Ai Editorial: Machine learning algorithms – are they sharp enough to deliver?

First published on 18th October, 2016

Ai Editorial: The explosion of data, backed up by affordable processing power and storage plus the expertise of data scientists is taking machine learning to a new level, writes Ritesh Gupta  


How accurate a prediction can be that a human wouldn’t be capable of making?

Finding an answer to this question is fascinating indeed, as machine learning algorithms are now more luring than ever.

Before probing into this, I would like to recount one remark from a speaker, who explained the meaning of machine learning during one of our conferences held in Fort Worth, Texas around the same time last year.

“Machine learning is not big data,” he said. The executive even spoke about the significance of being data-driven, right from being clear with what to look for, how to work around it with the help of a proficient team etc.

Today it seems 12 months is a long time.

Today we are talking of sophisticated use cases. Travellers’ experience, be it for planning a holiday or optimizing miles, is being shaped up by smart algorithms. Even combating fraud or revenue leakage is being controlled with the help of algorithms.

An algorithm that essentially is about learning patterns in data (and then foretells similar patterns in new data) can propel the operations of an airline. It is true that machine learning isn’t new – data scientists, analytical software developers, and even use cases have been around for a while. But the explosion of data, backed up by affordable processing power and storage is paving way for outcome that is being deemed as fast and reliable. As it turns out, machine learning often encompasses different types, and simply using one type (predictive analytics) is insufficient.

It is also important to understand that pattern recognition, deep learning and stochastic optimization all have a role to play. Yes, carries have been using algorithms for operational decision-making, such as maximizing the profit on each flight based on the types of fares offered on that specific flight. But new avenues are opening up. For instance, there is a different approach to airline ticket sales auditing leading to revenue protection, featuring knowledge algorithm to manage the complexity and dynamic changes in terms and conditions, flight schedules, reservations, check-ins and fares.  

Here we explore some of the areas in which machine learning algorithms are helping travel organizations:

·          Cross-device targeting: The world of ad tech continues to find an answer to cross-device advertising as travellers’ shopping journey is fragmented. This subject is complicated. Matching the intent of a traveller and delivering a marketing message/ ad with laser-sharp precision, irrespective of the device, location etc. isn’t easy. How well a digital consumer is being identified? Can machine learning understand a traveller’s immediate context and forecast the desired experience of the moment? Single-ID based targeting technologies that rely on probabilistic algorithms have been around for a while. Such type of matching is derived from algorithmically assessing anonymous data points – device type, operating system, location data etc. Yes, a machine learning algorithm can process data and assess if it’s more likely than not that two devices are linked. The outcome is shared in big data records comprising groups of pairs of matched devices. Specialists continue to refine their algorithms by learning what works and what doesn’t. Also, other than probabilistic, there is deterministic or exact match methodology. It is pointed out that travel isn’t similar to retailing, the decision-making is prolonged as there is lot more time to use more devices. So it would be worth following how algorithms play their part as cross-device targeting is still a work in progress.

·          Fraud management: It is being highlighted that airlines can rely on an algorithm –an optimized fraud risk management algorithm – to make decisions designed to optimize sales as much as possible while keeping fraud and chargeback rates under control. Fraud management is going beyond predicting future fraud based on historical data. With pattern recognition, even without any prior historical data, the machine is able to detect patterns across different transactions and diagnose if the transaction exhibited bot behaviour or human behaviour. Using big data, the system collects information from the merchant’s website, such as the user’s web movement behaviour, social media accounts, likes or comments on the website, e-newsletter subscription or alternative payment methods. Combined with pattern recognition, the system draws patterns (for both positive and negative behaviour) to map the DNA profile of the user, and determine if other incoming transactions exhibit the same (fraudulent) behaviour or not. The large quantity of information collected from big data makes it difficult for fraudsters to cover all of their tracks, therefore increasing the effectiveness of preventing fraud.

·          Loyalty: Machine learning is being spoken highly about in the arena of transaction monitoring. For instance, in case of loyalty programs, once a loyalty transaction take place in the blockchain environment —issuance, redemption, or exchange—an algorithm-produced loyalty token emerges. This token is foundation for all sorts of rewards, including points. The availability of this token and distinct identifiers are rationalized on each participant’s ledger. Various online protocol terms administer how points behind these tokens function.  On a basic level, interoperability between programs and partners that use the same dataset to record and transfer value will result in enhanced efficiencies in transaction processing and invoice reconciliation. For travellers, this will result in quicker availability and better usability of their points and miles.

·          Personalisation: Algorithms are being used to analyze and predict customer behavior, sharpening retailing by looking at historical data, buying pattern etc to evaluate propensity to buy. A major aspect is consideration of personal preferences. The blend of right algorithms and customer data can strengthen  personalisation strategies. As Boxever highlighted in a blog positing recently, with greater automation and reliable queries, filters and propensity models, companies can maximize their investment in developing their customer database, extending the value of having a single customer view by making the data actionable.  

 While machine learning, as a concept, always strives to improve, there are times when things can go awry. For instance, algorithms can’t be as meaningless as targeting wrong audience. Also, there are times when one comes across a challenging remark from the proponents of machine learning. For instance, it isn’t a routine task to keep pace with developments in the world of ad tech, analytics, ecommerce, mobile technology etc. Programmatic buying, deep linking, new payments options, progressive web apps, robots, mobile wallets, etc…it seems like a tough ask to manage digital assets and marketing. As one of the senior marketers from Lufthansa told me: It is very challenging to merge all data points, apply the right algorithms and have the right text and visual components come together to create a seamless flow of information to our customers. But yes coming across a relevant message from an advertiser on my chosen device or dealing with fraudsters is a welcome change, and it means machine learning is delivering today.

Interested in machine learning? Hear from senior industry executives at the upcoming 7th Mega Event Worldwide 2016, The Event for Loyalty, Ancillary & Merchandising & Co-Brands, to be held in Toronto, Canada. (25 -26 October, 2016).

Twitter hashtag: #MegaEvent16

Follow Ai on Twitter: @Ai_Connects_Us