1. Analytics top priority of enterprises
Analytics is the key enabler of superior customer experiences. Systems of intelligence that run on proprietary data are also the key drivers for future enterprise value, not only for US and Asian platform corporations, but for traditional B2C and B2B enterprises in Europe as well. New trends will not change the list of top skills requirements for marketeers in the near future. Numbers and technology will continue to rule.
2. Data, integration quality and data management issues are still big hurdles
Data management platforms and customer data platforms have become integral parts of Ad tech and Martech stacks. Data management in general, is the single most challenging area to master for marketeers. Many companies are now struggling to integrate Customer Data Platforms (CDPs) and marketing data platforms (DMPs). Vendors promise that technology will make it easier to manage, develop the value of your own data and enrich it with campaign, second, and third-party datasets. Some 78% of organisations either have, or are developing, a customer data platform. CDPs are designed to extract data from internal silos in order to create an enterprise-wide database with a unified customer view. While a CDP is built on data from identified customers, a DMP contains mainly anonymous data on customer behavior from digital channels. DMPs are mostly used to turn third-party data into a proxy for first-party data. In 2019-2020, many expect that the big promises of CDP and DMP integration will be delivered.
3. Automated marketing decisioning skills will make or break companies
Reinforcement loops will be at the core of future models for automated marketing decisions. They are algorithms where each action is associated with a reward (negative or positive). They run on data generated by customer actions in different environments. The models work on an intelligent trial and error basis and the quality of reinforcement learning models will decide the fate of many enterprises in the future. The learning models will succeed or fail based on whether they are able to trigger the right automated actions (such as up-sell, cross-sell, serve or reward) that deliver the best results in terms of experience, revenue, loyalty and retention. The algorithms of company A compete against the algorithms of company B in the global digital marketplace. The fastest and best reinforcement loop will score the best deals.
4. Data Scientists will increasingly become accountability officers for the CMO
The pressure of marketing accountability in digital ecosystems will only increase as investment and complexity continue to rise. It is not good enough to answer the question of “What contributes to customers converting, and by how much?” Marketeers will increasingly need the support of Data Science to be able to answer: “What contributed to this individual customer converting, and by how much?”
5. Zero-party data, sound and video – The new kids on the block
Marketers will start to collect data that is intentionally and proactively directly shared by the consumer. This is zero-party data. This type of data is never inferred through income or device matching, nor is it merely observed through spending behaviours or cookie data. Zero-party data allows brands to build direct relationships with consumers, and in turn, better personalise their marketing efforts, services, offers and product recommendations.
Deep Learning models and best practises open up the doors for new, more complex and information-rich data to be used in modelling applications for marketing. Sound, image and video are types of data which yesterday’s models have had a hard time coping with. Not anymore.
6. AI is becoming easier thanks to new entrants
The gap between AI experimentation and AI large scale adoption can be very hard to bridge. Many new players are addressing this challenge with new off-the-shelf AI solutions. Major marketing clouds come with their special purpose AI embedded (Newton, Einstein). On top of that, many general purpose AI solutions have entered the market quite recently: Peltarion, Valohai, Mesosphere, SageMaker, Dataiku, DataRobot. The challenge will gradually be shifting to figuring out how to apply AI in use cases that bring indisputable value to a business.
7. Platforms will take the Machine Learning lead
New Automatic Machine Learning (AutoML) solutions are beginning to solve many of the most time-consuming tasks of Data Science, such as bringing a sandbox proof of concept model into fully automated prediction models running in production. There are both new startup and enterprise gap-bridging AutoML solutions on the market. The open source community have put in lots of effort to earn pioneer status. In 2019-2020 we will see more players emerge in both worlds, and the role of AutoML solutions will become increasingly critical. Companies utilising ML platforms will see a substantial boost in the number of models deployed. However, custom coding will remain an important factor and a valuable skill.
8. Regulatory impact on business will continue to increase
The data ownership transfer from corporations to individuals has begun. GDPR is only the first large regulatory framework in a series, with more to come. Regulations create overhead on all work related to data. The marketing community needs to stay on their toes for many years ahead with the do’s and the don’ts of data usage.
9. Right to explanation – Trust will demand answers to many new questions
There is still no off-the-shelf definition for marketeers of what “explainable” really implies. Still, fully automated decisions must be explainable. Responsible companies will start providing explanations for how their machine learning algorithm work. As algorithms affect society more, we are entitled to make sure biases are mitigated, and their use is towards the benefit of the whole and not just a few.
This is just one example of what marketers need to spend time and effort on. It is a part of the big quest for developing customer trust, in relation to the data that their companies use in their core business processes. Hygiene levels need to be constantly re-evaluated. Here are some other questions that need company specific answers: What is trust in the context of machine learning? If we claim that a machine model is fair, what are its characteristics? Can we truly say that data is anonymous?
10. The granularity of Data Science job descriptions is on the rise
One or two years ago, companies would just publish a job vacancy as “Data Scientist”. This is starting to feel incomplete. Here are the present nine Data Science related job titles at Netflix: Business analyst, Data analyst, Quantitative analyst, Algorithm engineer, Analytics engineer, Data engineer, Data scientist, Machine Learning Scientist and Research Scientist.
11. Self-service Data Science in da House
The demand for data engineers is far greater than supply. No automatic machine learning will happen if the data is not in order. Citizen, i.e. amateur, self-learned practicians are bound to join the game. One of the tasks of expert data scientists will therefore be to oversee these citizen counterparts. Many hard-core data scientists will resist this. But, it happened decades ago to finance with Excel and in BI with tools like Tableau. Dataiku, DataRobot, are new tools for everyone interested in big sets of numbers. A fool with a tool might surprise many of us.
Ola Ottosson, Principal Analytics, Emma Storbacka, CEO, Tuukka Vakeasuo Principal Tech, Harri Tumanoff, Chief Data Officer