In a previous blog post 'Is AI simply a numbers game?' we investigated the relative strengths of Microsoft and IBM in deep learning, based on the hypothesis that future success in depends on having access to uniquely large data sets. Microsoft has been able, whether deliberately or not, to acquire some fantastic assets and thereby claim a strong position in deep learning.
This begs two questions: firstly are there any other organisations which might be able to stake a claim to a place at the top table alongside Microsoft, Google, Amazon and Facebook, and secondly which other organisations have sufficiently valuable data sets that they are acquisition targets?
For the former, we can look at various large vertical sectors. The large industrial manufacturers such as Bosch, GE, Hitachi, Samsung, or Siemens, have increasingly rich data on a variety of processes and increasingly on their machines themselves. GE, in particular, has been beefing up its AI capability in recent years with acquisitions of Bit Stew Systems and Wise.io. Retail is also increasingly data rich, for instance with Walmart holding large volumes of useful data. Finally, financial services organisations, including insurers also hold enormous sets of data. The thing that unites all three of these is that the data is typically focused tightly on the specific use-case of the organisation. There are numerous compelling use case for the application of deep learning to make their operations more effective, but not really for externalising as a broader deep learning offering.
Much more horizontal are the ERP, CRM and enterprise software companies such as Oracle, Salesforce, SAP or Software AG, which have insights on organisational operations which might be more universally monetizable assuming sufficient levels of anonymisation. All of these have the potential to step up.
Another name to throw into the mix, courtesy of some acquisitions, is Verizon. In Yahoo! and AOL it has two assets (now under its Oath arm) which, while they may not be the most dynamic web players, do have very rich data sets in terms of email, IM, web-browsing and so forth. It has also recently acquired a group of logistics companies and can claim a strong position on data in that space. This is in addition, of course, to the data it gathers from its own network operations. Telcos are in an enviable position for deep learning in terms of having rich location data.
In terms of acquisition targets, most enormous data sets will be held by disproportionately expensive organisations. For instance, acquiring an insurance company or a large retailer just to get hold of rich sets of time series data is out of proportion to the value of the raw data. However, there may be some exceptions. Automating investment decisions is a highly lucrative line for deep learning, where small variations in accuracy can justify large lay-outs. The credit rating agencies (S&P, Moody’s and Fitch) might fit the bill. There are also dozens of niches where there are companies holding specialist data sets, for instance media companies holding video which could be used for facial analysis, or mapping agencies (such as the UK’s Ordnance Survey) holding uniquely detailed geophysical data. We expect a flurry of acquisitions in the early 2020s focused on data assets, as well as a number of tactical joint ventures between technology vendors and owners of rich data sets.
There are also tremendous sets of data in the hands of government institutions which, while obviously not acquisition targets in themselves, offer potential rich rewards for anyone who can gain access. Examples would include the UK National Health Service, the Chinese government social credit system, or the Indian Aadhaar system. The NHS, just as one example, has an exhaustive set of healthcare records on almost the entirety of the UK’s 65 million population going back for decades. No wonder Google was keen to license it. Monetising that data set is a potentially highly lucrative opportunity.