© Dadann – Shutterstock – 2807618635
Forecasting is tough! Get it right and you set the stage for perfectly balancing customer demand with agent availability – but get it wrong and you could be left juggling understaffed shifts, frustrated agents, and spiralling costs.
So where are some of the biggest trip points? We asked our WFM panel to spill the beans on the forecasting lessons they learnt the hard way – so you can avoid making the same mistakes.
“Seasonal Events Don’t Always Create Similar Patterns in Customer Behaviour That Can Be Relied Upon for Year-on-Year Comparisons”

One of the biggest lessons we’ve learnt in forecasting is that historical data doesn’t always tell the full story. A good example is forecasting demand over Easter, Eid, Christmas, and New Year. Unlike many seasonal events, these do not fall within a consistent weekly pattern.
Some years it lands closer to the weekend, while in others it falls midweek. This significantly changes how customers behave and when they contact us.
We have underestimated this demand in the past because there is no truly consistent historical pattern to compare against. Customer behaviour also evolves over time, meaning that even when these events fall on a similar date pattern, previous years are not always reliable indicators of future demand.
What we learnt was that forecasting isn’t just about finding the closest historical comparison. It requires understanding the drivers behind the numbers and challenging assumptions.
We now place much greater emphasis on analysing customer behaviour trends, calendar placement, public holiday patterns, and broader business changes – rather than relying solely on historic volumes.
Another lesson was the importance of regularly reviewing forecast accuracy and understanding where variances occur. Rather than treating forecasting as a one-off exercise, we continuously refine assumptions as new information becomes available.
Contributed by: Tim Milburn, Global Workforce Planning Manager at Emovis
For information on successfully navigating seasonal peaks, read our article: How to Thrive Through Seasonal Peaks
“A Model Can Win on Average Error by Spending Its Accuracy on Intervals That Don’t Matter”
The forecast I trusted most was the one that hurt me most. For a long time, I reported forecast accuracy as one number: MAPE, averaged across every interval in the month. It looked great. And we still had weeks where service level fell apart on the busiest afternoons.
The average was hiding it! A Tuesday 2pm peak and a Thursday 4am trough carry the same weight in that number.
Miss the peak by 10% and the day’s service level is gone. Miss the trough by 20% and you pay for a little idle time nobody notices. A model can win on average error by spending its accuracy on intervals that don’t matter.
“When People Abandon and Ring Back, Every Redial Inflates the History That Becomes Next Year’s Baseline”
The second lesson came out of the same bad weeks. When service drops, people abandon and ring back, and every redial lands in the arrival count looking like a new customer.
At 10% abandonment, with 7 in 10 abandoners trying again (it varies a lot by queue), history runs about 8% heavy. That inflated history becomes next year’s baseline. Then the week you fix service, “demand” drops and it gets logged as a forecasting miss.
What I do now:
- Grade the peak intervals separately from the daily average. If the daily MAPE is fine and the peaks missed, the peaks are the story.
- Deduplicate arrivals by customer inside a retry window before anything goes into history.
- Once a month, ask whoever owns the forecast to explain the biggest miss without opening the tool. If they can’t, the forecast is being produced, not reviewed
A forecast is only as accurate as the thing you chose to grade it on.
Contributed by: Carlos Cordero, The WFM Architect
“You Need to Invite Others to Be Participants in the Process – Not Just Consumers of the Forecast”

A lot of WFM teams get themselves in trouble when they turn into a black box themselves, or their processes turn into a black box. When Operations simply receives the output, you’re not building trust or relationships; you’re building a consumer.
From my perspective, WFM should be working collaboratively with Operations, Marketing, Finance, and any other team that might be receiving from the forecasting process and/or providing input into it.
The best forecasting that I have done has typically come from my ability to hear what other teams want or could be doing, understand how that might impact my forecast, and then expose the variables that I see and can change, but that they can also see.
Now, they don’t necessarily have to be able to change those variables themselves. But being able to influence them and having me articulate why I’m doing what I’m doing helps build that relationship. It moves them from simply being a consumer of the forecast to being a participant in the forecasting process.
“No One Really Cares About Forecast Accuracy”
This one might be a little harder for WFM teams to get their heads around, but no one really cares about forecast accuracy. What they care about is how good we are at predicting the future. Which sounds like the same thing, but I think they’re different.
Forecast accuracy is a number. It’s a score. Think about it this way. No one really cares about the grades you got in college, but they do care about your ability to think, to collaborate, and to communicate.
Forecast accuracy is the grade. I think what happens is that WFM holds themselves to a particular grade. They report that score out, and in the end, operations just sees high = good, low = bad. And when the score is too low, they ask WFM what they’re going to do to improve it.
