On Target - Fall 2018

What Would You Do If We Could Automate Everything? 2018-11-26T00:35:28+00:00

What Would You Do If We Could Automate Everything?

By Ric Kosiba, Ph.D.  Vice President, Genesys

More AI, Automation, and the Like

There is a new resurgent theme among workforce management vendors: artificial intelligence and workforce management automation. Over the last year or so, in this column, I have spoken about AI and forecasting as well.

I believe that some of this talk is hyperbole, vendors hopping on the AI fashion train, but some of it is becoming real.  To be honest, I do not think that, for our workforce management problems, AI is the powerhouse technology. Rather, it is the introduction of cloud computing that allows us to run more sophisticated algorithms (including AI) in our workforce management systems. Pure cloud installations, or hybrid premise-systems-with-cloud-based-algorithms, will thankfully be coming to all of us.

Cloud computing is an enabler. It enables machine learning forecasting systems, like we discussed in the Spring 2018 edition of On Target.  It enables us to build schedules using true mathematical optimization, as we described in the Winter 2018 edition of On Target, with all the value add-ons we discussed. It enables us to get to the point that every one of our algorithms are customized to every contact stream, are proved accurate, and are shockingly fast. Instead of schedules taking 15 minutes to several hours to run, they can run in a few minutes. Cloud computing, and modern algorithms enable this.

The History of WFM Systems

Many of the WFM systems in use today were developed at a time when computers were slow, memory was expensive, and distributed computing was a fantasy.  Scheduling algorithms, which are computationally hard, had to be built in a way to solve in a timely manner on these slow computer systems. This generally necessitated that the algorithm developer trade off optimality for speed.  To get solve times to an hour or so, the methods developed required work arounds and an attempt to find “good schedules,” not necessarily the optimal best set of schedules.

As contact centers became more complex, with multi-site, multi-skill, and multi-channels added, these algorithms were not replaced, rather they were altered to schedule for all this additional complexity; this new, more complex scheduling problem was shoe-horned into an old solution process.  For example, outbound and email channel schedules were developed by determining staffing requirements, even though the concept of requirements is sort of absent for each of these operations.

But it is all changing with the advent of cloud computing.

Prerequisite to Automation

Workforce automation has a current buzz.  Many of the tasks of the workforce management team, like determining best training schedules, automatically detecting shortages before they happen and prompting agents for overtime, etc…  is very cool. However, there is a prerequisite to automating many of the workforce management functions, and that gets us back to our modeling discussion. Automation hinges on the ability to accurately and automatically develop forecasts, accurately and automatically determine staffing requirements, and optimally and automatically schedule and re-schedule our agents.

This, of course, implies our systems are fast, accurate, and optimal.  We won’t want a system to automatically offer overtime if the requirements calculation is off. We don’t want it to automatically approve time off if our system calculates slowly and cannot respond to changing conditions quickly. We don’t want to trigger any real-time changes if our forecasts are inaccurate.

So, to automate, we need to upgrade the building blocks of our workforce management systems.

All the WFM companies are investing in cloud-based algorithms

Actually, I’m not sure of this, but I expect so.  If automation is desired, a hybrid or cloud computing model is absolutely necessary.  We’ve done some experimentation and here are some results.

Like we discussed in On Target before, we can automate the forecasting process and use machine learning and cloud computing.  Our experiment shows that we can solve, using sophisticated methods (like different flavors of Holt-Winters), over 4 million forecasts in under two minutes.  This means that cloud computing and AI allow us to crank through almost every conceivable forecast method, every conceivable forecast using each method, and return that forecast with the least amount of errors in no time.

We can solve a complex scheduling problem, using cloud computing and mixed-integer programming in minutes, rather than hours, and return a provably least-cost set of schedules.

AI and cloud computing also can help us develop incredibly accurate requirements calculations, custom for every contact stream, regardless of the multi-skill complexity.  Our experiment shows that these models can be developed and recalibrated every night, for every call stream.

Cloud-Based Systems Can Run in Parallel

This is the big aha.  Not only can we run one schedule fast, but because cloud computing is virtually unlimited in scale, we can run hundreds at the same time in parallel.  Have you ever wanted to run a scheduling what-if? Soon, you will be able to.

It is an exciting time to be in workforce management.  As we continue to automate our function, and start by rebuilding our core algorithms, we will be able to get to the point that we can ask our systems hundreds of questions, and they will be able to accurately answer them all, in minutes.

Ric Kosiba, Ph.D. is a charter member of SWPP and vice president of Genesys’ Workforce Systems. He can be reached at Ric.Kosiba@Genesys.com  or (410) 224-9883.

