Following the introduction of our Erlang Calculator a couple of weeks ago, Dave Appleby explores the future of Erlang as a method for workforce forecasting.
I like to think I know my way round forecasting, I know the myth that is a marketing forecast, I know Erlang (well not personally as he’s dead), I know my teams, I know that for us WFM is not applicable, I know our forecasts are good. What I don’t know is if we are doing it the right way rather than just the standard way. There is a lot of work being done by people who are a lot cleverer than me at MIT and the University of Hashing that I think will revolutionise the way we forecast in the next 10 years.
Hands up who has heard of FPP or Fractal Point Processes? I’m about to use lots of buzzwords now, Fractal, Chaos, Self-Similar, Stochastic, Poisson, Superposition, Queuing theory, FMPPP (Fractal-Modulated Poisson Point Process), FBNDP (Fractal-Binomial Noise Driven Process) and traffic engineering. Whoever is playing office bingo may shout House! now.
Ok, buzz words out of the way. Nothing is new in the basic structure of how a queue works. The research at the moment is looking at the huge amounts of traffic across international switches and doesn’t apply, yet, to the smaller volumes we get in our centres. However, it is something to be watched. The theory is that unlike a traditional Erlang model, which applies a steady-state transformation to the data, a much more robust fractal model can be applied, which CONSTANTLY remodels the data, so intra-day becomes automatic, and ‘smarter’. [Trust me I’m going somewhere with this].
Now I know and can understand the maths behind the Erlang calcs. I’m even working on a way to break it down to ‘Math for Managers’. [Watch this space, but don’t hold your breath.] I know Mandelbrot, Julia and Lorenz’s work, hell I even understand about 40% of it, but now we all need to start looking to the next new thing. In our case it WILL be VOIP, we don’t have any choice, copper to glass, analogue to digital. We WILL be going there. The infrastructure is already in place in what is the biggest network in the world; why would telecoms companies continue to install and upgrade voice networks when they can bang the call across the spare bandwidth they already have on their shiny new data network. The first stages will be transparent with the voice end still being a standard phone but the transmission over IP. Moving later to IP Phones and IP-only call centres. This isn’t a warning, more of a heads-up that these things are changing.
Which leads me back to the point above.
In a VOIP environment an Erlang forecast won’t work, call-flow across a network is chaotic, network loading needs to be balanced to allow agents to take audible clear calls not just calls. Traffic engineering will be the new call-flow planning, FMPPP and FBNDP will replace Erlang B and C as the accepted methods for forecasting. Staff planning will move from agent vs. call planning to bandwidth vs. agent planning.
We are fortunate at this point to be in at the start of something new, possibly the biggest revolution in communications since the rotary switch. The drawback is for all of us it’s back to school.
The article below gives an insight and the second and third I can’t find a weblink to at the moment.
1) Fractal Point Process and Queuing Theory and Application to Communication Networks
Gregory Wornell MIT
2) A Case For Fractal Traffic Modelling
Ashok Erramilli, Walter Willinger MIT
3) Performance Analysis of a queue under FPP Self-Similar Traffic
Chen Chun Han, Cao Mingcui, Liu Erwu, Li Feng and Luo Zhixiang.
And as an introduction to fractal modelling.
A leap of faith with Fractal Analysis by John Conover
Have fun, time to hit the books.
Dave Appleby has been working as a planner, forecaster and analyst in the contact centre industry for the last 11 years, having been a chef in a previous life. Starting off working on the phones for the launch of a Grocery Home Shopping service, he has worked for a variety of in-house and outsource operations including Disneyland Paris, Seeboard, GIftaid, GM Finance and the Daily Telegraph. A keen diver (both instructor and cave diver), Dave is currently a senior analyst for a large UK insurance company.
Do any of the major workforce management solutions on the market use fractal planning algorithms in them?
Not at the moment,
The work is in a very early stage at the moment, whilst the FMPPP (Fractal-Modulated Poisson Point Process), FBNDP (Fractal-Binomial Noise Driven Process)Processes I discussed above are looking promising, they, at the moment, only work at a network traffic forecasting level. The challenge for the mathematicians (well it makes a change from Engineers doesn’t it) is to apply the additional variables of human nature in WHEN we initiate a call (the extant Erlang) to the network traffic modelling. This is something that we should see in the next few years. Most WFM products a modified, and VERY sophisticated, Erlang / Poisson models. Exciting times (If you’re a geek like me).
Do you have any articles (or pointers?) to the suitability of the Erlang B and Erlang C models to different types of call Centres. I am working on a project where the there is naturally randomisation of calls and a high level of duplication when they occur.
Interesting discussion, but I think you are off the deep end in extrapolating Fractal Point Processes to call queuing in a call center.
FMPPP and FBNDP apply to phone networks, VOIP etc, where traditional Erlang formulae no longer cater for some of the complexities (especially where a single voice conversation is split up into packets along multiple routes shared with TCP traffic).
However once a call arrives at the call center (from whatever source, by whatever means) then normal Erlang C applies – a certain number of incoming calls, queued and presented to a certain number of agents. That has not changed with VOIP.