Database Reference
In-Depth Information
As more and more people have started using the system and following our
scores, we've had to think about how to score new customers who are people
our customers have never seen in the past. We also want to think about how
to have salespeople do exploration. We've actually started taking that into
account as well. More than that, with more customers, we've started getting
more information from several customers that work together. For example,
we have data from two distributors of the same company and we can see that
both distributors are trying to sell to the same person. This then lets us turn
around and ask important questions such as, “Why are you doing that?” So for
us, what's amazing is that because we have all this sales data, we see insights
and things in the business world that wouldn't be available in a different way.
We're actually combining all of them together and mining them.
Gutierrez:
What are the most surprising things you've found from one of
these models?
Radinsky:
The most surprising element of the work we are doing deals with
the issue of our customer's perception of our model results and then how we
work with that perception. Everyone talks about human-computer interaction.
I think we're actually implementing it here. So for me, this is the next step. We
have to figure out how to produce great results based on our understanding
of our customers, as well as our understanding of our customer's customers,
so that we can combine them effectively. This is really exciting to me.
For example, a type of thing our system will see is that a salesperson is looking
to sell to a senior engineer. That senior engineer will have the characteristics
of having had their career grow very quickly and have their company grow-
ing financially as well. That is the perception that a salesperson will have of a
senior engineer they want to sell to. In order to find that person, we have to
figure out a few things behind the scenes, like how do we know that a com-
pany's growing financially. We want to figure that out in order to be able to
add the figures as a feature of our models for the classifiers. So we have to
take in a lot of data to be able to match the perceptions of how a salesperson
thinks of a senior engineer they would want to sell to.
Another thing that's been pretty interesting and surprising for us is the varying
levels of knowledge our different customers want about a new person who
comes into the sales funnel. Because we automatically install for each type
of customer and we don't know our customer's data at that point, we build
something called a “buyer persona” for each of their potential customers. But
different customers of ours care about different granularities of this buyer
persona.
The way it works is that a potential customer goes into their system and they
will want us to tell them who this person is. They want to know whether
that person is a technical person, or a marketing person, or something else.
One of our customers cares about the difference between this person being