What Would You Do If We Could Automate Everything?

By Ric Kosiba, Ph.D.  Vice President, Genesys

More AI, Automation, and the Like

There is a new resurgent theme among workforce management vendors: artificial intelligence and workforce management automation. Over the last year or so, in this column, I have spoken about AI and forecasting as well.

I believe that some of this talk is hyperbole, vendors hopping on the AI fashion train, but some of it is becoming real.  To be honest, I do not think that, for our workforce management problems, AI is the powerhouse technology. Rather, it is the introduction of cloud computing that allows us to run more sophisticated algorithms (including AI) in our workforce management systems. Pure cloud installations, or hybrid premise-systems-with-cloud-based-algorithms, will thankfully be coming to all of us.

Cloud computing is an enabler. It enables machine learning forecasting systems, like we discussed in the Spring 2018 edition of On Target.  It enables us to build schedules using true mathematical optimization, as we described in the Winter 2018 edition of On Target, with all the value add-ons we discussed. It enables us to get to the point that every one of our algorithms are customized to every contact stream, are proved accurate, and are shockingly fast. Instead of schedules taking 15 minutes to several hours to run, they can run in a few minutes. Cloud computing, and modern algorithms enable this.

The History of WFM Systems

Many of the WFM systems in use today were developed at a time when computers were slow, memory was expensive, and distributed computing was a fantasy.  Scheduling algorithms, which are computationally hard, had to be built in a way to solve in a timely manner on these slow computer systems. This generally necessitated that the algorithm developer trade off optimality for speed.  To get solve times to an hour or so, the methods developed required work arounds and an attempt to find “good schedules,” not necessarily the optimal best set of schedules.

As contact centers became more complex, with multi-site, multi-skill, and multi-channels added, these algorithms were not replaced, rather they were altered to schedule for all this additional complexity; this new, more complex scheduling problem was shoe-horned into an old solution process.  For example, outbound and email channel schedules were developed by determining staffing requirements, even though the concept of requirements is sort of absent for each of these operations.

But it is all changing with the advent of cloud computing.

Prerequisite to Automation

Workforce automation has a current buzz.  Many of the tasks of the workforce management team, like determining best training schedules, automatically detecting shortages before they happen and prompting agents for overtime, etc…  is very cool. However, there is a prerequisite to automating many of the workforce management functions, and that gets us back to our modeling discussion. Automation hinges on the ability to accurately and automatically develop forecasts, accurately and automatically determine staffing requirements, and optimally and automatically schedule and re-schedule our agents.

This, of course, implies our systems are fast, accurate, and optimal.  We won’t want a system to automatically offer overtime if the requirements calculation is off. We don’t want it to automatically approve time off if our system calculates slowly and cannot respond to changing conditions quickly. We don’t want to trigger any real-time changes if our forecasts are inaccurate.

So, to automate, we need to upgrade the building blocks of our workforce management systems.

All the WFM companies are investing in cloud-based algorithms

Actually, I’m not sure of this, but I expect so.  If automation is desired, a hybrid or cloud computing model is absolutely necessary.  We’ve done some experimentation and here are some results.

Like we discussed in On Target before, we can automate the forecasting process and use machine learning and cloud computing.  Our experiment shows that we can solve, using sophisticated methods (like different flavors of Holt-Winters), over 4 million forecasts in under two minutes.  This means that cloud computing and AI allow us to crank through almost every conceivable forecast method, every conceivable forecast using each method, and return that forecast with the least amount of errors in no time.

We can solve a complex scheduling problem, using cloud computing and mixed-integer programming in minutes, rather than hours, and return a provably least-cost set of schedules.

AI and cloud computing also can help us develop incredibly accurate requirements calculations, custom for every contact stream, regardless of the multi-skill complexity.  Our experiment shows that these models can be developed and recalibrated every night, for every call stream.

Cloud-Based Systems Can Run in Parallel

This is the big aha.  Not only can we run one schedule fast, but because cloud computing is virtually unlimited in scale, we can run hundreds at the same time in parallel.  Have you ever wanted to run a scheduling what-if? Soon, you will be able to.

It is an exciting time to be in workforce management.  As we continue to automate our function, and start by rebuilding our core algorithms, we will be able to get to the point that we can ask our systems hundreds of questions, and they will be able to accurately answer them all, in minutes.

Ric Kosiba, Ph.D. is a charter member of SWPP and vice president of Genesys’ Workforce Systems. He can be reached at Ric.Kosiba@Genesys.com  or (410) 224-9883.