Call centre forecasting tools

This graph uses the yearly contact volumes noted in the following table. Instead, the contact centre should look at the trend across previous years and find the average percentage of yearly calls that are expected each month. Then, divide this percentage by and times that figure by the overall contact volume for the year, as found earlier.

By this, we mean that there will be anomalies, where a lot more contacts entered the contact centre than would have been expected. So, when referring back to historical data, look out for any great spikes in call volumes that cannot be accounted for by seasonality. When one of these spikes is identified, it is best practice to remove the data from any future forecasting calculations.

From the most basic methods of forecasting to the most sophisticated and the most likely to give the most accurate results. While triple exponential smoothing has been used in contact centres since the s, the other methods highlighted below have really grown in prominence over the past decade and offer the most precise forecasts. This method can be used through spreadsheets and it involves splitting data into three components: Level, Trend and Seasonality see above.

AutoRegression — The ability to compare data with past patterns ii. Integrated — The ability to compare or difference the current observation from the previous observation iii. Moving average — The ability to smooth out data over a number of past periods. Neural networks have the ability to inspect a great number of calls and try to match the next item of data to the forecast. To find out more about each of these techniques, and how to use them, read our article: The Latest Techniques for Call Centre Forecasting.

Now that a precise forecast has been developed, we shift our attention to the supply side of the equation — the advisors. In an ideal world, there would always be enough appropriately skilled advisors to handle each call as it arrived to satisfy service level, customer satisfaction, cost and revenue objectives.

However, as we all know, the contact centre is rarely a perfect world. To state the obvious, every advisor is different. They have different skills, competencies, proficiencies and schedule preferences. To find the correct balance between schedules and forecasts, it is important to match the right advisors to the projected work based on these factors. Creating the initial schedule is the easy part, and the Erlang Calculator and commercial workforce management forecasting products do it well.

Powerful process automation tools can make this job much easier. For those advisors calling in sick, these tools can also identify whether or not that advisor had previously requested time off for that day that had been rejected.

Advanced WFM systems will also provide immediate feedback regarding impact to service levels, workload fit the relative balance between supply and demand and advisor costs. Use recurring and non-recurring events to create a precision forecast. Forecasting best practices suggest that more accurate forecasts enable more effective scheduling. Be prepared for frequent changes to your staffing plan. Streamline the schedule-change process to reduce administrative overhead and associated costs!

Consider your overstaffing and understaffing impacts independently. Have a keen understanding of what factors drive the business. They both examine Level, Trend and Seasonality but use different ways of analysing, training and modelling the outputs. One is called the Jackdaw Algorithm and the other is called the Kestrel Algorithm. We have named the algorithms after birds that are clever, observant and good at spotting small details.

In the current phase we require 2 years of data to generate a forecast. We are looking at a new algorithm that can generate a forecast with less data. These experimental algorithms have been tested on a range of data sources, but we are not sure that they will work in all circumstances. Please could you rate the accuracy of the forecast at the bottom of the results page. If you have any problems, please email us through the newsdesk newsdesk callcentrehelper.

A CSV file is a common way of exchanging data between programs such as Excel and other computer systems. The data column needs to be whole numbers with no decimal places or commas. Care needs to be taken that a large number such as 33, is entered without the commas, e. You can simply cut and paste values from 2 columns in an Excel worksheet and put them into the date box and press submit. After a short delay, you should see the outputs on the second screen.

This will produce a graph that looks like this. A neural network is a network that tries to model the neurons or brain cells in the human brain.

For example, they will scan in a series of numbers of calls and try to match the next item of data to the forecast. It looks like neural networks could have a lot of potential advantages for contact centre forecasting.

Some of the most exciting factors for neural networks could be in automatically isolating special days from the forecasts. The idea is that if I have a forecasting problem, I use a neural network, no matter the specific challenges of the problem, and it will help solve the issues. If you do not have multiple years of data available, take a look at our article on How to Forecast With Minimal Data.

The key to generating neural networks seems to be in how many nodes in essence, how much memory the network has. In theory more nodes should generate better results but much slower performance. A contact centre time series looks quite complex to me, but not in terms of mathematics.

In most contact centre applications a small number of nodes should be enough. This takes you through how the logic works along with a simple worked example:. The very latest thinking in call centre forecasting is Multiple Temporal Aggregation.

This is a method to combine both high-frequency data hourly daily, weekly with longer-term trends over time. You have completely removed the seasonality. In essence, this averages out the contacts and special events across the year. You will never be able to extract from a single viewpoint everything, but if you pull together all of the aggregation from different aggregation levels, then you have a holistic view.

The advantage with Multiple Temporal Aggregation is that you can focus both on the intraday and the longer-term data at the same time. What you do sounds a bit strange at the beginning and then it makes sense. So one will be one observation, at the other end it will be 8, observations. You can carry over information from the top level to the bottom level and vice versa.

To help with the understanding of how Multiple Temporal Aggregation works a software model has been produced in the statistical modelling package R, called MAPA — Multiple Aggregation Prediction Algorithm, which produces some promising forecasts. You want to consider:. You will also need to clean the data. If call volumes spiked in a given month because of a new product launch, but this trend is not repeated, you will need to remove this outlier to ensure an accurate forecast.

For new contact centres, you might not have a lot of historical data you can use to make forecasts. Or you may be missing this data if using a more basic technology solution. Bumping up your average handle time for example can result in a massive change in the number of FTEs forecast.

You might actually already know your future call volumes. For example, you might receive a consistent number of new records each month, and know that every call takes 2 minutes on average, excluding agent break time. In this case, you can skip to forecasting staffing requirements.

For example, if you mostly call inbound leads from your website. In this case, you can use the method below that inbound call centres use to forecast call volumes. The first thing you want to do is grab your historical data, and calculate period-by-period growth rates in call volumes. This helps to avoid some of the major problems with using simple averages. For example, in this contact centre, call volumes are growing each period.

If we take a simple average, the growth is not accounted for, and the forecast is likely too low. If, on the other hand, we find the annual growth rate, we can then find average growth by period, and use this to predict growth for Cell P7 is an arithmetic average of cells P4:P6, which is then used as a predicted growth rate for each month.

To solve the first problem with the aforementioned method, we can use a technique known as exponential smoothing. Exponential smoothing is a forecasting method that assigns increasing weightings to data in the recent past when compared to the distant past. This helps to calculate a more relevant average based on more recent data.