Forecasting is something used widely in contact centres, and it’s something I’m surprised not to see more of in the SaaS world, although I have come across examples of people doing this in SaaS.
Contact centres use forecasts to set their budgets for the year, alongside using them to plan staff schedules and plan for peaks. In the contact centre industry forecasting is a role in itself and forecast accuracy is what you strive to achieve.
Contact centres generally have a team called Resource Planning or similar. This team is responsible over creating these forecasts, scheduling all of the staff to meet the demand and managing the incoming work in real-time to ensure targets are met. I’ll talk about each aspect of that work at some point but for now we’ll focus on the forecast.
The purpose of your forecast will be to set your staffing levels and be able to budget for them, if you have someone that manages your queues in real time it also gives them an indicator of how busy they’re likely to be so they can plan for the day in advance. Personally, I’ve only ever gotten a forecast within about 8% accuracy at a daily level, but there are a lot of methodologies and the more you work at it the more robust the forecast will be.
Although throughout this post I’ll be referring to calls these processes can be used for tickets and chat as well. Ideally you will need three years worth of data on your volumes, split at a daily or weekly level. In this post I’ll outline some of the methodologies available, and go into more detail on them in later posts. NB: to forecast effectively ideally you’ll have at least three years of data available.
Time Series/Seasonal
Perhaps the simplest methodology, time series is simply the analysis of contacts over set intervals of time, be them weekly, daily or monthly.
Blending seasonal variance with time series is the most effective method, showing you your average ticket volumes vs the seasonal trends (effectively what’s your average volume over the last X weeks vs what happened at the same time last year). Here is a good excel based model that uses these methods and it will be a good starting point.
Regression Model
Regression models are really cool. Essentially, they take all of the data around your call volumes and you plot various variables against them, for example, let’s say you’ve mapped out 3 years worth of call volumes and against them you plot every major weather event that’s happened and maybe sporting events too, why not?
The regression model will analyse that data to understand any trends relating to those events and the volumes, so the next time you input a similar event in the current week the volumes returned will be influenced by the model.
This is just one example and this model can be used in various ways. One time that I have used it is when I worked for a holiday company, I took three years worth of data and plotted every cruise ship departure against it to understand the impact of these departures on our volumes. (They went down, yay!) So this way I could state with relative confidence that we were going to have a quiet week due to a departure.
BI Tool
A BI tool is probably the simplest way of creating a basic forecast. My go-to is Microsoft Power BI but you could use Tableau, Yellowfin, SAP etc. There’s a free, as far as I can tell, download of MS BI here.
In writing this I realise now that I’ve moved to Mac OS I can’t use MS Power BI. Fuming. Look out for the more detailed post 🙂
Python/R
I’m not experienced in this methodology but it’s something I’ve seen growing in the planning world recently, and I’ve seen companies use it to great success.
Programmatic forecasting is likely the most robust way of producing this kind of data with more confidence, I’m about to start a course in this and I’ll let you know how that goes!
Hopefully you can see why forecasting is a useful tool. You can give your team an understanding of workload and plan for it, there are a bunch of cool scheduling models you can implement to match volatile volumes.
