"Our forecasts are off."
That statement is always true. Forecasts are predictions about an uncertain future. They're never exactly right. The question isn't whether your forecasts are wrong—it's whether they're wrong in ways that hurt your business.
Why Forecast Accuracy Matters
Forecast error flows downstream into every inventory decision:
Underforecast → Stockouts → Lost sales and disappointed customers
Overforecast → Excess inventory → Tied-up cash and potential write-offs
The tighter your forecast accuracy, the leaner you can run inventory while maintaining service levels. That's real money—in working capital freed up, in storage costs avoided, in markdowns prevented.
But accuracy has diminishing returns. Going from 60% to 80% accuracy transforms your operations. Going from 85% to 90% is expensive and might not be worth it.
How to Measure Forecast Accuracy
Before benchmarking, you need consistent measurement.
MAPE (Mean Absolute Percentage Error)
The most common metric. Measures average percentage error regardless of direction.
Formula: MAPE = Average of |Actual - Forecast| / Actual × 100
Example: You forecasted 100 units, sold 80. Error = |80-100|/80 = 25%
Forecast Accuracy = 100% - MAPE = 75%
Weighted MAPE
Standard MAPE treats a 50% error on a 10-unit SKU the same as a 50% error on a 1,000-unit SKU. Weighted MAPE adjusts for volume.
Formula: Sum of |Actual - Forecast| / Sum of Actual × 100
This gives more weight to high-volume SKUs where accuracy matters most.
Bias
MAPE doesn't tell you if you're consistently over or under forecasting. Bias does.
Formula: Bias = (Sum of Forecast - Sum of Actual) / Sum of Actual × 100
Positive bias: You're over-forecasting (building excess inventory)
Negative bias: You're under-forecasting (risking stockouts)
Some businesses deliberately bias slightly high (to protect fill rates) or slightly low (to minimize inventory). Track it so you know what you're doing.
Forecast Accuracy at Different Levels
Accuracy varies by aggregation level:
- Total company: Easiest to forecast, typically 90%+ accuracy
- Category: Somewhat accurate, typically 80-90%
- SKU: Hardest, typically 70-85%
- SKU-Location: Even harder if you plan by warehouse
Measure accuracy at the level where you make decisions. If you order by SKU, SKU-level accuracy is what matters.
Forecast Accuracy Benchmarks by Industry
What "good" looks like varies by industry:
Grocery / Staples
Benchmark: 80-95% accuracy
Steady demand, low volatility, established products. These are the easiest categories to forecast.
General CPG
Benchmark: 70-85% accuracy
More variety, more promotional activity, more new product launches. Still reasonably forecastable.
Fashion / Apparel
Benchmark: 60-75% accuracy
Trend-driven, seasonal, short product lifecycles. Even the best forecasters struggle here.
Consumer Electronics
Benchmark: 65-80% accuracy
Product launches, rapid obsolescence, and competition make forecasting challenging.
Promotional Products
Benchmark: 60-70% accuracy
When demand is driven by promotions rather than organic patterns, volatility is high and forecasting is hard.
Benchmarks for CPG Brands
For the typical mid-market CPG brand, here's a realistic target framework:
Hero SKUs (Top 20% of volume)
Target: 80-90% accuracy
These products drive your business. They deserve the most attention and should be your most accurate forecasts.
Core SKUs (Middle 60%)
Target: 70-85% accuracy
Solid performers that need good forecasts but don't warrant obsessive attention.
Long Tail (Bottom 20%)
Target: 60-75% accuracy
Low volume means high percentage volatility. Don't chase accuracy here—focus on not running out of your best sellers.
New Products (First 6 months)
Target: 50-70% accuracy
New products are inherently unpredictable. Expect high error rates and plan your safety stock accordingly.
Why Chasing Perfect Accuracy Wastes Resources
There's a point where improving forecast accuracy costs more than the benefit:
The Effort Curve
Going from 50% to 70% accuracy might require:
- Cleaning up your data
- Implementing a basic process
- Consistent weekly reviews
Going from 70% to 80% might require:
- Better tools
- More sophisticated methods
- More analyst time
Going from 80% to 90% might require:
- Advanced statistical models
- Machine learning
- Dedicated data science resources
Each increment costs more while delivering less additional value.
The Uncertainty Floor
Some demand is simply unpredictable. Weather, competitor actions, viral social media moments—these affect your sales in ways no model can anticipate.
At some point, you hit a floor where further accuracy improvements aren't possible without predicting the unpredictable.
The Safety Stock Alternative
Instead of obsessing over forecast accuracy, you can often achieve the same business outcomes by adjusting safety stock.
If improving accuracy from 80% to 85% would cost $50,000 in tools and analyst time, but holding 5% more safety stock costs $10,000—hold the safety stock.
How to Improve Forecast Accuracy (When It Makes Sense)
If you're below benchmark, here's where to focus:
Clean Your Data First
Garbage in, garbage out. Fix data quality issues—incorrect sales histories, stockout periods counted as low demand, missing SKUs—before trying to improve methods.
Focus on High-Impact SKUs
Don't try to improve everything at once. Focus accuracy efforts on the SKUs where forecast error actually hurts: high-volume products and products with long lead times.
Add Demand Drivers
If promotions drive demand, build them into your forecast. If seasonality exists, model it. The biggest accuracy gains come from incorporating information you're currently ignoring.
Shorten Review Cycles
Forecasts that sit for months without updates drift from reality. Weekly reviews catch errors faster and allow course corrections.
Learn From Errors
When forecasts miss badly, do a post-mortem. Was it a data issue? A missed event? An unforeseeable demand spike? Each miss is a learning opportunity.
Accuracy Isn't Everything
Forecast accuracy is one metric. Others matter too:
Bias Matters
90% accuracy with consistent over-bias builds excess inventory. 80% accuracy with minimal bias might produce better inventory outcomes.
Response Time Matters
If you can respond quickly to demand signals (short lead times, flexible suppliers), you can tolerate lower forecast accuracy.
Process Discipline Matters
A consistent 75% accuracy you can plan around beats volatile accuracy that swings from 60% to 90% month to month.
Key Takeaways
- Perfect forecasts don't exist—the question is whether accuracy is good enough for business outcomes
- Measure accuracy at the level you make decisions (usually SKU level)
- CPG benchmarks: 80-90% for hero SKUs, 70-85% for core products, 60-75% for long tail
- Improving accuracy has diminishing returns—sometimes safety stock is cheaper than better forecasts
- Focus accuracy efforts on high-impact SKUs: high volume and long lead times
- Bias and consistency matter alongside raw accuracy numbers
Frequently Asked Questions
Q: What forecast accuracy should I target?
For CPG brands at SKU level: 80-85% for your most important products, 70-80% overall. These are achievable with good processes and deliver solid inventory outcomes.
Q: Is 70% forecast accuracy bad?
Not necessarily. It depends on your product category, demand volatility, and how you buffer for error. 70% accuracy on a fashion item is impressive. 70% on a staple grocery item suggests room for improvement.
Q: How often should I measure forecast accuracy?
Monthly at minimum. Weekly during critical periods. Track trends over time, not just point-in-time snapshots.
Q: Should I measure accuracy by units or dollars?
Units, at the level you make ordering decisions. Dollars can mask SKU-level issues and aren't as useful for operational planning.
Q: How do I improve forecast accuracy?
Start with data quality, focus on high-impact SKUs, incorporate demand drivers you're currently ignoring, and review forecasts more frequently. The biggest gains usually come from fundamentals, not sophisticated techniques.