We started Urudata R&D in April 2024 with a simple idea: to do Applied Research and,
through it, generate new business lines for the organization.
But what is Applied Research? Or at least, how do we conceive it at Urudata?
Fail fast, fail often, and create a significant impact with just a few successes.
If we don’t fail often, we’re not doing research; we’re just applying the state of the art in
technology, which is very valuable, but not for our team.
The fear of failure is the executioner of innovation in our organizations.
Let me tell you how the idea of Applied Research formed in my mind over the years. I
think it’s the best way to convey what we’re trying to do at Urudata.
The first time I became familiar with the concept of the ‘gap’ between Basic Research
and Product Research was during Ray Ozzie’s arrival at Microsoft and his idea of
creating the ‘Microsoft Labs’.
These connected basic research (developed by Microsoft Research on fundamental
topics with timelines of 10 or more years to achieve results) with Product Research
carried out by the product teams in Redmond (where the company’s major investment
was, with timelines of no more than 3 years and generally a horizon of a year and a
half).
Although we can link Ray Ozzie’s ‘Red Dog’ and the Platform as a Service initiative
(which became Microsoft Azure) as his legacy, in reality, we can say that the labs failed
spectacularly and could not break the barriers between Basic Research and Product
Research (only an insider like Satya Nadella, almost a decade later, managed to align
the organization with a clear vision and without hesitation to remove two or three
misaligned executives in a single morning).
When Microsoft Labs were created, I tried to move to that organization, and after some
visits arranged by my manager (Eduardo Mangarelli) to talk face-to-face with team
members and learn what they were doing, I clearly saw they were disconnected from
reality, and I didn’t even try to apply to Microsoft Research or Microsoft Labs.
My first opportunity to learn deeply about the topic and apply it came with the opening of
Telefónica R&D and CORFO’s international center of excellence in IoT and data science
in Chile, which was very focused on ‘Product Research’; and although it was a 4-year
project, it was unbalanced toward product and with little room for Applied Research.
So, to solve this problem, together with Loreto Bravo and Universidad del Desarrollo,
we created the Data Science Institute, dedicated to working on Applied Research,
solving real problems based on the data we had at Telefónica R&D, resulting from the
vast number of sensors deployed in IoT experiments in Mining, Precision Agriculture,
and Smart Cities in 8 Latin American countries, as well as cellular network telemetry in
Chile.
That's right, when we started the discussion with Loreto, she came up with this slide:
That was exactly what I was looking for—not an academic collaboration where a
researcher ‘sells’ you their research to apply, but rather takes one of your unsolved
problems with the current state of technology and evolves the technology to solve it.
That was exactly what I was looking for—not an academic collaboration where a
researcher ‘sells’ you their research to apply, but rather takes one of your unsolved
problems with the current state of technology and evolves the technology to solve it.
The key to success? A research team focused on the problem and not on ‘their
research topic,’ high-quality data available, and an engineering team with a startup
mindset to experiment and fail fast.
And then we arrived at Urudata R&D—a young team (except for the writer),
multidisciplinary and ambitious to solve problems, paired with a young engineering team
led by Ignacio Cattivelli (who brings both engineering excellence and a startup mindset),
both teams focused on solving real problems for our clients that require pushing the
state of the art in technology. These are problems where, a priori, one cannot find an
adequate solution and wouldn’t even dare to estimate a time to solve them.
Today, we are already 16 people on the team, and we have solved several ‘impossible’
challenges:
In various types of fraud models.
- We aligned the origin-destination matrices of IMM with those of ANTEL.
- We used a combination of AI and transformations based on formal logic to transform legacy code for our clients.
- We turned several processes that ‘couldn’t’ be done in real time into processes with minimal latency.
- We turned several processes that ‘couldn’t’ be done in real time into processes with minimal latency.
And the most demanded in almost all projects: agentic and resolutive AI, that is, AI that
observes and takes actions (agentic AI today is trivial, but the one that makes decisions
requires very high levels of excellence and validation to deploy).
What is our process today?
We receive requests from our clients or other teams within Urudata and classify the
situation: problem, state of the art, process mining, data exploration, etc.
And if we validate that we are facing a relevant problem that cannot be solved with the
current state of technology, we formulate hypotheses and experiments to fail fast (and
cheaply) and focus on addressing the problem by pushing the state of the art in
technology. Thus, if in a limited time we achieve a solution, the engineering team builds
the final solution; if not, we go back to the beginning…
Fail often.
Fail fast.
Fall in love with problems, not solutions.
Move the needle significantly in cases that end up being implemented quickly, providing
real value to our clients.
That is what defines the Applied Research we do at Urudata R&D. Or rather, it is what
defines us as a team.