Email a colleague    

May 2019

Mobileum on the Move: Attracts Several Operators to its IRSF-Analyzing & Opex-Saving Counter-Fraud Machine

Mobileum on the Move: Attracts Several Operators to its IRSF-Analyzing & Opex-Saving Counter-Fraud Machine

When it comes to IRSF fraud solutions, there are two opposing strategies on what kind of platform to buy.  And I’ll present these as pure black or white choices even though we know the world is polka-dotted:

  • Keep & Maintain the CDR-based Fraud Solution that’s been the mainstay of IRSF fraud fighting for decades.  The CDR approach is a proven, affordable solution and has been enhanced to include better customer profiling, stronger case tools, integration with blacklist databases, email alerting, and modularity within a larger business assurance platform. 

    A key premise is that the CDR solution takes strong direction from a trained and experienced team of fraud analysts.
  • Opt for a Machine Learning-based System, a more expensive investment but featuring real-time detection/blocking and finer grained analysis of the fraud “event”, enabling specific treatments of each fraud event.  More importantly, the machine promises to greatly reduce the analysis and response time, thereby providing more bandwidth to the fraud-control-trained manpower.

Two years ago, you’d have to say these choices were not real in the marketplace — so few machine learning fraud solutions were deployed.  But now Mobileum has turned a successful Tier 1 U.S.-based client deployment into significant repeat business.

Joining us to discuss his company’s wide range of machine-learning analytics solutions is Mobileum’s CTO, Avnish Chauhan.  Besides explaining the value of machine learning in fraud control, Avnish also discusses complementary technologies such as biometrics, voice service analytics, and their applications in fraud, roaming and security.

Dan Baker, Editor, Black Swan: Avnish, what is Mobileum’s value proposition in fraud control?

Avnish Chauhan: Dan, it’s about stopping IRSF by applying an analytic model to the problem.  In many cases we can detect fraud much faster — and that’s a big issue when operators often don’t detect a major IRSF hit for 2 days or more.  Our solution, however, works day and night because it uses machine learning.

In some cases, we achieve zero-day detection, blocking the fraud the first time it’s seen — even before the number range is blacklisted by fraud databases.

While we are talking about IRSF, the underlying analytics and machine learning technology works equally well for domestic revenue share fraud, voice and data bypass, and other related fraud types.

When I spoke to your colleague Jason Lane-Sellers a couple years ago, he briefed me on the great results you achieved for a large customer.  How is that same solution being received in the larger marketplace?

Our aim is to make every customer referenceable, especially the early adopters who proactively help shape the product.  Each such success generates additional opportunities at the same customer, and others.  In this particular case, we have been successful in taking what was built, fine-tuned, and operated over a 3 to 4 year period, and converting it into a packaged solution that is helping another 20 or so operators around the globe, with many more in the pipeline.

The big story is our solution is more efficient and cost-friendly for the operator because the machine does the work rather than a human.  And of course, a machine can work 24 x 7 without asking for a pay raise!

Another factor, frankly, is the fraudsters have gotten smarter and are constantly discovering new ways to fly under the radar, and new ways to commit fraud.  And what our AI/machine learning technology does is bubble up the outliers: it detects the outliers and abnormal behavior — and it does so very objectively, no emotions involved.

Now some will say, “Stopping zero-day attacks may not be necessary.” For instance, Equinox IS tells me many of its U.S. customers lose only a few hundred dollars a year in toll fraud, so the cost of a more robust analytics solution may be not be worth it in all cases.

It depends on the geography and fraud type.  So, if a customer is losing only a few hundred dollars a year in toll fraud, they are very fortunate to either not have fraudsters operating on their network, or to have an effective solution to stop the leakage!

Avnish Chauhan, CTO of Mobilieum

CFCA estimates global fraud losses in tens of billions of dollars.  Hence, it can be a significant problem for some operators.  However, their individual ability to detect and fight it is highly dependent on the staffing of their fraud department.  This is where new AI techniques are coming mainstream and that will significantly move the needle towards more machine-based detection and blocking in the future through industrialized systems.

Take the SIM Box problem, the illegal termination of calls on a mobile network to avoid the operator’s interconnect tariff and government taxes.  Now a consumer doesn’t walk into a retail store and buy 100 SIM cards, but that’s exactly what the SIM Box fraudsters do.  They buy plenty of cards to populate their SIM boxes.  So, the first thing you automate is watching the linkage between IMEI IDs, IMSI, call graphs, location etc. of the SIM cards purchased in bulk.

