First things first: assuming a predictive model will be a “good model”, how can you tell if it’s a good way to spend money and improve your business, given all the *other* ways you can spend money to improve your business?

What if the model ends up being “less good” in fact, compared to how it sounds in a proposal?

***

Predictive models can help solve material business problems by answering difficult questions like:

  • Which of my customers are probably getting ready to leave next? (Customer Retention)
  • Which of my best employees are most likely to resign? (Employee Turnover)
  • What do my “Best Customers” have in common that sets them apart from the rest?

Understanding probability, by analogy with games of chance, is the best way, I’ve found, to understand how to assess a likely ROI from building and using a predictive model.

And it’s kinda fun!

So how do you decide if investing in a predictive model is a good investment? Let’s look at coins, dice and cards to understand how.

The flip of coin

I offer you $1.50 every time I flip a fair coin and it comes up tails.
I ask you to pay $1.00 for every coin flip.
You must play 100 times.
Do you play?

Knowing that tails is likely to come up 50% of the time, you should see ~ 50 tails if you play 100 times. You pay $100 to play, and likely earn 50 * $1.50, or $75 dollars.
Now you might see 48 tails in 100 flips, or you might see 53. The more times the coin is flipped, the more likely you are to see 50% tails. But even with 53 tails, you only make 53 * 1.50, or $79.50.

You don’t play.

A roll of the dice

I offer to pay you $7 every time I roll two dice and the total of both dice is 7.
I ask you to pay $1.00 for every roll of the dice.
You must play 100 times.
Do you play?

Knowing that two dice adding up to 7 is likely to happen 6/36 (~17%) of the time, you should see about 17 combinations equaling 7 in 100 rolls. You pay $100 to play, and likely earn 17 * $7 = $119.
Now you might only see lucky number 7 come up 15 times, or maybe even 20 times, in 100 rolls of the dice. The more the dice are rolled, the more likely you are to see 17% of them come up 7.  But even with 7 coming up only 15 times, you still earn 15 * $7 = $105.

You would likely make money if you play…but is it enough to want to play? Is earning $5 worth your time and money to do so?

Pick a Card, Any Card…as long as it’s a King

I offer to pay you $20 every time I deal you a King.
Each deal consists of 13 cards, face up.
I ask you to pay $15 for every deal.
You must play 100 times.
Do you play?

Knowing that 1 King is likely to show up anytime 13 cards are randomly dealt from a 52 card deck, you should see 1 King every time you play, or 100 times. You’ll pay 100 * $15, or $1,500, to play, but you will likely earn 100 * $20, or $2,000.
Now you might not see a King with every deal, and sometimes you might see 2 or more.  The more times you play, the more likely you are to have gotten 1 King, on average, for all the hands dealt. Even if you only see 95 Kings in 100 deals, you still earn 95 * $20, or $1900, while only spending $1,500.

Sounds like you should play!

But you will have to put $1,500 to work in order to play and win…maybe that’s a lot of money for you to spend on anything, even with a high likelihood of a positive return. Should you do it?

And what if someone comes up to you just before you decide to play and offers you $2,250 dollars to be repaid in 1 week for the $1,500 you have right now. What do you do then? Which is the better use of your $1,500?

A Predictive Model

Just as you had to think through cost to play, probability of winning, and the amount of payout when you win, you need to do the same calculus when evaluating the potential return on building and using a predictive model in your business.