How to implement artificial intelligence in your business
We all know artificial intelligence (AI) is the next big thing in business. But what is the best way to start its implementation in your company? Which essential building blocks are required? How do you set targets and create a strategy for achieving it? How can you effectively gauge possible quick-wins and transformational potential?
These are just some of the myriad questions executives need to table – and of course answer – in order to proactively define a value-creating AI strategy. While the answers are not easy to come by, managers do not have to face these issues alone. Along with third-party AI development companies, Roland Berger consultants execute high-impact projects to specifically address all of these issues. This is achieved through an assessment of your companies' AI fitness and readiness levels, identifying opportunities, and drafting a roadmap for achieving your goals.
Andrew Hanff, financial services partner in Roland Berger's Montreal office, is one of the firm's global experts. In this interview, he discusses his experiences with various projects, pitfalls to avoid and the expertise of the Terra Numerata network.
- Andrew, you have worked with several clients on the subject of artificial intelligence. We would appreciate if you could share some of your experiences with us – especially regarding the fundamental question of how you have helped companies "get down to business" with AI. Do companies feel threatened by artificial intelligence or do they appreciate the opportunities it affords them - what is your prevailing impression?
We're experiencing a range of reactions across organizations. Some clearly see it as an opportunity – particularly those with a higher level of maturity regarding data management, governance and quality. These companies have already been experimenting to a certain degree with advanced analytics, and subsequently view AI as the natural evolutionary next step. This group includes, for example, many banking and insurance firms, which are capitalizing on prior years' efforts to digitalize end-to-end-processes. They have also exhibited adeptness in applying advanced analytics across their various value chains, such as loan adjudication and pricing, policy risk determination and cross-sell migration trends. Their broader functions have integrated data scientists, Ph.D. engineers, and they master the basics of languages such as Spark, Hadoop and 'R'. Such players have clearly made a "psychological shift" in recognition of the value of information and its manipulation, versus that of their balance sheet per se. They clearly see this as an opportunity. At the other end of the spectrum, some smaller institutions, and other sectors in general, may still be struggling with workflow digitalization, and appreciation of rigorous data discipline. In these cases, AI is less understood or viewed as another potential pressure point on their business and delivery models.
- What is the most critical managerial pitfall to avoid?
Overestimating the pace and scope of adoption that is realistically achievable, as well as the results which are 'guaranteed'. Globally, there is a lot of 'buzz' around this topic, as well as misinformation regarding how fast AI applications can be up and running in order to deliver results. These efforts should not be confused with low-impact automation like RPA or the optimization of traditional "develop-biased" programming. Managers often think they will get it done quickly and broadly across the organization. Yet they will soon find it involves a rigorous application of various techniques – such as search and optimization, constraint satisfaction, probabilistic reasoning – across a set of highly discrete problem sets. This is why one needs to be very astute in defining where to concentrate efforts. Managers soon discover there is a very significant research and development component to this. You do not know a priori what the results are in terms of the relevance of AI for a given business problem. For instance, we have been tracking the returns on AI-enhanced versus 'human-managed' active ETF funds and to date the latter group has not shown differentiation.
- That sounds rather disappointing…
Perhaps, but that sentiment could be completely foregone if managers understand that AI is not a hammer looking for a nail. One needs to first understand with high precision what your business problems are across the enterprise's value chain, and subsequently evaluate whether AI likely provides the right solution, or whether you need to either consider other disruptive technologies, or simply tweaked ways of working. That's why in some cases we're actually discouraging our clients from concentrating their efforts in certain areas of their company where we feel that traditional advanced analytics or human problem solving is best applied for those problems. They need to concentrate AI in the areas where it makes the most sense from the perspective of data availability and accuracy, what's at stake, and very importantly – relation causality. From a decision or reasoning standpoint, in an accretive manner AI will help describe the relationship between inputs and outputs, enabling firms to make better future – and likely automated – decisions. This works well in situations where causality exists, but not in situations where it turns out be random – such as certain types of market movements). It requires a pointed effort to make sure that companies get the best bang for their buck.
- How do you support your clients in choosing the right solution for their specific needs?
By creating a programmatic view of AI and how it can be applied to a client's specifics. We also actively evaluate likely impacts, implementation and challenges to success, which permits an enterprise-wide triangulation to a set of prioritized applications. This often results in a roadmap which encompasses a set of considerations beyond only AI. You would be perhaps surprised to learn how critical the analysis of challenges is, in comparison to the actual prioritization of potential applications. We often uncover a need for remediation related to data and workflow management, as well as components of the IT infrastructure and architecture. There are important corollary benefits to this as we are able to uncover a whole series of areas in which getting ready for AI will help clients with other aspects of their business that might not necessarily be initially impacted directly by AI.
- How do you actually measure a client's AI readiness?
There are over 120 variables that we exhaustively review, seen through the context of both the company's business as well as their overarching AI ambitions. These fall into three main buckets. Firstly, there is organizational readiness, secondly data, workflow and technology fundamentals need to be analyzed, and finally specific AI deployment characteristics must be taken into account. Our analysis covers the range from highly technical considerations related to software and hardware, all the way to cultural considerations such as tolerance for failure. Importantly, we perform the readiness assessment in the context of the target applications. This ensures that the definition of a minimum viability threshold is in sync with our clients' requirements.
For this part of the exercise, we eschew an outside-in approach. Our teams are very much "down in the weeds", working side-by-side with their client counterparts. This phase consists of extensive on-site and documentation reviews, and is managed as a highly interactive and participative effort together with client stakeholders. We really leverage this opportunity to federate new paradigms and drive alignment, and build consensus on the transformation and roadmap forward. In parallel, supporting the thought process of the board, CEO and executives regarding how AI fits into a general disruptive technology program, is also a key part of what we do.
- You just mentioned the CEO and the board. Do you feel there is enough interest and attention from their side regarding AI?
There is a lot of interest, but there isn't necessarily all the expertise, and less so experience, required to place AI in the context of overall enterprise plans. Nevertheless, the tone has to be set at the top. CEOs need to drive a forward-looking mindset, openness to innovation and experimentation, a paradigmatic shift from incremental to wholesale change that might be afforded by AI, the setting of an agile organizational culture, a tolerance for failure, and so on. Often executives don't necessarily understand all the direct and peripheral implications arising from AI, implementation key success factors, and the way forward. This is where I think consultants add significant value. We leverage direct AI expertise and deep business understanding to help them navigate this arena.
- What about any missing expertise? How can you remedy this?
During these mandates, we leverage our Terra Numerata network. Via this platform we partner with specialized providers and data scientists who are deeply experienced in machine learning. The strength of our co-delivery on specific projects is underpinned by the complementarity of our expertise. Following the strategic opportunity prioritization and readiness assessment, we channel the issues related to AI – or other disruptive technology, and can subsequently maintain close proximity (for example project management and interfacing with third-party developers) to ensure that development moves forward in a way that is attuned to business requirements. Hence we offer end-to-end assistance, performing the front-end strategic and readiness assessment by leveraging our deep understanding of different sectors, and continue our involvement as necessary until the client has achieved direct value benefit.