Event attendees were asked to fill out a short survey about the current status of their firm's data science effort and how they see it changing in 2017. A short summary is provided on the right.

We may be seeing bifurcation: those who are behind keep falling further behind because they under-invest while those who are already ahead keep widening their lead.

Starting small

Several firms looking to establish data science teams have been pursuing top-down build-outs: a senior strategist designs an overall architecture, and it is then realised through aggressive hiring.

The panellists were broadly in agreement that it is much more effective to build out a data team bottom-up. The effort should begin with a few small projects and the firm should then scale into the ones they deem most successful. Even as the build-out continues, it is important not to overcommit prematurely to any particular initiative - given how new the space is, it is hard to know what will and won't work out.

This also applies to data sourcing. It is easier to justify purchasing complex and expensive datasets once you have a record of extracting value from more readily available data.

Data sourcing

There are a few areas our panellists saw as particularly promising for data sourcing:

• Marketing data – A lot of vendors traditionally catering their data offerings to marketers can redeploy their assets and sell into the buy-side as well.

• Shipping – While datasets have been available in this space for some time, we are now able to get much better insight into what is being transported.

• Internet of Things (IoT) – Anything from farmland to oil rigs is going to be putting off exhaust data as connected sensors become more common.

Note that discussion around alternative data tends to overemphasize the discovery of new datasets as that is where vendors get their incremental revenue.

Extracting value

Several multi-manager funds and a couple of vendors have been building out data platforms for use across multiple investment teams. While these efforts can be helpful for nowcasting (tracking) revenues, big data’s real value lies in being able to replicate the CEO’s dashboard – and different investment teams can have very different ideas on what that dashboard should look like.

One solution several firms have explored is pull-based research. Analysts and PMs tell data science what it is they want to understand and the data scientists come back with answers. The problem here is that traditional investors often don’t know what kinds of questions to ask. Conversely, data scientists possess a wealth of information but have no idea what really matters to analysts.

The solution is to make sure at least one person in your firm can bridge fundamental research and data science. They should be able to work closely with analysts to understand their research process but also interface with data scientists to understand the possibilities and limitations with each dataset.