• Unlisted

How to overcome the artificial intelligence stumbling block

We don't want to claim to be the ultimate experts in the area – while we are data scientists we’ve been around the block enough to respect there's a good dose of wisdom in the adage that "pride comes before a fall” — however, given the progress that we've made with the unlisted platform, in particular its use of artificial intelligence, we thought we could shed some light on why we have got to where we want to be, and why others might struggle. Firstly, we think there is enough evidence to say that many do struggle to implement successful AI projects. A cursory look of the data available shows that companies themselves recognise that more often than not their AI projects don't achieve what they set out to. Commentators on the industry, such as research firm Gartner, claimed in a 2018 paper that 85% of projects will fail to deliver; of course failure can cover a range of outcomes including lateness, budget problems as well as more substantive issues such as not reaching the desired outcome.

So what are the problems with AI projects, and importantly, what did we do to overcome these challenges?

1. Lack of expertise/understanding of AI AI skills are few and far between, so it is important to ensure that you have the specialist knowledge and talent before you get going.

We were lucky in that not only did we have specific machine learning knowledge, but machine learning knowledge particularly focused on the private company valuation problems (sometimes it’s handy to begin your life as an academic project!) 2. Lack of understanding of the domain and/or the quality of the data available AI cannot magic away the inherent need to understand the area in which you're working and to be able to assess the tools with which you can address your project. Our ability to overcome these hurdles was due to the fact that there exists a large amount of reliable data regarding company valuations, which is our raw material. Additionally, we have vast amounts of data covering the solutions we were looking to provide (existing data on private company valuations) which facilitated the task of training the AI machine, as it knew exactly what it should be looking for. 3. Inability to contain AI within an end-to-end solution/lack of clarity around the project The term AI has been bandied about far too much in recent years, sometimes where there is no discernible AI process present at all. Other times, the AI is there but the engineering is unable to make it function properly within the end-to-end solution. However, the biggest mistake with AI is to believe that it can be used to address all problems. Our approach was to set out a roadmap from the start that showed clearly where the AI process fitted in with our project. Additionally, our AI is focused on a single task (to constantly improve on the accuracy of discrete private company valuations) which means that in building the engine we were able to concentrate and contain our efforts. 4. Lack of trust/lack of transparency This is a problem in how a project relates to its external stakeholders. Failure can come either where stakeholders believe they can do as good if not a better job themselves than the AI, or where potential users are suspicious because they cannot perceive clearly the workings of the AI machine. Our solution has always been to deal openly and honestly with our stakeholders. We do not deny that investment professionals are already able to successfully value private companies – what we do is make the process far more efficient for them, by making it quicker, cheaper and improving on the accuracy through the AI model. And there's no reason for our clients to believe there is a hidden process in a blackbox – we provide our clients with full transparency of all the variables and parameters used to generate our valuations. In summary, our approach to AI and our ability to deliver is based on the right expertise, reliable data to give guidance on inputs and outputs, a clear and limited agenda, and total transparency on how the platform works. As a result we have an engine that allows our clients to produce their private company valuations at a fraction of the price, in real-time, and with ever-increasing accuracy. Now, that's what we consider a successful AI project.