Corios helped a vehicle manufacturing client apply predictive algorithms to connected vehicle and service center geofence data to help them pursue fleet uptime of 99%, and to reduce mechanical failures and warranty claims. This was a fascinating application of predictive analytics to the internet of things (IOT).

Our client is a major truck manufacturer who has the strategic global objective of increasing vehicle uptime from 96% to 99%. How they intend to get there is to predictively identify vehicles at risk of mechanical problems before they happen, by combining real time data from telematics alerts, service center geofencing, repair histories and travel and road conditions.

 

Corios brought in a top-down view of the problem; we gathered data from a variety of sources, including vehicle telematics alerts, warranty claims, vehicle repairs, vehicle registration, vehicle service histories, and equipment supplier profiles. This became a massive collection of data, and we rationalized all this information so we could see a complete 360 degree view of every vehicle and its detailed history over time. Then we applied predictive analytics to rapidly identify the leading indicators of the equipment failures which led to the increase in warranty claims.

 

Predictive assets aren’t academic exercises

The biggest challenge that confronts predictive modelers and data scientists is that it doesn’t matter how good any model is, if it isn’t used by other people in your organization to improve the way they run your business.

A recent Harvard Business Review article noted at a recent industry conference, that out of 150 data scientists, roughly a third had developed a model in the past year, but not a single one had deployed it into production, nor monitored that model’s effect on business value. That’s just sad.

We believe there are some common reasons that predictive models struggle to get used. They’re complex, they’re arcane, and there is no common process or vocabulary in most organizations that enables people across the domains of data science, customer-facing representatives, and technology to make daily business decisions using models. That’s a problem we are committed to solving.

Corios Tempo: our process for building an analytic model factory

We’d like to help the industry define the discipline of rapid model development and deployment. Our own internal name for this discipline is Corios Tempo, which reveals that we think maintaining a stable rhythm of activity is an important design element.

In this presentation, we’ll discuss:

• the problem we aim to solve,
• the design principles of a robust solution,
• the benefits of doing so,
• the roles and responsibilities inside the analytics organization that need to be harnessed to solve this challenge,
• the way we organize data,
• how we build models,
• how we validate and deploy these models,
• how we administer the platform on which all this asset development and management activity takes place, and
• the best practices that we have developed over my 30-year career in analytics as well as the decades of collective experience developed by the Corios team.

If you’d like to read more about how Corios Tempo solves this problem by providing enterprises with a repeatable process for building and operating an analytic model factory, you can get in touch with us to find out how you can build an analytics model factory.

During a recent industry briefing call, the Chief Analytics Officer of a life insurance carrier asked me an interesting question: “Should I try to get funding support for our new analytics initiative on the basis that it’s cool and sexy?”

When I prompted her about what’s sexy and cool about her proposal, she replied, “Well, machine learning is sexy, because especially in our industry, it’s the new hot thing. And it’s cool because I think we can improve how we make underwriting decisions on new policies.”

Then, I followed up with an important clarifying question, ”Do you have the technology support to implement that model, if you did indeed find a better underwriting mousetrap?” She had to admit she didn’t yet know how or whether she could implement such a new model in their production systems, which is maintained by the technology organization. She also didn’t know the best way to work with the underwriters to adopt the pricing policy changes that her upgraded model would support.

Why cool isn’t enough

Here’s why I have found cool doesn’t win the day: In many financial services firms, few executives want to be a guinea pig for an investment that is novel in their industry, especially in fiscally and operationally conservative environments like the life insurance industry. Cool simply isn’t enough.

Read More

At the gym last week, I saw someone wearing a company-branded t-shirt proclaiming “Data is cool”.

I had a lot of reactions to that statement, not the least of which was the subject-verb disagreement (since every Latin major knows “data” is plural, but let’s get past that). This might sound strange coming from an analytics professional, but no, in my opinion, data are not cool.

To many people entering the domain of data science, data might be novel and mysterious, but really, by itself, data are the measurement of a process. Measurement is only academic unless it’s informing a decision, so by itself, data is not actually cool.

What is cool? Our ability to capture data at ever-greater frequency, latency, and accuracy, and how it allows us to make decisions faster, more efficiently, and at a more fine-grained level—which produces stronger outcomes. When those decisions allocate resources towards strategies that maximize our chances to produce wins… that’s pretty cool.

It’s fundamentally the role of leaders to make decisions that allocate resources towards winning strategies. That role as the decision-maker is hard, risky, requires persistence and faith, and is the domain of courage and enduring values. When people make decisions about allocating resources towards those outcomes producing the best results, for as many people as possible, and reducing the pain of loss as much as possible, while still being efficient and utility-maximizing… now that’s more than cool, it’s sexy.

So there really ought to be a t-shirt that proclaims the core truth: “Data might be cool. But decisions are sexy!”

When I attend industry conferences or speak with Chief Data Officers (CDOs) and Chief Analytics Officers (CAOs) of large financial institutions, one popular question that arises is, “What do you hear about open source analytics in other large banks? Is it ready for production?”

While I’ve encountered substantial growth of exploration and analytics development occurring in private and public clouds using open source analytics, I’ve also been a little surprised (in two ways) at the findings of these CDOs and CAOs when it comes to actually deploying analytics assets using open source.

First source of surprise: large financial institutions with whom I’ve spoken are getting nasty wake-up calls about failures of their selected open source platforms to provide robust, reliable results. The initial attraction was the price of open source tools; the subsequent feedback is a broader appreciation for total cost of ownership, which isn’t as attractive as they first anticipated.

Second source of surprise: that collectively, we didn’t see this coming.

Read More