Lately a term that has been thrown around regularly is A.I. or artificial intelligence. Immediately an image of a robotic boy with the name David appears who is trying find his place in the world or if you prefer the doom and gloom version as that of the film Ex Machina where the machine outsmarts the creator and ultimately results in the death of the lesser humans. Rough right? Well, we’re not there quite yet.

Firstly, let’s define artificial intelligence. When I type 7 to the power of 7 into my calculator and hit the equal sign the number 823543 instantaneously pops out. As I see it, that’s quite clever or intelligent rather and as my calculator is created by man, it’s artificial. So artificial intelligence then. But no, not really.
According to John McCarthy Intelligence is the computational part of the ability to achieve goals in the world. An example of this would be for a computer to play chess and decide what the best next move would be. The challenge lies in the fact that the opponent’s next move is indeterminable. Intelligence is therefore needed in the decision making process.

A.I. becomes of great interest where it can be used to aid or even replace white collar jobs. Like the industrial revolution but the jobs at risk are those of the professionals. So back to doom and gloom then but let’s focus on the positive for now. For one, time consuming activities could be done autonomously and/or even improved. Let’s take risk assessment for example: A couple of years ago you had a personal relationship with your banker, let’s call him Bob, and Bob decided if you would get a loan based on how he felt about you (This is not completely true but Bob had a big say with regards to your loan). Bob made some bad character calls and was replaced by fancy tables and some mathematics as things needed to be standardised. So no more Bob then and a thing called credit score becomes important. The new risk assessment method works well but has a lot of room for improvement as it is very robust and slow to react to changes in real life. Here A.I. would be an improvement as large data sets could be analysed to give a more bespoke product. On top of that the program would learn as data is added to the knowledge base causing alterations and adjustments as needed to output. This also frees up time for you to get to the other aspects of your day to day.

This is all very well and is sure to have a great improvement on the way we conduct business but we will need to proceed with caution as creating something that essentially thinks for itself may have complications. Already we are very reliant on Google for the answers we seek and the knowledge of how the mechanics behind something works is no longer of interest as long as we get to the end result. Being too dependent on something is dangerous as consideration has to be given to what would happen if systems fail. We therefore need to fully understand how A.I. works and use it correctly as a tool to improve efficiency rather than a replacement for a professional’s expertise.

We all want to be using models in many areas in our businesses. At Dynamo, when we think about insurance modelling, we classify model use into 9 broad categories, and each of these categories has its own “personality” or a “flavour”. At the one end of the spectrum is the regulatory capital model, a large lumbering elephant, used for infrequent submissions to the regulator. At the other end are business models, used to help management understand implications of changes in the business plan, and these are more a coalition of fleet footed cheetahs. Both of these do what they are meant to do, and they do it really well. It’s just that they’re meant to do different things.

We’ve all had to build the regulatory model. It is a large and wide-ranging beast. It considers as many risks as you can possibly pack into the framework, with estimates or loadings for the things we can’t quantify. It gives a passing nod to materiality, and focuses on a single remote risk metric. Formal, peer reviewed parameterisation processes are in place; expensive and slow validation processes are built; and despair if you want to change the model, because that will cost you a lot of time and money. For a regulatory model, this is necessary. For a business model, this is really unhelpful.

By contrast, the business models we would love to have need to be more nimble, flexible and useful than a regulatory model. They must be sufficiently responsive to keep chief executives’ attention and specifically tailored to the questions the business is asking. If you don’t have all the risks in, that’s probably ok, as long as you have the main ones. We pay far more attention to the events that could cause the business trouble next month and next year, than to a remote risk metric. And most of all, they are engaging and understandable.

The current regulatory framework is forcing us (or at least really strongly encouraging us) to put the elephant and the cheetah (and the 7 other members in our modelling menagerie) into one mythological hybrid, and so we’ve had to build compromises into our “universal model”. In order to make the model tractable for decision making, it does not consider all the risks, interactions and complexities that the regulator would like. But it does not really meet the needs of the business either. The regulatory demands mean the model is unlikely to be nimble enough for real decision making support.

It is this one-size-fits-all requirement that means businesses are not deriving anywhere near the value that they could have from the substantial investment in regulatory models. Any attempt to “extract business value from the regulatory model” is likely to be a difficult task, as we’d be battling to pull down the modelling structures which are necessary to it being a regulatory model. Is it really surprising that models have never lived up to the hype?

In future posts I’ll share some of our thinking around how to solve this, and to actually make modelling useful in business.