Earnix Blog > AI
Machine Learning for Financial Services: Hype or Reality?
Reuven Shnaps,
March 8, 2018
- AI
There’s an ongoing debate as to whether new trends in machine learning are mere hype or are actually providing tangible business value and helping shape financial services pricing and offering strategies.
A survey of about 200 global insurance professionals conducted by Earnix in 2017 showed that more than half of the respondents are using machine learning technologies in their business. At the same time, only 14% view machine learning as a core strategy that all areas of the company are encouraged to use.
That begs the question, why, despite its relative broad use, is machine learning still not viewed as a core strategy?
During our recent 6th Annual Earnix Summit in London I posed a similar question to an audience of about 150 financial services professionals, trying to see whether our customers and prospects share the same sentiment. The polling results validated the findings: two thirds of the audience said that either machine learning is overhyped or that they see no real tangible value from it.
A recent article in the Harvard Business Review, entitled, “Why you are not getting value from your data science?” also deals with this same question.
One of the main observations in the article is that machine learning experts are often focused on later parts of the pipeline. They also don’t tend to ask the right questions such as “what value does this predictive model provide and how can we measure it?”
It seems as if machine learning experts want to spend time building models and not thinking of how to process huge data sets or translate the business problems into analytical ones. They are too busy tweaking the models and the optional parameters or putting together an ensemble of models, rather than making sure the models’ insights makes sense to business people and can actually be deployed in the market.
Surprisingly, at least for me, a recent survey conducted by Kaggle of nearly 16,000 people around machine learning practices supported this key observation. When asked to specify the barriers in the workplace, among other things, more than 30% mentioned a lack of clear question to answer as a barrier.
So what should financial services companies do to make sure they are deriving tangible value from their machine learning technologies and data science programs? The solution includes four main pillars:
My analytics team is focused on leveraging technology, data and analytics to address business questions and challenges. We accelerated our research and development efforts around machine learning, and developed a solid methodology along with the technology to enable the deployment of machine learning models into the market, and to ensure these models would provide tangible value.
Our customers successfully use our integrated machine learning systems to address business problems such as:
A survey of about 200 global insurance professionals conducted by Earnix in 2017 showed that more than half of the respondents are using machine learning technologies in their business. At the same time, only 14% view machine learning as a core strategy that all areas of the company are encouraged to use.
That begs the question, why, despite its relative broad use, is machine learning still not viewed as a core strategy?
During our recent 6th Annual Earnix Summit in London I posed a similar question to an audience of about 150 financial services professionals, trying to see whether our customers and prospects share the same sentiment. The polling results validated the findings: two thirds of the audience said that either machine learning is overhyped or that they see no real tangible value from it.
A recent article in the Harvard Business Review, entitled, “Why you are not getting value from your data science?” also deals with this same question.
One of the main observations in the article is that machine learning experts are often focused on later parts of the pipeline. They also don’t tend to ask the right questions such as “what value does this predictive model provide and how can we measure it?”
It seems as if machine learning experts want to spend time building models and not thinking of how to process huge data sets or translate the business problems into analytical ones. They are too busy tweaking the models and the optional parameters or putting together an ensemble of models, rather than making sure the models’ insights makes sense to business people and can actually be deployed in the market.
Surprisingly, at least for me, a recent survey conducted by Kaggle of nearly 16,000 people around machine learning practices supported this key observation. When asked to specify the barriers in the workplace, among other things, more than 30% mentioned a lack of clear question to answer as a barrier.
So what should financial services companies do to make sure they are deriving tangible value from their machine learning technologies and data science programs? The solution includes four main pillars:
- Relevant data. If the data has no meaning to business users, it will be of minimal value.
- Solid analytical methodology. The analytics applied to the data needs to be rigorous and make sense.
- Enterprise grade technology. If the IT infrastructure doesn’t support the deployment of machine learning technologies and data science models, the effort will likely fail.
- Deep business knowledge. It’s all about business, not just crunching numbers and spitting out results.
My analytics team is focused on leveraging technology, data and analytics to address business questions and challenges. We accelerated our research and development efforts around machine learning, and developed a solid methodology along with the technology to enable the deployment of machine learning models into the market, and to ensure these models would provide tangible value.
Our customers successfully use our integrated machine learning systems to address business problems such as:
- Rank Optimization
- Price Setting (Eg: Earnix's bank pricing software solution)
- Predicting customer product choice
- Predicting customer utilization and volumes
- Focus on the business problem and seek the right modelling approach. Sometimes it is better to stick to simple models and sometimes there might be a need to combine the traditional models with the new integrated machine learning models.
- Covering more ground and addressing various building blocks with simple models is often better than trying to perfect one piece of the puzzle.
- Don’t always rush to work on massive quantities of data. When you develop your models, it is sometimes better to work on small samples to get business intuition and validate it with business colleagues.
- Always make sure to involve the business people and get their “buy-in” from the inception of the project.
- Invest in acquiring the right technology and automation that can help you deploy the models and business strategy in the market quickly and respond to market changes rapidly.
Share article: