DTC brands, fill rate expectations are set by your own service standards. Most brands aim for 98%+ on orders — a 95% fill rate on your own website means roughly 1 in 20 orders ships short or delayed, which creates customer service volume and return risk.

Why fill rate belongs alongside MAPE and bias:

A brand can have excellent forecast accuracy and poor fill rates. This usually means the supply chain isn't executing to the plan — lead times are longer than modeled, safety stock is miscalculated, or the 3PL has receiving delays that compress the available inventory window. MAPE tells you the forecast is right. Fill rate tells you the plan isn't translating to execution.

Conversely, a brand can have high fill rates with poor forecast accuracy if they're carrying large safety stock buffers that absorb the forecast error. The fill rate looks fine. The balance sheet has too much inventory. MAPE surfaces the hidden cost.

You need both numbers to see the full picture.

How to Use All Three Together

Each metric is a diagnostic tool. The pattern across all three tells you where the problem is and what to fix.

High MAPE, balanced bias, acceptable fill rate:

Your forecast errors are large but random, and you have enough safety stock to absorb them. The plan is imprecise but not directionally broken. Focus on improving forecast methodology — better historical baselines, promotional modeling, seasonality adjustments.

Acceptable MAPE, high positive bias, declining fill rate:

You're systematically over-forecasting but running into production or supply constraints that mean inventory isn't arriving despite the inflated plan. Or your fill rate is actually fine but you're carrying more inventory than you need to maintain it. Look at whether the forecast is being set to match a target rather than a demand signal.

Acceptable MAPE, high negative bias, low fill rate:

Classic under-forecasting pattern. Demand is consistently coming in above plan, safety stock is getting consumed faster than expected, and stockouts are following. Revisit your baseline demand assumptions and check whether promotional lift is being underestimated or ignored.

Good MAPE and bias, low fill rate:

The forecast is good but execution is failing. Lead times are longer than modeled, 3PL receiving is slower than expected, or reorder points are set too low. The demand plan is working — the supply chain isn't.

All three healthy:

Your plan is accurate, directionally honest, and translating to good customer outcomes. The only remaining question is whether you're measuring at the right level of detail — by SKU and channel, not just in aggregate.

How to Start Measuring

If you're not tracking any of these metrics today, start with MAPE. It requires only two data points: your monthly forecast and your monthly actuals. Calculate it for your top ten SKUs, going back as many months as you have clean data.

Once MAPE is running, add bias. Same data, different formula — it takes ten minutes to add to whatever model you're already maintaining.

Fill rate requires your order and fulfillment data. Your 3PL or order management system should be able to produce this report. If it can't, that's worth addressing separately.

Run all three monthly. Track them over time. A demand planning process that's improving will show it in the numbers — MAPE coming down, bias converging toward zero, fill rate holding steady or improving. A process that isn't improving will show that too. The metrics don't lie about the process, even when the process feels like it's working.

Not sure how your current forecast accuracy compares to where it should be or what's driving your forecast error?

Izba's demand planning audit looks at your data, your method, and your outcomes. We'll tell you what the numbers are saying and what to fix first.

Book your audit at izba.co