You ever notice how the best predictions don't come from fancy math, but from paying attention to what's actually happening right in front of you? Most forecasting systems drown in spreadsheets and forget the obvious. That's the gap focus forecasting walks into.
Quick note before moving on Worth keeping that in mind..
Here's the thing — focus forecasting is based on the principle that the most relevant information for predicting demand is already sitting in your recent sales and operational data, not in some distant economic model. In practice, it sounds almost too simple. But in practice, that simplicity is what makes it work Simple as that..
What Is Focus Forecasting
Focus forecasting is a method that builds forecasts by testing a small set of forecasting rules against recent actual data, then picking the rule that would have performed best. That's it. In real terms, no crystal ball. No PhD required.
The short version is: you keep a handful of simple forecasting techniques in your back pocket. Think about it: every forecasting cycle, you ask, "Which one of these would have guessed closest last time? " Then you use that one to project forward.
It was popularized by Bernard Smith back in the 1970s, and the name says it all — it keeps your focus on what the data is telling you instead of on the model itself Surprisingly effective..
It's Not One Model, It's a System of Choices
A lot of people hear "forecasting" and assume there's a single equation doing the heavy lifting. Because of that, with focus forecasting, there isn't. You might have five or six candidate rules: a straight moving average, a simple trend line, last period's number repeated, something seasonal. Each is dumb on its own. Together, they cover enough ground.
The Selection Step Is the Engine
What makes it different is the feedback loop. You simulate each rule on the most recent periods you already have answers for. The one with the smallest error gets promoted to make the next call. That's the whole trick It's one of those things that adds up..
Why It Matters / Why People Care
Why does this matter? In practice, because most companies either over-forecast with complex systems or under-forecast by gut feel. Both cost money.
When a retailer uses a giant statistical suite and ignores that a simple three-month average would have been more accurate last quarter, they end up with dead stock. Or stockouts. Either way, the customer loses and so does the margin.
Turns out, focus forecasting matters most in messy real-world settings — places where demand jumps around, where new products show up, where promotions wreck your baseline. In those cases, a rigid model lags. A focused one adapts Simple as that..
I know it sounds simple — but it's easy to miss how much money is quietly lost to forecasting pride. Honestly, this is the part most guides get wrong: they sell complexity when the win is in the loop.
How It Works (or How to Do It)
Let's get into the meat. Setting up focus forecasting isn't hard, but you do need discipline. Here's how it actually goes Easy to understand, harder to ignore..
Step 1: Pick Your Candidate Rules
Start with a small library. Four to six is plenty. Examples:
- Naive: next period = last period
- 3-month moving average
- 6-month moving average
- Linear trend on last 6 months
- Seasonal naive (same month last year)
You don't need them to be smart. You need them to be different.
Step 2: Define Your Error Window
Choose how many past periods you'll score them on. Still, usually 3 to 6 months of real history. This is your test track.
Step 3: Run the Back-Test Every Cycle
For each candidate, calculate what it would have predicted for those past periods. Compare to what really happened. Measure error — MAPE is common, but even simple absolute difference works.
Step 4: Select the Winner
The rule with the lowest error on the window becomes your forecast for the next period. That's your focus pick And that's really what it comes down to..
Step 5: Repeat, Always
Next month, roll the window forward. Worth adding: re-test. The winner might change. Consider this: often it does. That's the point — you're always using what's working now, not what worked in 2019.
A Quick Example
Say you sell garden hoses. Consider this: in March, you test your rules on Oct–Feb actuals. The 3-month average wins. You forecast April with it. In April, you roll to Nov–Mar, and the naive rule wins because demand went flat. So you switch. On top of that, no meeting. No model rebuild. Just the loop Small thing, real impact..
Common Mistakes / What Most People Get Wrong
This is where you can tell who's actually run one of these versus who read a slide.
Mistake 1: Too many rules. People add twenty techniques thinking more is safer. It isn't. It dilutes the signal and creates noise. Keep it tight Not complicated — just consistent..
Mistake 2: Wrong window length. A 12-month window in a volatile category just averages away the pattern you needed. Too short and you chase flukes. Match the window to how fast your demand actually turns.
Mistake 3: Forgetting to automate. If you're doing this in Excel by hand every month, you'll quit by quarter two. The loop has to be automatic or it dies.
Mistake 4: Mixing in opinions silently. Someone "adjusts" the pick because they feel a promo is coming. Fine — but then track that adjustment as its own rule. Otherwise you broke the system and won't know if it helped.
Look, the biggest miss is treating focus forecasting like a one-time setup. It's a habit, not a tool you install.
Practical Tips / What Actually Works
Real talk — here's what I've seen separate the teams that get value from this versus the ones who tried it once Turns out it matters..
- Start with your dumbest rule. The naive "last month repeats" is shockingly competitive. If your fancy stuff can't beat it, you don't need the fancy stuff.
- Log every selection. Keep a tiny table: month, winner, error. After a year you'll see which rules earn their keep in which seasons.
- Use it for the messy 20%. Focus forecasting shines on sporadic items. For stable baseline SKUs, a normal stat forecast is fine. Don't force it everywhere.
- Watch the handoffs. If sales and ops don't trust the number because "the computer picked it," show them the back-test. Seeing the rule beat the others last cycle builds buy-in fast.
- Pair with a promo flag. Add a simple rule that says "if promo, use X." Not brain science, just honest segmentation.
Worth knowing: the goal isn't perfect. It's less wrong than the alternative, consistently That's the part that actually makes a difference..
FAQ
What is the main principle behind focus forecasting? It's based on the principle that the best forecast comes from selecting, each period, the simple rule that performed best on your most recent actual data — not from one fixed complex model That's the part that actually makes a difference..
Is focus forecasting only for inventory? No. It started in demand planning, but any repeating decision with history — staffing, web traffic, cash flow — can use the same loop That's the part that actually makes a difference..
How many forecasting rules should I use? Four to six is the sweet spot. Enough to cover different shapes of demand, few enough to stay readable Easy to understand, harder to ignore..
Do I need software to run it? Not strictly. A disciplined Excel setup works. But anything past 50 items and you'll want it automated in your planning system.
Why not just use the most accurate model overall? Because "overall" hides the fact that different rules win in different conditions. Focus forecasting lets the data tell you which condition you're in right now Worth knowing..
The weird part is, once you run focus forecasting for a while, you stop trusting forecasts and start trusting the process. And that's healthier. The principle — that recent reality should pick your method — keeps you honest when the world shifts. You don't argue with the loop. You just let it point, and you move.
