How Etihad Enhanced its Financial Controls using Microsoft AI

Etihad Airways has successfully deployed Artificial Intelligence (AI) to identify erroneous payments to suppliers. In turn, this has saved the finance team precious time by reducing false positives, while correctly identifying payment anomalies that could have been missed manually. Bhavesh Shivshanker, Head of Financial Operations, Etihad Aviation Group, gives TMI the inside story.

Having set up an internal Innovation Committee, the Finance function at UAE flag carrier Etihad Airways was keen to explore how technology could be applied to improve the way it operated. The aim of this exploration was to drive new and perhaps unexpected efficiencies, ranging from relieving heavy workloads through automation to improving decision making using advanced analytics. One of the pain points identified was around transaction anomaly detection, as Shivshanker explains.

“Most organisations will have questioned what more could be done to strengthen controls across their payments processes. Given our transactional volume and spend, we saw this as an area where automation and AI could really add value.”

Getting visibility on the problem

The first step in tacking this issue was to build a data profile of any overpayments from the past that had been identified. To accomplish this, the Etihad team created a central database that detailed these payments, what had been collected back and what was still out there.

Shivshanker recalls: “Our data sat across various internal teams; some with Audit, some with Business teams, some with Legal and some within Finance. The first step was to centralise and standardise. This helped us understand where things had fallen over in the past and ensure we had closed those operational gaps. It also gave us a lot of transactional training data, which we later found to be a critical step in building AI supported controls”.

Deploying artificial intelligence

In search of an efficient and elegant solution, Etihad turned to one of its trusted technology partners, Microsoft. The team explained the issue to Microsoft, described the work they had already undertaken on centralising the payments data, and asked if there was a way that AI could provide a smart solution. Fortunately for Etihad, this proved to be the case.

Microsoft brought its partner Predica into the project, and together with Etihad they mapped out the existing business challenge and manual business processes. Predica is the developer that would go on to build the AI algorithm for Etihad and champion the proof of concept.

Shivshanker recalls “We quickly discovered that whilst there are many off-the-shelf tools available in the market, these often generated too many false positives i.e. stopped healthy transaction flows or were linked to specific banking platforms. Using such solutions would potentially prevent us from making real-time decisions or leave us with coverage gaps across our multiple banking partners. Leveraging our great partnership with Microsoft and Predica, we looked to an in-house solution which could mitigate some of these issues.”

“The project got off to a good start thanks to the work that Etihad had already done in collating the data,” Shivshanker notes. “With this type of AI, you need a set of data examples which can be used to train the algorithm. Using those examples, the AI can learn trends or markers in the data and then identify similar transactions from a much larger dataset. It does this very quickly, far faster and more accurately than a human could.”

The project that the partners decided upon was a proof of concept, which was deliberately restricted to a couple of years of data so it could be quick and agile. The training data set was given to the AI, which then looked through Etihad’s enterprise resource planning (ERP) system,  and flagged any of a million plus transactions in there that looked very similar to the ‘bad outcomes’ in the training data set.

Shivshanker recalls: “To get accurate results, it’s important to also provide the AI feedback. This is an iterative process whereby the AI proposes a set of results i.e. it thinks certain transactions look out of the ordinary and then a human validates. Once the AI has the validation input, it automatically refines its identification criteria. Whilst this initially takes some time, it is a powerful feedback mechanism and the workload quickly drops off as the AI learns”.

Building on results

Following the two-month proof of concept period, the Etihad Finance team was delighted with the results. Shivshanker explains: “The project results went above and beyond what we had expected. Within two months, the AI was able to identify outlier transactions which, using simple rules-based detection methods, may have slipped through the net. We are also saving 600+ hours per year thanks to a significant reduction in false positives and the corresponding review workload”.

Following the proof of concept, Etihad plans to move into production and run the analysis before the payments run every single day. This will see a shift away from a detective control environment to a preventative one. Shivshanker elaborates: “The AI looks at each transaction, goes back through the ERP to see if it looks like an anomaly and provides close to real-time feedback. This gives us one more layer of control before releasing payments”

Next steps

With the success of AI in tackling the issue of anomalous payments, Etihad is keen to pinpoint further use cases for advanced technology in finance. One such issue Etihad is exploring is the time-old challenge of cash flow forecasting.

“This is a fairly standard use case, but Etihad approached the POC in a novel way,” Shivshanker explains. “We organised a mini-hackathon whereby historical cashflow data was shared with a group of participants, ranging from established market leaders to Fintech start-ups. We asked each to produce a forecast using AI, which was compared against actuals. This not only allowed us to quickly see whether AI could improve forecast accuracy but also compare the capability of the different Data Science teams. The results so far have been extremely promising.”

Etihad is also exploring how AI can be applied in the bank reconciliation space, highlighting how many areas in Treasury and Finance can benefit from smart automation.

“These are very generic business problems and I’m sure, in the near future, every organisation will be looking to AI as a solution,” Shivshanker concludes.