What’s more, analyzing the voices associated with SIM cards can be quite revealing.  In a normal household, we can expect two or three different family voices using a particular mobile phone.  But on a SIM box, every voice is different because the idea is to rotate the SIM card across hundreds of calls (users).

So without even listening to the conversations — focusing merely on the machine analysis of voice characteristics such as pitch, relative energy, and the space between syllables — you can determine with a 90% or even higher accuracy, whether it is a male voice, female voice, and the same voice.

That’s highly interesting.  So this is biometrics.  And so far, I haven’t heard any other telecom fraud solution vendor talk about this.

Yes, Dan.  Nobody talked about it and nobody has it.  It’s brand new, but we are working with a couple of our operators to get this in place and the early results are promising.

Most solution vendors deal primarily with call detail records and that’s an off-line process whereas Mobileum can provide a better solution and often complement what the operator already has.

It’s in technologies like these where Mobileum is investing on the fraud side.

Mobileum recently announced a partnership with network analytics specialist Allot Communications.  How is Allot helping you?

Basically, we are getting Allot’s help to pre-process or filter certain data from the network.  Mobile networks are very heterogeneous with 2G, 3G, VoLTE, LTE, and so on.  Allot’s SmartVisibility solution filters out certain events which are of no use at a higher order layer, and also enriches the data with some other parameters.

As operators move to all IP networks, and data traffic demand spikes, it is necessary to analyze hundreds of gigabytes per second on the network plane.  Certain KPIs, such as VoLTE call completion rates etc. are necessary for operational purposes, but to address certain other problems, it is necessary to filter out lots of “noise” and hone in on the traffic of interest.

Great, so Allot filters out maybe 90% of the data pipe so you can focus on the meaningful part.

Yes, and the kind of filtering varies by the use case.

In the case of measuring the roaming customer experience, the operator is interested in VoLTE call statistics.  Which of these calls was below a certain level of quality?  Which of them were high quality?  And so on.  Now to do that, you need to bubble up all the VoLTE roaming data as well.

Now, if the use case is fraud or security, maybe the customer is registered on a VoLTE network and you see calls coming on the SS7 network.  That’s not supposed to happen so it’s an issue.

As you know, Dan, we have been active for years in the roaming space.  That’s our genesis, and as we applied behavioral analytics to roaming problems we quickly found that the same kind of analytics applies to the fraud domain, the network security domain, and so on.  So, we expanded and now we widely apply behavioral analytics to these problems.

In the roaming area, for example, an operator typically has programs to offer a roaming subscriber only after he goes to a foreign country and switches on the phone.  But what if — using predictive analytics — you could identify the customers who are likely to travel abroad in the next seven days.  Well, that opens up a nice window of opportunity for the operator to promote its services.

And would you anticipate someone will roam by looking at the SMS messages going back and forth with an airline company?

An SMS message could certainly be one indicator, but more likely we would identify a future roamer by the pattern of communication.  For example, let’s say you call and receive calls from London very frequently.  Well, that establishes you have a strong connection with people in the U.K.  If you received calls during off hours, then we would know it is likely a friend or a family.  If you receive calls over the weekend, the chances of having a friend or family there goes up.

The frequency and duration of the calls is another giveaway.  Business communication could be a very long conversation, in the range of an hour or so, and often start on hour or half-hour boundaries.  Friends and family calls, however, may not be that long, are sporadic in nature, and often on weekends and holidays.

So, all of those things contribute to determining which countries you have a strong connection to.  And then we notice that before travel, there’s often a change of pattern.  So, those are a few of the behavioral factors we use to predict future roaming traffic.

Thanks, Avnish.  I loved hearing about the many kinds of advanced analytics methods you’re using at Mobileum.

Well, we are embedding analytics in everything we do — we’re about using new technologies like voice trends and voice biometrics so we can hopefully stay on the cutting edge of delivering value to our clients.

Maybe a few years back, you could not do biometrics in real-time, but today you can.  So, any company that can quickly adopt such technology can gain certain advantages.

Long and short of it, at Mobileum, we’re busy working some interesting market opportunities in the areas of roaming, security, and counter-fraud.  Fortunately, that’s a happy problem to have.

Copyright 2019 Black Swan Telecom Journal

Avnish Chauhan

Avnish Chauhan

Avnish is Co-founder and Chief Technology Officer of Mobileum, a recognized provider of innovative solutions in Roaming, Security, and Counter-Fraud.

Avnish lives in Silicon Valley, and has been instrumental in defining, building, and launching many “firsts” in the industry.  He studied Computer Science and Electrical Engineering.   Contact Avnish via

Black Swan Solution Guides & Papers

Recent Stories