Customer Support Forecasting: Seeing the Future

Support Ops

How to develop a forecasting model for your customer support needs.

“It’s a great problem to have,” says everyone who hears about the rapid growth of a successful company. However, it doesn’t always seem so great when the DMs and live chats are piling up, your email backlog is a week old, and your team is buried in work.

When your company is taking off, it’s easy to fall behind and get into a situation where you can’t catch up no matter how hard you try. When this happens, “great” is probably not a word you’d use to describe your situation, and it’s likely not a word your customers would use to describe your service. 

When you’re rushing, quality inevitably suffers, and when you’re understaffed, wait times can skyrocket. This ends up costing your business a lot of money when customers abandon, refrain from purchasing, cancel their service, or spread negative word of mouth that steers other potential customers away.

In times like these, you may feel guilty working on strategy when there’s so much that needs immediate attention. But a little long-term planning is exactly what it takes to finally get out of the weeds.

Taking a careful look at the supply and demand of your operations and developing a forecasting model will help you plan your hiring and scheduling in a way that ensures a more manageable workload for your team and a better experience for your customers. 

Like some sort of magical customer support crystal ball, a forecast can reveal what to expect in the future. And when you can see ahead, you can get ahead.

Forecasting can become quite complex and takes many different variables into account. As team size grows and the stakes get higher, you might choose to hire an analyst to oversee forecasting, along with your other data activities. You might also choose to implement specialized software. For small to medium-sized teams, however, the following model is a great start.

History Repeats

Start with historical data. More data will typically give you better accuracy and a better picture on variability, so at least a few month’s worth is optimal. But, you can start with less and refine your model over time as you collect more.

At the minimum, you’ll need some idea of the number of contacts and the average handling time per contact. For email and written contacts, be careful to account for each exchange rather than using a “resolved” metric that only shows the number of closed contacts. This can hide time spent on the back-and-forth leading up to the final resolution. Instead, look at something like total outbound interactions. With phone, be careful not to undercount by ignoring abandoned phone calls that may have been picked up if you had been operating with shorter call waiting times.

Setting up a Basic Customer Support Forecasting Model

Follow along with all the formulas here.

1. Collect the number of contacts from several consecutive time periods. For the period, you can use month, week, day, or hour, depending on how granular you want to get. This is where having more data can be especially useful, as you’ll start to see trends around seasons or other differences like monthly billing cycles. Be on the lookout for outliers that can throw off your averages; holidays, feature releases, or other unusual occurrences should be considered separately.

2. Calculate a ratio to help you envision how the volume of contacts will increase as your business continues to grow. Divide the number of contacts by another number that’s relatively predictable for you, like number of users or shipments. Do this over several time periods to find an average growth ratio.

3. Predict future customer contacts. Let’s say you choose to use contacts to shipments for your growth ratio in step #2. Map out the number of shipments you’re expecting to send over the next few months and multiply it by the ratio to predict customer contacts.

4. Multiply the customer contacts by your average handling time per contact. This is the total handling time you’ll spend working through all the contacts.

5. Convert the total handling time into total handling hours.

6. Find how many people you’ll need on the schedule by dividing the total handling hours by the number of hours an employee works in the time period. For this example, we’re forecasting by month, so we’re dividing by 168, the average number of monthly hours worked by a full-time employee.

7. Account for shrinkage. Your team can’t operate at 100% productivity all of the time. There are human factors like bathroom breaks, vacation and sick days, and a few minutes here and there of watching the latest TikTok dance trend. There’s also off-call related work like meetings, trainings, and filing bug reports. This is called shrinkage, and it usually ranges between 20–35%. To factor this into your forecast, divide the number of employees needed by one minus the shrinkage percentage. In this example, you need 60 employees without shrinkage, but with a shrinkage factor of 20%, you’ll need 60/(1-.2) = 76.

Take your Forecasting Model a Few Steps Further

New hires won’t have the same handling time as your average employee. Hire in advance based on how long it takes for someone to be fully trained and at the same level as a tenured employee. For example, if your typical ramp up period for new employees is one month, hire a month ahead of when they’re needed.

Customers don’t contact at a steady rate. This isn’t covered in the example above, but will come into play once you start forecasting with smaller time periods. Say you get 1000 calls from 9:00–5:00 PM. It’s highly unlikely that you get a nice and even 125 calls per hour. Therefore, you’ll need to have a few more people on your 9–5 shift than your average might suggest.

Very large teams rely on complex mathematical models like Erlang C to help them make staffing decisions based on their service goals. For typical teams, simply staffing closer to what you need for your peak instead of what you need for your average will keep your call waiting times under control.

This is especially important to take into consideration if your team experiences a wide variance between your average time and busiest time, which can often happen if your product has a lot of downtime, if you do a lot of marketing promotions, or if you have other activities which create big peaks and valleys in your volume.

Team schedules don’t have to be one-size-fits-all. As your forecast becomes more sophisticated, you can experiment with different shifts. It could be more efficient, and potentially more desirable for your team, to work nontraditional shifts, like four 10 hour days, three 10 hour days and two five-hour days, or early morning and late night hours. Days off don’t have to be consecutive, and not everyone has to start at the same time. If you’re open to hiring part-time employees, you can try building this into your forecast as well. Part-time shifts can add tons of flexibility and ability to meet a fluctuating workload.

Closing thoughts

After you start using a forecast, compare your predictions to what actually happens so you can assess your accuracy and improve your model. Data visualization can be especially useful here to highlight variation and pinpoint large differences.

Just like any fortune teller, a forecast doesn’t decide your fate. It does, however, equip you with the insights you need to take control of your team’s future. Knowing how much work there should be and how many people you’ll need to handle it will help reveal a path to brighter days ahead for your team and customers.

Do you like our content?

Subscribe to our newsletter to get our best posts in your inbox!
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Keep reading...

Insights on building top-tier support teams and delivering great customer service