TiVo builds its corporate data democracy with Splunk - aims to identify customer problems before it gets the complaints call
- Summary:
- We speak to Todd Kulick, TiVo’s VP of Technology, about how Splunk is helping the streaming service keep on top of its data estate.
I got the chance to speak to TiVo’s VP of Technology, Todd Kulick, who has overseen a ten-fold increase in the company’s usage of Splunk, and explained how the platform is helping it solve a wide array of its data challenges.
The company has five instances of Splunk globally, both in Europe and the US, all of which are interwoven to meet TiVo’s GDPR and technical needs, in terms of holding, finding and searching the data, according to Kulick.
TiVo is ingesting close to three quarters of a terabyte of data a day into Splunk.
On why and how TiVo uses Splunk, Kulick explains that the platform is fundamental to providing an enhanced customer experience to its users. He says:
One example, is just time to insight. We often have bugs in our code, believe it or not. We are able to find defects better and faster, with less engineer time, which means we ship cleaner, more reliable releases. We use it to monitor our running TiVo service. So if you buy the TiVo video service product line, our ability to keep the TiVo service up and running impacts many of your features, whether its searching for products you want or us predicting which shows you might want to watch.
We also use it to measure a bunch of aspects of our release readiness, related to performance and other things. So, basically monitoring the service and making sure that the product is as reliable as possible. In a maybe slightly more meta way, we use Splunk as a tool to understand all of the data that’s coming out of our software development organisation.
So we have 1,000 software developers and they generate a bunch of machine information, in terms of bugs, code reviews, check-ins, builds, test runs. And that information itself is complicated and large. We use Splunk as a tool to understand and retrospect on that information. For example, which areas of my software are most likely to generate defects when changed? That kind of information is super useful to allow us to make sure we do additional code reviews or testing of areas that are more difficult, or likely to lead to problems.
Why Splunk?
Kulick explained that when identifying a data platform tool, TiVo opted for Splunk as it was the largest company in the market, which Kulick found meant that its feature set was ahead of its competitors. Kulick also favoured Splunk’s UI and says its useful to be able to do ad-hoc analysis on the back-end.
Given TiVo’s huge software development capacity, Kulick did consider building an in-house developed data platform using open source tools. However, he found buying off-the-shelf was a far quicker route to market. He says:
We are a company that has a lot of software development capacity, so we could build a lot of these things ourselves, and we did for quite a long time. But we find that the time and investment required to turn open source infrastructures into tools that would provide value, far exceeds the cost profile of just buying Splunk and getting straight to work on the problem we want to solve, which is making TV better.
This formed a key part of TiVo’s value assessment of Splunk, which is quite often referred to as a ‘pricey’ option by customers. However, Kulick found that the money saved on human capital more than made up for the money spend on the platform itself. Kulick explains:
I was involved in the value assessment of Splunk. They’re quite good at helping you do that. Their value assessment forms and worksheets had five or six dimensions on which you could try to capture and measure value in terms of usage of the tool. I found their modelling to be relatively fair. A lot of time our vendors have models that estimate things in their favour. But I found Splunk’s model to be relatively fair.
And after only using two dimensions, I was able to convince myself we should invest. I am maybe the unusual person that doesn’t think Splunk is super expensive. It’s certainly not cheap, but I spend a lot of time recruiting and trying to maintain and mentor Silicon Valley software engineers - they’re quite expensive. If you look at the human resource that we are not investing because we have these tools, the decision is quite easy to make. I was able to justify it to my C-Suite relatively quickly, based on my value assessment. And that’s held up quite well.
The value of Splunk at TiVo
Kulick was keen to highlight that Splunk is more than a logging tool to TiVo. He says that this is a common misconception about Splunk in the market, which historically sits on the ‘techie’ end of buyers’ perceptions. Kulick says that Splunk is able to underpin a company’s entire data activity. He says:
The advice I would give another executive if they asked me about our implementation would be, make sure you don’t understand Splunk just as a logging tool. It really is an infrastructure on which you can build your entire data corporate data democracy and organisational strategy. It’s a platform for doing that.
I meet many people, including in our organisation, although we’ve changed a lot of minds at this point, that think it’s just a logging or security tool. I also think they think about the cost wrong, they don’t think about the human capital cost. I think those are the things that are misunderstood about Splunk most.
And TiVo has some big and interesting use cases for Splunk going forward, very squarely aimed at improving the customer experience. Kulick heads up a small data science team, which is made up of four people that focus on SWAT analysis. And that team has a long list of aspirations for Splunk, according to Kulick. However, one that they’re hoping to roll out this year is focused on identifying customer problems before the customer is even aware of them. He says:
I’ll give you an example of the kind of aspirational things that we think about. We have machine data in our customer care tool, that collects information about every time a customer calls us, if you have any kind of challenge with your TiVo devices. Every one of those calls, we know the caller’s device and we code the reason we called us. We also know the time and duration. We also have a lot of machine logs, about 10 megabytes a day from every set-top box.
I believe it’s possible for us to figure out to some extent which customers are going to call us from those logs, by building machine learning models. And then fixing the problem they were going to call about before they ever call. We aren’t quite there yet. We have machine learning capability in-house and we are starting to predict things.
We are trying to use those tools in a bunch of exciting ways, but that’s the dream scenario. I expect we will try that before the end of the year.