MediaSlap blog

“Why Monkeys Are Better at Scheduling Music”
by Todd Zarnitz

Why do people hate radio? Is it an image problem? It must be! We research the music listeners want to hear, and then we rotate the songs through the day. And the cost to the listener? Totally free of charge! We should be the most beloved industry on the planet.

So what is the problem. Not enough new music? Not enough deep cuts? Not enough personality between the records? We have tried to address all of these areas, with overwhelmingly poor results.

The actual problem is two-fold. And the solution will knock your socks off.

Problem one: the DJs cannot access the music. DJs can interact with the audience between the records, but they cannot play a request or play a song that fits into an immediate conversation. Scheduled music makes this kind of interaction difficult, if not impossible.

Problem two: by scheduling music we are making radio far more predictable and repetitive then we ever imagined. I will prove it with math.

MUSIC DISTRIBUTION

If you forgo music scheduling and let the DJs pick the music, what will happen? Songs are going to start coming up at random. You will not see a nice song/artist exposure crosshatch in Selector, that’s for sure!

However, the random DJ spins will still produce an even artist/song impression across dayparts. What??

Take a 10 x 10 inch piece of paper and draw some type of random blob. Now, tell me how many square inches the blob takes up. Ouch! How are you going to do it? I bet you wish you had drawn a nice even square.

One way to find the area is to grab a ruler and draw a square around the blob. Calculate the area of the square (height x width, dummy!). Now between the square and the blob start drawing smaller squares. Subtract the area of the little squares from the big square, and we start getting closer the to answer. These are the basics behind calculus. It is insanely complicated and now my head hurts.

Let’s try another approach. Find a monkey and make him throw darts at your piece of paper. Add up the number of darts the monkey landed on the blob. Take this number over the total darts thrown, and multiply it by the area of the paper. You now have the area of the blob.

Total area of the paper = 100 square inches
169 monkey darts hit the blob, 1000 total monkey darts thrown on paper
100 square inches x (169/1000) = 16.9 square inches of blob

We have just utilized the Monte Carlos method. The Monte Carlos method was advanced during the development of the hydrogen bomb at the Los Alamos National Laboratory and has become popularized in the fields of physics, physical chemistry, and operations research. [1] The key to the math is simple: random patterns will produce an even distribution across a target area.

The implication for radio is staggering: If you pick songs at random, they will automatically distribute themselves evenly across the dayparts.

PROGRAMMING PLAYING CARDS

Let’s program a radio station, not with music, but with a deck of playing cards.

Here are the rules:
We draw five cards every hour.
the Rank of the card (2 through ace) = artist
the Suit of the card (spade, diamond, etc) = song

This is a pretty hot station. There are 13 artists and each artist has 4 songs.

The goal is to give each artist an even distribution throughout the day. We will try two methods. 1) Scheduling the cards, and 2) letting a monkey pick the cards at random.

For simplicity, we can ignore the suits. For every card that comes up, we will just assign the next song from that artist’s four song rotation.

CARD SCHEDULING

Let’s try to schedule the cards. Keep in mind that we want an even rotation for each artist across the day. First, I am overwhelmed by the complexity of the task. Take a minute before continuing to figure out how you would attack the problem.

The best method I can come up with is an artist schedule that cascades down the day.

http://spreadsheets.google.com/pub?key=p2Xd4yhCK5X31BOKg4JfeZQ

You can see that we have achieved a nice even rotation for each artist. But there are huge problems.

Patterns keep repeating. Look at how many times the 4 and K card come up in the same hour. Even more surprisingly, look at how many hours the Q, 9 and 6 cards come up together! Find the J card, and see how many hours are followed by the 2 card…

Look at the embedded patterns. At the bottom of the spreadsheet I have graphed the distribution of the King card for the week. As you move down each hour, the accumulated 5 day King card exposure repeats back and forth between 3 and 1 spins.

Words that come to mind:
Predictable. Boring. Repetitive.

MONKEY POWER

Now shuffle the deck and let a monkey draw 5 random cards for each hour. After the 52 cards run out, re-shuffle the deck and put the monkey back to work.

We have an immediate problem. On average, we end up with a pair of cards every four hours, a three-of-a-kind every 70 hours, and a four-of-a-kind every 1700 hours (once every 3 months).

Since we do not want more than one song by the same artist in an hour, let the trainer help the monkey by moving a repeat card either up or down by one hour.

http://spreadsheets.google.com/pub?key=p2Xd4yhCK5X0U72ivS_0hMQ

What are the problems? The artist impressions seem to bunch up occasionally. But how big of a problem is this? Which is more dangerous: an artist bunching up down subsequent hours or across the week on the same hour?

I have graphed the King card impressions at the bottom of the spreadsheet. Look at the K exposure across each hour after 5 days. Is this more even or less even than you expected? Now look at the King card impressions per daypart and compare it with the scheduled station. Over a short amount of time, everything starts to even out.

Words that come to mind:
Unpredictable. Interesting. Complex.

SCHEDULING vs RANDOM

Both models are working with the exact same deck of cards. The music is just as tightly controlled. But look at each day!

Compare Monday on the scheduled station with Monday on the random station.

On the scheduled station, how similar are Monday and Tuesday? The days are almost identical! Can you spot any similarities between Monday and Tuesday on the random station? There are none.

Which station would you rather listen to?

ECONOMIC REALITIES

How much time do we invest in scheduling music? If you spend 3 hours/day on Selector, here is the breakdown:

15 hours a week
780 hours/year
97.5 days/year (workday = 8 hours)
20 weeks/year (work week = 40 hours)
5 months/year

All this effort to produce boring, predictable radio?

Meanwhile, by picking the songs at random, we could be producing an exciting, unpredictable and complex rotation that also ends up exposing the songs evenly across dayparts.

CONCLUSION

Give the music back to the monkeys. They do a better job.

footnotes
[1] http://en.wikipedia.org/wiki/Monte_Carlo_method

by Todd Zarnitz

###