New Possibilities for Workforce Algorithms according to Ric Kosiba of Interactive Intelligence.
Every year for the Society of Workforce Planning Professionals (SWPP) conference, I help design the Interactive Intelligence Interaction Decisions™ T-shirt. Sounds sort of weird, but I really love doing this. My favourite T-shirt so far said, “Plan for the Possibilities,” with a superstructure of a bridge.
For those of you who’ve never heard of Interaction Decisions™, it’s a strategic planning application for long-term analysis and fine-tuning of staffing requirements for contact centres.
Let’s chat about “the possibilities” today in terms of algorithms. Contact centre workforce management, planning and forecasting is a discipline built on the back of very sophisticated algorithms. These algorithms can do a lot, such as:
• read historical volumes
• apply different forecasting techniques
• present the best forecast
• estimate how many agents are needed to hit service goals
• simulate multi-site and multi-skilled workgroups
• schedule efficiently
• develop hiring plans and overtime plans
In the past, there’s been significant conflict between building algorithms that were accurate and optimal, and building algorithms that were fast. You could get accurate, but not fast and vice versa. Algorithm developers have worked on a variety of techniques made possible by each technology advancement to make algorithms produce good results in a timely fashion.
Erlang-C Equation: One early attempt, circa 1917, was the old Erlang-C equation. Its purpose was to calculate a quick approximation of the number of staff required to hit service standards. As contact centres became more complicated, the Erlang approximation became less and less accurate. Surprisingly, many systems still use this fairly inaccurate approximation.
(Check out our white paper, Contact Centre Planning Calculations and Methodologies: A Comparison of Erlang-C and Simulation Modelling).
Processing Power: Around 15 years ago, there was a great improvement in processing power. With the advent of multi-core processors, algorithms could be coded to allow different processors to handle different parts of problems separately. For problems that allowed a separation of its algorithms, math modelers were able to produce better solutions to complex problems in reasonable timeframes.
The Cloud: We’ve all heard the positives of moving a business to the cloud. I believe for analytic systems it’s something to get very excited about. The advent of the cloud has resulted in an even more impressive leap in processing capability. Through technologies like Amazon Web Services, software that’s been designed using a cloud architecture can access not just the four or eight computing cores available to modern desktop computers, but can turn on new processing cores as needed, and then turn them off again when the math model is finished. If constructed properly, algorithms can solve incredibly complicated or CPU-intensive problems very quickly.
But there is a rub. In order to take advantage of the new possibilities available through the cloud, you have to rewrite the algorithms in modern computer languages. The initial investment can be pretty large. I expect older workforce management systems will be too hard to upgrade and companies will choose to skip the investment.
My prediction is this: we’ll see a mix of both, making it hard for buyers to discern the best choice for them. We’ll end up with some systems in the cloud that have very accurate models, fantastically efficient schedules and hiring plans. And for those built with old technologies, the systems will be inefficient and inaccurate schedules and plans. One easy question for buyers to ask during the buying cycle is, “When were your algorithms written and coded?”