These days it has become common sense that Artificial Intelligence (AI) can deliver business value to organizations and their customers.
However, many organizations face difficulties in achieving the expected results of an AI strategy. Companies feel the need to be part of the ‘AI game’ but find it difficult to execute and derive tangible business and social impact.
This is not surprising as AI is a broad and complex field involving complex technologies, complex mathematical and statistical methods and last but not least new ways of working and thinking.
I had the privilege to speak with Doron Reuter and asked him about ING’s approach towards AI. Doron is responsible for AI Products & Partnerships within ING’s Advanced Analytics (AA) tribe of Wholesale Banking (WB).
Hopefully, this blog helps you with finding the right approach, when it comes to setting up or improving your AI department.
Topics of this interview:
· How to get started with AI?
· What do you do with AI?
· What do you need to build an AI product?
Before diving in, what are the origins of the AA tribe of WB within ING?
In 2014 ING announced the Think Forward Strategy with the purpose to empower people to stay a step ahead in life and in business. One of the pillars of this strategy is to use Analytics to better understand our customer’s needs and continuously improve the way we empower them. As a result of Think Forward, ING set up a team to research what big data meant for WB, resulting in the AA tribe.
After doing research, the team came to the conclusion that within the Wholesale Bank, we would be able to deliver most value by developing end to end software products with embedded machine learning algorithms. This approach enables our clients and users to continuously apply valuable insights to augment their work. For us it means we can continuously enhance the insights that we delivered through our products.
Since 2014, ING’s AA tribe has grown to 83 professionals in 4 countries comprising a state of the art custom data analytics platform, data engineers, data scientists, software engineers, ux designers and product managers, all working to delivering products.
Some examples of successful AI use-cases within ING:
· Early Warning System: Using AI to assess credit risk
· Katana: AI tool sharpens investor decisions on bonds
· Axyon: AI boost for syndicated loans market
If you are interested in reading more about the core activities of ING Wholesale Banking Advanced Analytics, you can find an overview here.
One of the success factors in building AI products is an agile way of working in multi-skilled self-steering teams so that you can maximize execution speed, flexibility and the development of craftsmanship within the team. ING is well-known for its agile organizational structure – referring to the famous Spotify model. However, Doron determines that there are also a number of additional aspects to consider when setting up an AI department and developing AI products .
So you want to develop AI products, but where do you start?
Firstly, you need to strike the right balance between top down strategy and bottom up initiative.
Without Think Forward it would be much more difficult to get organizational momentum for an AI department. However, you also need people in the business who are dealing with clients on a day to day basis to dedicate the time to working out the most valuable applications of AI in their client interactions. A strong mandate from the top will give you access to funding, the platform to talk about the value of AI and access to key people in the business. But it is crucial in your “evangelization” of AI that you link the realistic possibilities of AI very specifically to the business context of your people and clients. It is also crucial that you team up with your business and client counterparts whose knowledge is important to articulating what to build as you will need to make choices and you want to build the most valuable AI products.
What do you do with AI?
Assuming your objective is to use the possibilities of AI to maximize the value to your clients then it is important to consider: (1) the time horizon of your innovation i.e. do you want to incrementally improve the current business or do you want to develop a completely new future proof business model and (2) the interaction of user problems and business problems.
When trying to create value with AI it is easier and less costly to take an existing process, tech platform or software product and improve that by integrating an AI component (for e.g. automated actions on the basis of insights to reduce churn, increase up and cross-buy or price optimally) as opposed to building a completely new process, tech platform or software product around a core, or a number of core AI algorithms (e.g. a new trade platform based on AI driven supply demand matching or a new financing platform based on AI driven on-boarding, risk assessment and disbursement). An organization needs to make its own cost, benefit analysis and decide what the optimal mix of near term incremental optimisations and new business models is most desirable.
Another important aspect to consider is how you solve both a business and user problem. If your client, a CFO, is trying to reduce losses due to unpaid invoices then you will only achieve that risk reduction if the operators in the receivables department actually uses the AI product (which may provide a prediction of invoices that are at high risk of default) to better manage these high risk invoices. Solving a user’s problem is very much a design endeavour and in some cases may be completely different to the business problem. In the invoice example, a user in the receivables department might be completely overloaded in administrative processes which prohibit him from giving the right attention to high risk invoices, so unless you solve that administrative burden as well in your product, it will never get used to solve the business problem of reducing losses.
Once you have identified a number of potential user problems that will ultimately solve a business problem that matters it is important that you fully understand and prioritise the user problem that is most acute and define the most simple prototype possible that will prove that you can solve the problem using AI. You want to involve your user base very early on and work with them in the process. When you are satisfied that you are solving the right problem you can assess whether the business case actually works and the investment is worth the outcome.
What do you need to build an AI product?
A key pre-requisite to building an AI product is data. AI products as we refer to them today are driven by Machine Learning. In order to train Machine Learning models effectively, you need to have sufficient data. Apart from data you will of course need the technology and tooling to work with this data but in the first phases of the project this is not usually the challenging part. Of course you also need the right team, which will very much depend on the type of product you want to build but will almost always include a Data Scientist who can explore the data and develop the core Machine Learning models, a User Experience Designer who can really discover & articulate the most critical user problems, and a Product Manager who can link the process to the business problem and necessary client and organizational stakeholders. The team mind-set should be open and entrepreneurial, working exhaustively to prove their hypotheses wrong! Later when you are starting to scale the product then you will need to consider more data and software engineering capabilities.
Building AI products is not simple and involves a lot of hard, and in some cases very operational work. However, you get to work with cutting edge tech and science, think creatively about how you can impact the world and inspire new ways of working. Above all, you get towork with brilliant and in most cases humble technologists whose inspiration lies in changing the world we live in for the better!
Ben je op zoek naar advies? Klik hier om in gesprek te gaan.