Putting it simply, Anomaly Detection uses advanced mathematical modelling to accurately detect unexpected changes in activity over time even when the activity is quite ‘up and down’ in nature. For example we may see an hourly or daily pattern that means comparing it with a fixed value or rule would lead to a significant number of false positives.
The Anomaly Detection system effectively uses a curve fitting approach which means the detection rule can become dynamic by tracking the curve and an exception is then based on there being a significant variance at some point in time. For a fuller description of Anomaly Detection please read the first article in this series.
Anomaly Detection is an extremely powerful technique and although the range of its application is very wide the thrust of this article is around performance, quality of service and clients.
Ironing out the Performance Wrinkles
The overall performance of the platform is a major factor in determining the response times to individual transactions. The challenge is that the API operator has relatively little control over how individual agents behave in terms of traffic volumes. An agent with the best of intentions can run a campaign or offer which results in a significant surge in price and availability requests which has the potential to not only impact the platform but also other agents’ service levels and overall bookings. Tracking the load in terms of overall number of searches and bookings is a basic defence against this scenario but the challenge is detecting a genuine surge which is over and above daily traffic peak or indeed fall.
The actual response time experienced by Agents is driven by the culmination of a wide range of factors including ambient platform load, database performance, third party suppliers, and network transit times and so on. Once again there will be natural cycles in response time measurements which need to be discounted whilst any deviations are highlighted. Ideally the response times are tracked at the destination level since this allows the Anomaly Detection to be significantly more specific regarding delays with larger responses or more third parties.
Overall general error levels should be tracked using Anomaly Detection at IT, connection and XML levels and each API platform will have its specific error concerns one of the most critical is timeouts. Most agent system will set a timer when they send a request to an API, if the timer expires before they receive response they will abandon the request and just use the product offers they probably already have from competitors. This is seen as a timeout on the API, the client simply break the connection. Network based analytics tools can detected these timeouts and specifically report on them. Losing the chance to respond to that request is bad enough, but the Agent system usually track timeouts over time and if you are repeatedly too slow (for them) then they will escalate to a cut-off – stop sending search requests all together. The Agent system is necessarily being destination selective so this means that a series of slow requests for a subset of destinations can impact your overall business.
The Clients’ Ups and Downs
The online travel business is a numbers game. Most travel distribution platforms rely on the long tail – a large number of Agents with higher than desirable look-to-book ratios that do make modest but steady bookings that collectively can drive often 50% or more of Total Transaction Value (TTV).
All the factors considered as performance wrinkles can equally be tracked on a per Agent basis. Specifically this means that Anomaly Detection can be used to monitor and highlight any issues in load, response times and errors being generated or experienced by a specific Agent. Anomaly Detection will allow a significant number of clients to be tracked with little human intervention.
Depending upon the role of the clients accessing the API e.g. OTA, it is highly likely they will be handling traveller searches from specific geography/market and as a result their traffic pattern will have a clear daily pattern with super imposed weekly pattern. This is ideal scenario for Anomaly Detection and its capabilities to detect unexpected dips or peaks in traffic and their timing.
Bookings are obviously the end game. Drop offs in bookings are clearly a significant problem and if not preceded by a corresponding drop in searches i.e. the look-to-book ratio has changed then it may be a sign of booking error or too high a price. A casual idea that booking increases are all good news is equally misguided. Sudden increases in bookings, so a change in the book rate, might indicate an inappropriately lowly priced offer. Many travel providers have made major pricing errors e.g. misplaced decimal point only to start offering products at a dramatic loss.
Travel suppliers and distributors can lose a lot if they don’t find and act quickly on business problems or opportunities. Clients can be lost, new sources of revenue missed, or profit margins dented.
Many anomalies are symptomatic of problems that affect (directly or indirectly) a company’s revenue stream. The longer that issues remain unfixed, the more money is being leaked. This is just as true for striking anomalies as it is for more subtle ones. Striking anomalies may cause bigger losses in the short term, but they are also easier to spot because they tend to cause larger deviations in the key metrics being monitored. Subtle anomalies however, can go undetected for longer because they’re harder to pick out in the data. A subtle anomaly that persists for a long time can do just as much damage as a drastic, but brief one.
Success in travel hinges on making the right decisions at the right time. This is where an analytics platform with anomaly detection capabilities can help make that crucial difference. Whether helping companies to chart a course through a storm or take advantage as the tide of business shifts in their favour, quick detection and analysis can enable them to adjust course in time to generate more revenue or avoid losses.
The examples covered here are just some of the many applications of Anomaly Detection in both a technical and business context. This article is design to highlight the opportunities in a platform performance and basic client management context.
The next article in this series will look at Product Availability, Pricing and related business areas associated with Offer Management that form the next step up in deployment of automated technologies such Anomaly Detection.