Looking for insight from the gurus out there:

We schedule/forecast using good ol' Erlang C plugins in Excel.

My company has as real-time chat skillset. The agents that work this skill have 2 chat windows open at any one time.

We've experimented several ways to factor the multiple chats but none have been successful compared to actual results.

Just wondering how this is done elsewhere?

Before I make a lengthy post...do your people multitask with other duties simultaneously? E-mails or phonecalls?

The only simultaneous activity is the Chats.

If they conduct Email/Inbound then they are removed from the chat skillset.

Dan:

Forecasting for chat is fundamentally incompatible with Erlang for a variety of reasons:

1) Erlang assumes that an agent handling a call is unavailable to handle other calls until the call and the wrap time are over. Obviously chats don't work this way.

2) Some chat agents can handle more simultaneous chats than others. Erlang has no way of representing this.

3) Chat sessions have often have lengthy periods when the agent is just waiting for the caller to respond. The agent can multi task during these periods. Erlang assumes the agent is fully committed to one event at a time.

4) Chat sessions can be much longer than phone conversations. If you use the average duration of each event then you may be grossly overestimating the resources required.

5) Erlang assumes that events that start in an interval also end in the same interval. Because chat sessions may last much longer than phone calls, this presents a problem. Chat sessions that arrive late in the interval will carry over into the next interval but Erlang will ignore the carry over effect. Some chat sessions are long enough to span many intervals.

6) Erlang assumes that call volumes are random (relatively even across the planning interval). Chats can be much more volatile, arriving in sharp spikes or arrive in patterns that are heavily skewed. This can cause serious under forecasting, missed service levels, dropped sessions and other quality problems.

7) Some Chat response systems automatically tell callers that no chat agents are available when the chat queue reaches a threshold. This type of functionality cuts the top off of your handled chats meaning that if you are understaffed for chat, you may never know about it.

Having studied chat forecasting extensively, we have developed a solution that forecasts and schedules chat agents faithfully to all of the unique characteristics of chat. If you would like to try it I would be pleased to set you up for free. The solution is entirely web based so we can host it for you or it can be installed on-site.