Database Reference
In-Depth Information
for me. Another thing that excited me about PlaceIQ was being able to work
in a data-intensive world where not everything is a time series. It was a little
overwhelming at first, but I learned and continue to learn new techniques.
Gutierrez:
Any other challenges you encountered moving from finance to
ad tech?
Lenaghan:
One of the differences I noticed when moving from algorithmic
trading to mobile advertising was the infrastructure. The infrastructure in
finance is much better. The financial industry has been around much longer,
so the infrastructure is much more developed and built out and is much
more intolerant to failures. In the ad tech industry the standards are a little
bit looser.
Gutierrez:
In finance, if you make an error, such as overfitting a model, then
you lose money. Is there the same sense of urgency or pressure in your work
at PlaceIQ?
Lenaghan:
No, there's not the same sense of urgency or fear. If you overfit
your trading model, it loses money. When making money is your job, losing
somebody else's money is one of the most horrible sinking feelings in the
world. So you are very incentivized to not overfit your models. If you overfit
your ad tech model, the repercussions are less dire.
When an ad misserves, it's bad professionally, but the revenue hit is more
incremental. I do try to instill in my ad-tech data science team a sense of—to
put it not entirely facetiously—healthy paranoia about the quality of our data
and the robustness of our models. But the feedback on my ad-tech team when
something goes wrong is not, “Oh my God, what a disaster!”—but instead,
“Let's fix it.”
So in terms of the penalties for error and the allowances for correction, the
two worlds are somewhat different. Internally, the urgency at PlaceIQ is not as
relentlessly instantaneous as it was at the trading desk. Externally, we still have
to be very careful to get whatever we publish right, because if something in a
mobile campaign goes wrong it could be a very large revenue hit. At PlaceIQ
we work mostly on large, direct deals rather than a lot of small channel deals,
so losing one partner is always a big deal. A serious error can incur the very
serious cost of losing a large client.
Gutierrez:
What was the first large data set you worked with at PlaceIQ?
Lenaghan:
The very first big data set that I worked with was anonymized
location histories from an ambient background location app. We selected this
very large data set as a test set to see how well we could actually contextual-
ize movement. We wanted to understand at a very high level where people
were living and what types of behaviors we could correlate with in-home
demographics of social data.
