Forecast accuracy sits near the top of most planning scorecards.
MAPE. Weighted MAPE. Bias. Tracking signal.
Targets are set, variances reviewed, and considerable time is spent explaining what's changed. None of this is wrong. These measures have a place, and they provide a useful signal about where the business needs to look.
The problem starts when forecast accuracy becomes the objective rather than an input to better decision-making.
A forecast has no value on its own. Its value comes from the decisions it informs and the actions that follow.
It can inform purchasing, production, inventory, capacity, distribution and commercial activity. But a more accurate forecast does not, on its own, result in a better outcome.
That outcome might include stronger availability, healthier inventory, improved margin, reduced waste and more effective use of working capital.
A good forecast can still produce a poor outcome
It's possible to forecast demand accurately and still disappoint customers.
If a supplier fails to use the forecast, can't respond to it, or hasn't aligned its capacity and material planning to it, the quality of the number is largely academic. The same is true if that supplier has no mechanism to allocate available stock fairly, and instead distributes it on a first come, first served basis. The forecast may be accurate, but another customer can still secure available supply before you do.
In a different scenario, a supplier might work with two forecasts: the number agreed with its customer and a lower number used internally.
The reasons can be behavioural.
Someone measured against sales performance may prefer a lower internal forecast because it makes overperformance more achievable. However, while the agreed customer forecast may influence the supply requirement, it's the lower internal number that drives the supplier's purchasing, production and inventory decisions.
The result can include increased costs, stock-outs, lost sales and penalty charges.
Worse still, an inability to supply may suppress sales to a level more in line with the lower internal figure. If availability and lost sales are not considered, that constrained result can then become the baseline for future planning.
Collaboration is what turns a forecast into an outcome
A forecast passed from one organisation to another is not collaboration. It's a data transfer.
Collaboration requires a shared view of what sits behind the forecast and what each party is expected to do with it. That means being clear about the assumptions, the degree of confidence, upcoming events, constraints, decision points and the point at which the forecast becomes a commitment.
It also means acknowledging that the relationship works in both directions. Customers and commercial teams often hold information about promotions, listings, pricing changes, contracts or market activity. Suppliers hold information about capacity, materials, production schedules, allocation rules and emerging constraints. A better outcome depends on those signals being brought together early enough to influence a decision.
Useful conversations therefore need to extend beyond whether the forecast moved up or down, or whether it was accurate. They should explore questions such as:
- What's changed since the previous view, and why?
- Which assumptions carry the greatest risk?
- Where is supply constrained, and how will available stock be allocated?
- What flexibility exists if demand is higher or lower than expected?
- Which decisions need to be made now, and which can remain open?
- What will each party do when agreed triggers are reached?
This doesn't eliminate uncertainty. It makes the uncertainty visible and gives both parties a better chance of managing it.
Measure the forecast where a decision can still be made
Forecast accuracy needs to be measured at an offset that reflects how the business and its suppliers can respond.
A forecast measured one week before demand occurs may look strong. But if the supplier needs twelve weeks' notice, a key material requires sixteen weeks, or the minimum order quantity locks in several months of cover, the one-week measure has limited practical value.
The right offset should relate to the decision horizon. It should reflect the longest relevant lead time, the planning time fence and the point at which meaningful action can still be taken.
Different decisions are made at different points in time. As a result, different forecast horizons matter for different reasons.
Long-range forecasts often support supplier, purchasing and production decisions. Shorter-range forecasts support executional decisions such as allocation, replenishment, pricing and promotions.
Combining them into a single forecast accuracy measure can hide where planning is genuinely working well and where it is not.
The important question is not simply whether the forecast was accurate. It's whether it was sufficiently reliable when a decision had to be made.
Not every product deserves the same response
A single forecast accuracy target assumes a level of consistency that rarely exists across a product portfolio.
Some items have stable, repeatable demand. Others are seasonal, promotion-led, intermittent or inherently volatile. A poor accuracy result may reflect a weak forecasting process, but it may also be telling you that the current supply and inventory model is not suited to the demand pattern.
That should lead to a decision, not simply another conversation about improving the number.
For a highly volatile item, the right response might be to move towards make-to-order, agree a different service level, change safety stock, shorten the replenishment cycle or rationalise the range. For a seasonal product, the business may be reasonably confident in the total season but much less certain about the weekly sales profile. It may choose to secure more of the season's volume in advance and accept some inventory exposure in return for availability.
For a low-value, non-core and highly volatile item, the opposite decision may be appropriate. The business might deliberately accept a lower service level because holding enough inventory to cover every possible spike would destroy the economics of the range.
In each case, forecast accuracy remains useful. But its value lies in shaping the planning response, not in becoming the response.
When the stock is already here
The distinction becomes particularly clear with imported and seasonal products.
Long lead times and large minimum order quantities can deliver a lower unit cost, but they also increase exposure. By the time the selling season begins, much of the inventory may already have been ordered, shipped and landed.
At that point, forecast accuracy is mainly diagnostic. The inventory risk already exists. What matters now is whether sales are tracking to expectations and what the business is going to do about the gap.
The questions become more commercial:
- Do we need to adjust the price or promotional plan now?
- Can the product be bundled with complementary lines?
- Is there an alternative channel or customer segment that can absorb the stock?
- Should we slow or cancel later orders where that is still possible?
- How quickly do we need to act to avoid carrying inventory beyond the season?
A forecast accuracy score will not answer those questions. Good planning disciplines will. The business needs clear visibility, an effective review cadence, agreed decision rights and the willingness to act while there is still time to influence the outcome.
The measures have to work together
This is why forecast accuracy shouldn't be viewed in isolation.
It should sit alongside on-shelf availability, inventory levels, stock cover, turns, GMROI, supplier service, waste, obsolescence and working capital. The exact suite will vary by business, but the principle is consistent: no single measure tells the full story.
A business can report strong forecast accuracy while holding excess inventory and buffering every piece of volatility. Availability might look excellent, but the cost is tied-up cash, ageing stock and avoidable markdown risk.
Equally, a business can have more modest forecast accuracy and still deliver strong service, good turns and healthy margins because it has segmented products intelligently, collaborated well across organisational boundaries and built the right responses around uncertainty.
That's the better outcome, even if the forecast accuracy line on the scorecard is not the one attracting praise.
Use forecast accuracy to improve decisions
MAPE, weighted MAPE, bias and tracking signal all have value when they are correctly defined, understood and applied at the right level. They can expose weak assumptions, persistent bias and products, customers or categories that need attention.
They are also easily misunderstood. MAPE, for example, is an error measure, so a lower result is better. Yet many people naturally expect anything described as forecast accuracy to improve as the number rises. Definitions, aggregation and presentation matter if the measure is to support action rather than create confusion.
But even a technically perfect measure remains only one data point.
The purpose of reviewing forecast accuracy is not to reward a better percentage. It's to understand what's driving the error, what the parties involved can do differently and which decisions will improve service, inventory and commercial performance.
That requires good planning disciplines rather than allegiance to a particular framework: the right information, shared assumptions, clear accountabilities, timely decisions and a regular cadence for reviewing what's changed.
Forecast accuracy matters. But the forecast is not the end goal, and neither is the metric.
Better decisions, stronger collaboration and better business outcomes are.