Lessons learned from artificial intelligence projects

CXN LIVE: Customer Insight Analytics speaker Thierry Derungs, Chief Digital Officer at BNP Paribas Wealth Management explores embracing artificial intelligence to better understand and interact with your customer.

Add bookmark

The best ways you have witnessed AI powered insights improve CX strategies:

Thierry: “There are many ways to work on insights through artificial intelligence. The first, classic, is about the power you can get in analytics thanks to artificial intelligence. By working in new ways to analyse your data, using machine learning in your models, is for sure a very rich area.

“Artificial intelligence [can also provide] richer interactions with your clients 24/7.  So that’s all about the bots, the voice driven interfaces, everything about natural language processing or natural language generation. 

“Last but not least, it’s also what you can get for fraud detection, especially in the very rich digital area which needs trust for the clients, trust for the bank, security and a high capacity to detect fraud. Artificial intelligence is a key asset in that area.”

What sits behind all the hype around AI? What can people tangibly expect from AI and what it can bring to CX initiatives?

Thierry: “Firstly I would like to underline that artificial intelligence is a kind of an old-new story. For a long time companies [have been] working with artificial intelligence – let’s say, heavily with the classic algorithmic approach.  Nevertheless, it has been there for a pretty long time and the hype that we had before was about advisory and robot advisory.

“What has changed a lot in the last few years is the power and the capacity that you can have and the speed of those technologies at an affordable price.

“Before, working with artificial intelligence was something that requested a huge investment.  Today you have many companies that can help you.  The computer power that you need is pretty cheap [too].  

“Bots are also a very well-known domain.  Before it was more script bot, now with artificial intelligence you have much more powerful capacities for those chat bots and again this is something that you can afford pretty easily.  Today, you see [many] new ways of interacting, Alexa, Siri – for example, and artificial intelligence [often] sits behind those new methods. 

“[There is also a lot of hype] around robotic process automation which is focused on gaining efficiency by automating processes. Having artificial intelligence in that area just improves what you can do, you can have more complexity in terms of processes that [were not achievable before].

Reflecting on your personal experience with AI – what are your top lessons learned?

Thierry: “I will start by stating something which could be pretty obvious but [helpful nonetheless]. Artificial intelligence is always starving for data and its hunger is gastronomic. Data quality is compulsory and you also need the best people to ensure that highest quality. 

Data quality – non negotiable

“[The classic phrase is] s*** in = s*** out, but with artificial intelligence it is even stronger because it’s s*** in = total mess out. You really need to understand what your intelligence is doing, especially if you have some machine learning or deep learning. If you cannot be sure that your data at the entry is of the top quality, then understanding what your intelligence is indicating or building as a model will be very difficult.  So, if there is one key lesson, it’s having very high quality input for artificial intelligence.

Proceed with care

“Also in terms of lessons learned – when moving from the classic fraud management with some filters that are highly manual to a behavioral model to detect and prevent fraud cases, of course you need the technologies and the data, but you also need the people.

“Fraud management is a good example because for that kind of project you work with people from the risk team, permanent control, compliance and the regulator.  Of course, they understand the constraints and the global rules to implement, but translating that into new models based on behavior, that’s something you really need to explain to them step-by-step [so they grasp] the new ways of modelling and [the relevant] changes in the organisation.

“For instance, a classic compliance officer will be pretty uncomfortable to rely on artificial intelligence based on client behavior or internal employee [behavior] regarding internal fraud.

“These topics around artificial intelligence are pretty difficult to put inside the classic business and the usual stakeholders, and you have really to take care.

Manage expectations

“You must manage expectations. Many people talk about artificial intelligence without really understanding what’s behind it. It’s a very wide domain – you have image recognition, natural language processing, machine learning – there are a lot of different technologies and [people] can tend to think that [the tech acts as] a black box that will solve their problems.

“But you have to manage these boxes. It cannot be opaque, especially in banks.  You must be able to explain everything, even when using artificial intelligence. You have to make clear distinctions between what is today’s reality and what is just fantasy with artificial intelligence.”

Please name one top mistake to avoid when looking to use AI to power CX strategies?

Thierry: “I would like to underline the effort that you have to spend.  Do not believe that you can buy a black box, put it in your legacy systems and, hocus pocus, it’s done! There’s no magic. 

Be prepared to work

“As for many things, you have to train. Training in artificial intelligence takes time, even if you work with a very good company. For example, there is a lot of hype around Watson – a really amazing system.  Nevertheless, it’s an empty intelligence at the start.  So, even with the most amazing technology you still have to spend time to train it. 

CX Analytics – Bite-sized chunks

“You have to be focused and you have to start small. So, [instead of] ‘too big to fail’, in AI projects, it’s really ‘too big to succeed’.

“You have to try, you have to learn and you will fail for sure at some point.  Retry, learn better and make steps forward. Always small steps and do not plan to [revolutionise your entire] business through one artificial intelligence project. If you try that, I’m pretty sure you will fail.  It’s as simple as that. 

Always understand your artificial intelligence

“You must [always] understand your artificial intelligence. I mentioned the black box – if a company is promising you some magic with a black box, don’t believe them. You are going into a dangerous area because you have really to be in control of the [intelligence]. 

“Perhaps in some years there will be magic available, but today to be realistic you must know from the very first day that you will have to put in time, effort and money to understand what the machine is doing. A mistake to avoid is to be a bit too ‘Uncle Scrooge’ and not put enough money into the people that will work on [your AI projects] and will be in control of what you are doing.”

Save your seat today in Thierry's session: Applying AI Powered Customer Insight to Innovate your CX