Fred; I have been pitched over the past 25 years in franchising all manner of snake oil for "profiling" and many have said that they can measure top performers and we could use the model to recruit high caliber and similar franchisees and employees. 

What do you say to a franchisor who wants a reliable predictive performance tool, other than buy yours? 

How should they approach it practical manner?  What should they expect from using such a tool?

fred berni1.jpgJoe - Let's start with the 5 basics of all good design.

1. Predict Performance:

Make sure that whatever system you're considering was actually designed to predict performance for the job, in this case, owning a franchise.

2. Designed for Selection or Recruitment

Make sure the system under consideration was designed for selection. Several of the most common personality profiles specifically state on their websites that they were not designed for selection.

One even goes so far as to say it's unethical to use it for selection. Even so, people are out selling these profiles for selection purposes.

3. Don't Discriminate in an unreasonable manner.

Third, make sure that the system you're considering does not discriminate. Even unintentionally. See Griggs v. Duke Power Co. A good summary is at http://www.answers.com/topic/griggs-v-duke-power-co#ixzz32NTJOd4J

4.  Verify Performance Carefully, using an Independent 3rd Party.

Unfortunately, there's no simple way to run an analysis between performance and "profile scores". No matter what you're told.

There's a fair amount of complexity involved in comparing scores and actual performance.

And by performance, I don't mean simply a gut feeling of they're good or bad.

That's subjective criteria because the ratings can change depending on who's doing the rating and their relationship to the franchisee. That's not to say you can't use subjective data, just that if it's available, use objective data with subjective thrown in.

Objective data is something to which you can relate to hard numbers.

Things like actual $ sales per location, % increase one year over the next etc. I've even had clients in auto repair use $ sales per bay.

It all works as long as you can put a hard number on it. Then you have to use rigorous scientific methods to analyse the relationships between "profile" scores and performance. Hopefully this analysis is done by a 3rd party with no stake in how the results turn out.

5.  Include Everyone and Increase Sample Size.

You'll need to include good, bad and average performers.

Here's an example of what I mean: Let's assume you've just made a movie and want to project what your ticket sales will likely be.

If you only include good reviews you could say that 100% of the people that saw the movie like it.

True, but not accurate. Only by adding the bad and average do you get an accurate idea.

In the same scenario, how many people would you need to ask before you are comfortable with a projection? Five? Ten? Twenty-five? One hundred?

Generally accepted sample sizes are a minimum of 100 before you can accurately predict performance. Anything less than that and the best you could do would be to assume you've identified a possible trend.

As far a reliability is concerned, my definition is to be able to say with 95+% confidence that the there's a definite causal relationship between performance and profile scores.

With larger sample sizes, our confidence level could be 99.5% or even higher.

Having said that, it really boils down to what you're willing to accept as being reliable. Being able to predict accurately 50% of the time? 75%? 95%? What's your comfort level?

The bottom line is that it's simply not appropriate, accurate or legally defensible to just pick your  1. or 20 franchisees and base your decisions on that small a sample. Even if you include poor performers in the mix.

Having said that, if the system you're considering was developed to predict performance of franchisees in similar type of franchises, and can provide you with documentation of such, in all likelihood you'd be safe just going with the "template" already designed by the developer of the system.

Hope that answers your question Joe. Did I miss anything?

Authors

Archives