7 FAQs for the Experts: “Effectively Target Micro-Moments with Real-Time Decisioning and Machine Learning”
Comviva, a global leader in the digital transformation space, has long been known for their innovative solutions and services. Within their MobiLytix Customer Value Management (CVM) suite of solutions, Comviva uses advanced Machine Learning for real-time decisioning, which allows users of the MobiLytix platform to target and provide the right experience to customers at their micro-moments of intent. To make this happen, Comviva leverages VoltDB to execute intelligent decisions in real-time, allowing leading banks and telecommunications providers to create better, more profitable customer experiences.
Recently, VoltDB teamed up with Comviva, to host the webinar, “Effectively Target Micro-Moments with Real-Time Decisioning and Machine Learning”. In the presentation, Greg Armstrong, Chief Marketing & Strategy Officer of Consumer Value Solutions at Comviva and Seeta Somagani, Director of Field Engineering at VoltDB, explored the technology building blocks that empower the capabilities within these next-generation solutions.
This blog highlights some of the questions asked by those who attended our recent live webinar. If you missed the webinar the first time, or would like to view it again, you may do so here on-demand.
Editor’s Note: Unless otherwise noted, the question is coming from a webinar attendee via our chat functionality during the session. Answers were provided by the presenters [Greg Armstrong and Seeta Somagani]. These answers have been edited for clarity and grammar.
Q: What industries is Comviva operating in?
Greg Armstrong: So today, our focus has been in communications and recently, moving into banking financial services. So that’s our heritage. The platform itself is not industry-specific. The things that really change when you’re moving between industries is the behavioral DNA dictionary because that has to be industry-specific, as well as the integration points and the external events that we would use and integrate with. But historically, we’ve been in communications and moving into banking.
Q: How does VoltDB differentiate from other database platforms?
Seeta Somagani: So there are lot of database platforms are out there today. I think it’s a very significant challenge to understand where VoltDB can be a fit and what technology might be the best for you guys. Typically, we are compared against traditional databases because VoltDB in itself is a relational database. The comparison with traditional database is done very often, but the main difference with Volt is that it’s written from the ground up from scratch. It is not meant to do everything in one database.
When your data-processing strategy expands from, say – putting everything in Oracle and maybe another data warehouse and then doing transactions in one and analytics in the other, to expanding to tier your database layer and handle the most important and most critical events as soon as they occur and handle them right away, VoltDB plays a big part in that area. So if you have a high throughput, high-performance application that requires consistent decision-making at the time of ingestion of the event, so you want to make the right decision at the right time, that’s where VoltDB is a great fit.
Q: What sort of gains or benefits are people getting out of the MobiLytix platform?
GA: This is a good question and never an easy one to answer. Often, when – as a supplier of capabilities, we’re not necessarily privy to all this information from other clients. When you talk about comparatives, it’s always challenging with different markets and different methodologies for doing those comparisons. But just to give you a bit of an idea, we have got a client who is consistently getting incremental revenue returns of around 2.5% a month and this is very robustly measured. This is basically a universal target group versus a universal controlled group. So taking all the revenue made from the target group in a month, subtracting from that the scaled up revenue from your controlled group and that then represents the net incremental revenue. And this is in a pretty developed market, so I know that one I feel very confident about.
There’s other examples where we’re demonstrating the power of machine learning and real-time information; we have a program where moving from sort of marketing intuition-based business rules to using modeling to predict offer uptake – and using real-time balance check – has increased the conversion rate of the program by about sevenfold. It went from mid-single digit [conversion rate] to 30% [conversion rate]. Also, the participation rate on the program grew dramatically – about 3x – which indicates the customers were getting a lot more value out of it.
So I think those are two pretty good examples. We also had done some benchmarking of looking at comparative performance of triggering campaigns off of real-time events versus using traditional batch-planned campaigns, and we saw a significant improvement in performance. So that got us to a point where we believe that being at the right time is very useful in getting a next-step of performance in CVM [Customer Value Management] campaigns. The one after that is when you’re basically looking at how you make sure that you’re improving your decisioning around what to offer.
Q: What is the vision for VoltDB in the coming days? Are things like expansion towards CDP or maybe prescriptive analytics plans?
SS: As a technology company and being relatively industry-agnostic, we’d like to stay out of anything specific to any industry, but I can talk a little bit about our vision for the future. We started as this very solid relational database that offers all the guarantees that you come to expect from a traditional database but at the speed of the NoSQL, combined together as a NewSQL, VoltDB database. So as we have progressed through – over multiple iterations of the product – we’ve added many benefits but again, our roadmap is taking us more and more towards a stream processing platform.
If you take something like Kafka, which has a very simple benefit of just being able to hold logs of events from multiple publishers to be consumed by multiple consumers, is something that VoltDB pretty much has the capability to do. In that regard, we would like to make changes and upgrades that allows different consumers and publishers to work with VoltDB. So, nothing in terms of anything industry-specific industry, but [more in terms of] making it easier for you to build the platform for your own industry and to essentially make the right decision at the right time.
Q: How modular is MobiLytix as platform? Is it possible to integrate just the decision engine with an existing marketing automation platform?
GA: Yes, it does, actually – it’s designed to be modular. Those modules I spoke about in the architecture. We’ve done that deliberately and ensure that we maintained separation between them. We’ve got scenarios where we have got, for example, the event detection and offer allocation modules and integrate with a legacy campaign management platform for managing the orchestration with the customer.
Q: What changes do you witness when creating the BDNA across industries? Is it related to the identifier or is there something more to it?
GA: I don’t have a lot of cross-industry experiences. I’d say we’ve worked on communications. We’ve worked in banking. This is what is specific to the industry because effectively you’ve got different profile attributes. I mean, there are some [profile attributes] which are common, but a lot of them are actually industry-specific based around the type of service that you’re selling. So, it’s a mixture of that.
The ones that are consistent are attributes that are sort of customer-labeled ones; associated with channel identifiers and things like that. Also, the behavioral DNA that we generate internally in the platform associated with the execution of use cases. For example, counts of offers being sent, lists of current open offers, offers that are being rejected, offers that are being accepted. But it is, as I said, going into different industries. It’s the behavioral DNA dictionary that differs primarily. The configuration of the platform itself, and the configuration of the use cases is industry-agnostic.
Q: What are some of the key differentiators of stateful stream processing versus traditional stream processing?
SS: Yes. So as the name says, the lack of state essentially. I mean, I think I need to expand on what state is.
In a typical stream processing solution, you can have some sort of a state wherein you would have some constant or static data; data that could be changed but it doesn’t really mutate with each coming event. A lot of stream processing solutions do their processing in a micro-batch manner and do not process events per event. So they lack the capability to act on every single event with a 360-degree view of the world in which that event is accurate. So all of the difference might seem simple and easy to fix but it’s rather something fundamental that either you are looking at the events coming in as a bunch of signals – each of which needs to be processed – or you’re looking at the events coming in as opportunities which you need to take the most advantage of, keeping in hand the full view or the full context in which these are accurate.
So the stateful stream processing takes the challenge that stream processing tries to address – which is handling the loss of events – but it doesn’t lose the intelligence that you typically get from the traditional database processes. So being able to put all those together, I think, can make the difference in competing in the market.
Check out the full, pre-recorded webinar in its entirety here.