Supply curve: servers for your restaurant
1. Supply curve: servers for your restaurant
Previously, you examined the demand for your restaurant. The data scientists shared the forecasted demand, but you noticed some issues with that demand. While they are fixing it, let's examine the supply side of the operation. Remember supply and demand doesn't solely apply to restaurants, any system containing queuing can benefit from supply and demand forecasting and decision making.2. Determine an appropriate supply to meet the demand
Once an operation knows the expected demand, there are many techniques to determine an appropriate supply to meet that demand. One common method is to use a formula from the telecom industry called an Erlang-C. Before we do that, let's understand it conceptually. Suppose your operation expected 100 phone calls in a thirty minute period. Each phone transaction takes 5 minutes to complete. On one hand, you could hire 100 call center agents. No customers wait and they all work for 5 minutes out of the 30 minute total. Obviously that seems like a waste of your company's salaries since 100 agents are sitting idle for 25 minutes.3. Determine an appropriate supply to meet the demand
On the other hand, let's only hire 1 call center agent. The agent can only help one person at a time so within 30 minutes, only 6 customers can be helped, 5 minutes for each call. That means the operation will have a lot of unhappy customers. In fact, some customers will just hang up out of frustration... and if your operation is taking telephone orders that means lots of lost revenue. On top of that, your employee will be miserable. It's just non-stop work!4. Erlang-C formula to the rescue
An Erlang-C formula, can solve this problem more elegantly. This formula has four inputs and will return the number of servers, agents, or workers needed. First is the rate, what is the expected demand within the interval? Next, the duration input is the average amount of time needed to complete a transaction, like a phone call or checking a customer out at the store. The nuances of the Erlang-C is that it incorporates a target waiting time and service goal. The target waiting is the number of seconds an incoming demand is willing to wait. The service goal is a percentage of much of the demand should be serviced within the target.5. Math? Ain't nobody got time for that
The formula itself is a little complex for this course. This is because the total incoming demand is measured in erlangs and assumes a poisson arrival distribution within the interval. In our example, this means 100 calls wouldn't show up all at once in the 30 min interval. Lucky for us, R, python, other programming languages and even online calculators contain Erlang-C functions. Plus, our exercises will just need the previous inputs.6. Balancing incoming calls, wait time, and agent occupancy
If you input 100 calls, 5 minute average transaction times coupled with people willing to wait 20 seconds and your operation's desire to answer 80% of calls in that timeframe, the Erlang-C calculates that 30 agents are needed. In fact, all 100 calls get answered, some just having to wait longer than 20 seconds. Specifically, 82% of calls are answered within 20 seconds and employees are occupied about 79 percent of the time, which is 24 minutes of the 30 minute total. This means, the organization saves money by not over-hiring, the agents have a little time off, and customers are generally happy.7. Hurry up and wait
Now you know the secrets of inheriting a demand forecast and using it to arrive at an acceptable supply. Next time you're waiting in line for your coffee, remember a lot of data-driven decisions were made including how long you are willing to wait. Let's apply this new knowledge!Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.