Forecasting call volume & accuracy

[Canada]

I'm a Director for a relatively large call center in Canada and I would like to know how other call centers measure their forecasting accuracy ? What are your objectives and do you meet them ?

We receive between 7,000 and 10,000 calls daily and are open 18 hours a day. We measure accuracy for each interval (30 minutes), day, week, month, year. Our target is a 95% accuracy rate on a daily basis but we are far from it.

Is it realistic to reach 95%+ accuracy almost every day ? Am I being too agressive ? What should I do ?

Thanks for your help.

John
Montreal, Canada

[JoeB]

Before I attempt to answer, IÂ’d like to make sure I understand your question. You want to forecast the daily call volume for inbound calls. You measure call volume at 30 minutes intervals and it is way off the forecast. What method do you use to forecast the call volume or is it based on historical averages?

Joe

[Canada]

Hi JoeB,

we use historical averages to forecast our volumes. We forecast one month at a time since we provide CSRs with 4 weeks of schedule per production.
More precisely, here's what we take into account in our forecasting:
- Same month of the previous year
- % difference between same month last year and the month before
- "Special" days for our industry (banking) like paydays, first day and first monday of the month, last day of the month, days when we ship statements, hollidays, ...
- We look at the last 4 months to see if we can see any new paterns in the calls distribution (% of daily calls per interval)
- Marketing campaigns

Of course, we use a workforce management software to help us do all that.

I hope this will help you out and I look forward to your response.

Thank you

[JoeB]

Hi Canada

I suspect that there is no (significant) seasonal effect in your industry, and that customer behavior changes from day to day due to events you mentioned (pay day, statement inquiries, etc.) If this is correct, then same month data may not be the right reference. Moreover, there must be a significant difference between “normal” days and “event” days to render monthly (or weekly) average inaccurate.

To start understanding the call volume, I suggest running a long-tem moving average of “normal” days and a separate one for “event” days. You want to see if this average is fairly stable and if you can detect any long term (12 months) trend such as increase or decrease in call volume (normalized to accommodate for changes in number of customers, if necessary). This will also detect seasonal effects, if any.

If the data is still very “noisy” you’d need to look at a day to day behavior even for “normal” days (e.g. do customers call more often on the first Monday after receiving their statements, having the weekend to balance their checkbooks?)

Hopefully these simple analyses will start unavailing some of the drivers for changes in the call volume, which should be a good start to develop a forecasting model. Failing the simple method, you could try a more intensive analysis of variance and attempt to find a model that can be verified through regression testing etc., or even use AI techniques to develop a model, but I doubt that the later will be cost-justifiableÂ….

Hope this helps a bit. IÂ’d be curious to know what you have discovered and please feel free to contact me with more questions.

Joe

[Canada]

Thank you very much Joe. I'll ask my team to give it a try. As for the objectives, is it realistic to aim for 95% precision on a daily basis ?

Thanks again and have a great week-end.

John

[JoeB]

John,

I am afraid I am not ready to risk my reputation on that just yet... The accuracy of the forecasting will depend on the stability of the call volume, our understanding of the drivers, and the level of noise in the data, and the sophistication of the forecasting tools (e.g. historical statistics vs. behavioral modeling). My gut feel is that the effort to reach such high accuracy on a daily basis is going to be difficult and perhaps not cost-effective, so devising an alternative plan might be appropriate. A contingency plan might be based on less aggressive forecasting supplemented with techniques to respond to peaks in demands.

BTW, feel free to email me at jbarkai@DiagnosticStrategies.com if you do not wish to discuss details in a public forum.

Have a great holiday weekend

Joe

[Jbartolucci]

I agree with JoeB that you may not be ready to define an accuracy goal. The most important thing for you to do is compile at least 12 months of reliable historical data to begin analyzing.

As part of this analysis phase, defining the transactions and/or events that trigger call volume to your center will also be key. I would also try analyzing your system database/CRM coding to help understand the transactions with more granularity. This may help identify correlations in data sets that can be very useful for forecasting calls. Ultimately, you can develop usage rates/calling patterns by activity. For example, 100k customers received statements on the 30th of the month. Call disposition coding shows that 15% of the customers call with statement related questions. You will also be able to determine the behavioral calling patters for these transactions when enough data is collected.

Historical call volume analysis is very important. Moving average trending will help set the basis for projections. However, historical volume is not the only piece of the puzzle. It's a great tool if your business is static - realizes a constant growth rate. A solid understanding of your business growth/marketing in terms of member base and usage rates will help produce the greatest accuracy.

Hope this helps.

Joe

[Karsten Fuhrmann]

Hi all,

in my opinion, it's more important to forecast Workload than calls only. Workload is equal to calls offered x average handling time. That's the driver of needed resources as well as performance.

You should give it a try ...

Karsten

[JoeB]

Hi John,

I was curious to know if you have had a chance to look at long term data by averaging it as we discussed a couple of weeks ago, and what were the results.

Regards,

Joe

[tetron]

Canada

I run a call center that is currently taking 25000-30000 calls a day. Our standard is an 80% service level (80% of calls answered in 30 sec or less). I think the more research that you do you will see that 80% overall is the industry standard.

Now obviously I dont know what kind of calls you are taking, the overall importance of each different call, the length of the call, etc, but I think that 80% is the bench mark overall.


hope this helps!!

[Bender]

What workforce application do you use??

[Joxter]

Our center handles around 15-20k calls a day and using 2years of historical data I have managed to maintain an accruacy rate of 96.86 percent every month for the last 18 months (since I took over as the forecaster) so the goal is achievable but it depends highly on the stability of your customer base and a good amount of historical data to rely on. My target goals are set by upper management and they give a bonus if I can keep the error rate below 3.5% (haven't missed a bonus yet :) )

The formulas I use I designed myself and are in an excel spreadsheet, basically an exponential calculation where same day last week is more weighted than same day 8 weeks ago. The forecast is also updated daily so that trends can be identified quickly to allow schedule changes to handle excess volume or offer time off to agents when volume drops.

Our center has an ASA of 7sec. or less with LT10 of 85% or higher and we have not missed that goal in over 16 months.

Karsten is right on the mark about making sure the Average Handle Time is included in all Agent calculations. (Calls Offered) * (Handle Time) = (Time to be filled by agents).

I would post the formulas I developed but due to legal hassles I cannot :(

-Joxter-

[karlbermel]

John

I was unable to send you an email directly as per your profile but I would be interested in talking to you.

Are your calls all coming into the same location?

What mechanisms are you using to route the calls with the contact center?

Karl.bermel@sympatico.ca