Fighting Fraud and Delivering Frictionless Customer Experience in the Contact Centre, Abhinav Anand, Chief Product Officer at Smartnumbers.
Abhinav provides an excellent quantified review of the role Smartnumbers plays in call center fraud and authentication. For inbound calls they are in the signalling path and use over 200 parameters in their machine learning model. Gathering data beyond signalling using behavioural models and fraud consortium data. From that, their model produces a score and reason for that score. Which the call center uses in how the call is answered.
The challenges Smartnumbers addresses are:
- Fraud Prevention (stopping the cheats)
- Telephony fraud. For example, fraudsters use the IVR to validate customer information, such as recent transactions, which is then used to conduct fraud through other channels.
- Card fraud. Catch and release, fraudsters contact the bank to clear blocked fraudulent payments.
- Scams/APP (Authentication Push Payment) fraud. Fraudsters trick their victims into believing that they are their bank. The techniques such as spoofing of emails and SMS messages. Capturing their login information. The are also cross-channel and cross-bank scams, it’s amazing how creating criminals can be in discovering ways to cheat banks systems.
- Caller authentication (helping customers get what they need faster, as well as saving time / costs)
- IVR self-service
- Reduce Average Handling Time
- Improve caller experience
The impact Smartnumbers has on fraud is impressive. Out of all incoming calls, about 0.2% are suspect, and they achieve 50% net new fraud caught by Smartnumbers on top of layered fraud defences.
The size of the UK fraud market, which is likely underreported is 1.8 billion GBP; with card and APP fraud dominating those numbers.
Abhinav also provides some great quantification on the authentication impact, such as IVR containment saving 3.50 GBP per call. That is when the confidence is high on the call, additional IVR options can be provided. And Agent Acceleration of 0.13 GBP per call.
Thank you Abhinav for some great insights and quantification of the impact Smartnumbers has in fraud prevention and caller authentication for its call center customers.
Slideshare is having difficulties today, here’s a direct link to Abhinav’s slides.
Thanks for some great insights and quantification of the impact Smartnumbers has in fraud prevention and caller authentication.
When you expand the data set to include omnichannel, what are the most useful ‘non telecom’ related information to include in your model?
Hi Alan,
Thanks – hope the insights are useful to the community.
The models that underpin Smartnumbers Protect already derive some insight from non-telecom related information, such as frequency and time of the call. When the data set grows to include other channels, similar features from other channels like device id, location, time, frequency, and type of transactions are included in the model. This allows a natural evolution of the model from a call-centric model (i.e. “does this call look genuine or fraudulent”) to a customer-centric model (i.e. “does this customer transaction look genuine or fraudulent”).
Putting this into practice varies by the size and complexity of a customers’ existing fraud and authentication estate.
In smaller customers where Smartnumbers is often the only line of defence, we will ingest the data attributes from other channels along with anonymised customer identifiers and use these to provide this customer-centric model.
Larger and more complex customers such as retail banks will typically own the customer-centric model themselves; Smartnumbers will have a two-way link into the central fraud detection systems (e.g. FICO, Feedzai, Featurespace), feeding out the risk insights from the telephony channel and taking in non-telecoms data (e.g. fraud category, prediction feedback) to improve future predictions.