But what Operations really cares about is whether they feel WFM is producing reliably accurate numbers. Can they trust what WFM is telling them? Forecast accuracy is certainly an important measure, but I think WFM sometimes sees it as the end, when it’s really just a signal toward the end. And that end is trust.
So as you’re working to improve your forecast accuracy, make sure you’re translating that work in a way that helps Operations see they can rely on you, trust your numbers, and ultimately make better decisions.
Contributed by: Dan Smitley, Founder of 2:Three Consulting
“You Should Treat a Change in Client Mix as a Forecast Event in Its Own Right – Not Just a Membership Update”
In health insurance, our volume forecast was built on two drivers: forecasted membership multiplied by forecasted utilization, with utilization calculated as actual cases divided by actual membership. On paper it was clean and easy to explain.
The problem was that we calculated utilization at regional level, and one of our regions was made up of international organizations whose members used their cover in very different ways. A blended utilization rate hid all of that.
As long as the client mix stayed stable, the forecast held. The moment one organization grew its membership, or a new client was added, the blended rate stopped reflecting reality.
A thousand new members from a low-utilization client and a thousand from a high-utilization one looked identical in the model, but they generated very different volumes for the team. We knew forecasting by client would solve it, but with thousands of clients, it kept being postponed as too big a job.
The lesson I took from it is that an average is only as reliable as the mix behind it. I started building a client-level view before I moved on, and the approach I’d recommend doesn’t require forecasting every client individually.
Identify the small number of clients that drive most of the membership, or whose utilization sits far from the average, forecast those separately, and keep the long tail as one pool.
Before a new client goes live, ask for their expected utilization or use a comparable client as a proxy, rather than letting them inherit the regional rate. Most importantly, treat a change in client mix as a forecast event in its own right, not just a membership update.
Contributed by: Luísa Gonçalves, Founder, WFM Intelligence
“Sometimes the Outlier We Remove Is Actually the First Evidence of a Change We Need to Understand”
A forecast can be mathematically sound and still be completely wrong if the assumptions behind it are wrong. Historical demand tells us what happened; it doesn’t automatically tell us why it happened or whether it will happen again.
That changed the questions I ask. Instead of simply asking “What was the variance?”, I want planners to ask three things:
- What happened?
- Why did it happen?
- Is it likely to happen again?
A 20% increase could be genuine growth. It could also be a marketing campaign, a billing problem, an outage, repeat contacts, or customers chasing work we haven’t completed. Those scenarios may look identical in a spreadsheet but should lead to completely different forecasting decisions.
That is also why I’m cautious about automatically removing outliers. Unusual doesn’t mean irrelevant. Sometimes the outlier we remove is actually the first evidence of a change we need to understand.
The lesson I’ve learnt is simple – understand before you predict! The best forecasting teams I’ve worked with aren’t necessarily those with the most complicated models. They are curious. They build relationships across the business, challenge assumptions, and can explain the story behind their forecast.
Contributed by: Ben Heslop, Director of Planning and Data at ResQ and Founder of Smooth Planning
“Flexible Capacity in the Model Is Not Always as Flexible in Practice – Especially When Some Colleagues Haven’t Worked Certain Queues for a While”
One of the hardest lessons I’ve learnt is that a forecast can be mathematically right and still be operationally wrong.
I remember building a forecast for a contact centre where the numbers looked solid. We had used historical demand, expected growth and average handling time, then matched that against our multi-skilled colleagues. The model balanced, the plan was agreed and, on paper, we had enough people in the right places.
Then the week arrived, and reality had other ideas.
Contacts did not land neatly across the queues. Handling time rose as customers needed more support, and the flexible capacity in the model was not as flexible in practice. Some colleagues had not worked certain queues for a while, others needed support, and moving people created gaps somewhere else.
The calculation was not the problem. I had treated assumptions as facts and skills as switches that could be turned on whenever we needed them.
Since then, I have changed how I forecast. I spend more time speaking with the people closest to the work, understanding what has changed, and testing whether the flexibility shown in the model exists. I make assumptions visible, build realistic scenarios, and agree the early-warning signs that tell us when the plan needs to change.
Contributed by: Gavin Rogerson, Senior Global Workforce Planning Consultant, Xplor Technologies
★★★★★
What Lessons Have You Learnt the Hard Way With Your Forecasts?
If you have any stories or tips to share about the forecasting lessons you’ve learnt the hard way, we’d love for you to join our LinkedIn community discussion on forecasting and tell us more about your experience so far.
If you are looking for more information to improve forecasting in your contact centre, read these articles next:
Reviewed by: Xander Freeman


