Thank you for Subscribing to CIO Applications Weekly Brief
From Prediction to Foreseeing and Shaping the Future: The Required Evolutions in Machine Learnin
Sunando Das, CMI Director, Predictive Marketing and Retail Analytics, Unilever
Given the changing dynamics, the marketing questions are where the future growth will come from, how to sweat the investments more, how to predict the future consumer behaviours to drive sales. Machine Learning (ML) has been instrumental in addressing these challenges, driving business impact, and bringing certainty to uncertainty which I have spoken about earlier.
The business impact of these applications will be enhanced with evolutions in ML. This article outlines these eight likely ML evolutions in the next 12-18 months.
1. From Predicting the Future of a Known Past to that of an Unknown Past: Whilst the uncertainties imposed by COVID in predicting the future consumer demand has been addressed by leveraging ML, what remains unresolved is the prediction by consumer segments where sales data does not exist. Solving this challenge will help to move from predicting the consumer demand to shaping the future consumer demand. Evolutions in neural network and game theory applications are helping address this challenge.
2. From Prediction to Shaping Future Sales: Consumer Lifetime Value of categories has reduced significantly in FMCG (Fast Moving Consumer Goods) since the onset of COVID. There are categories with significantly increased consumption in 2020 but the rate of increase will decelerate in 2021. The focus will be on predicting and identifying consumers who are likely to change their future consumption patterns to shape their behaviour rather than looking at past or present behaviour as proxies of future behaviour. This is where evolutions in ML models will be critical.
3. Transferability of Findings: Deployment of any ML application across the business requires significant budget and resources. Hence, ML capabilities to learn from a defined set of markets, categories to extrapolate to the rest of the business will gain relevance. This capability has been in existence and has been applied for several applications over the years – what will evolve is the scale across all ML applications.
Convergence: skill sets, business applications
Machine Learning (ML) has been instrumental in addressing these challenges, driving business impact, and bringing certainty to uncertainty which I have spoken about earlier
There has been resurgence in the budget optimisation applications where econometrics plays a huge role. However, econometrics has challenges that are overcome by using a combination of ML models to drive higher precision, consistency, and granularity. Getting skilled econometricians to apply ML models without losing the past learning will gain more prominence in the coming months as budget optimisation applications move towards the platform as a service solution to accelerate scalability.
5. From Continuous Updates to a Dynamic Transfer Learning Closed-Loop System:
It is a common practice to continually refine the models with new in-market data to bridge the gap between in-market and validated predictions. However, with the quantum of deviations from the expected trends since the emergence of COVID, the need for a self-correcting closed loop learning system has become more important. This evolution will require coordination across the multiple models from the different functional units (example – Marketing, Supply Chain, Finance, R&D) where the outputs from one model will help course-correct the inputs of another model. This will also help drive the informal co-operation between the different functional units for ML applications.
6. The Convergence of Personalisation with Strategic Market Measurement Models: The two sides of measurement models are top-down Marketing Mix Models (MMM) and bottom-up campaign optimisation attribution models. MMMs optimise spends in different marketing levers for strategic planning and drive the ROI (return on investments). Campaign optimisation attribution models enable better personalisation to drive campaign ROI. The two areas are inter-linked but rarely get integrated as a seamless system of each feeding into the other continually. This convergence will be enabled as strategic MMMs move from aggregate measurement to consumer segment level measurement models made possible by data sharing eco-system (between manufacturers and retailers)and segment-specific ML models with sparse segment level data.
360 Consumer View: New sources of data and data enrichment
7. Enriching Big Data by Learning from Small Data: The ability to project small data onto big data, via propensity models, will gain more prominence to increase the depth of first-party datasets. This will help drive 360 view of consumers for a better basis of activation. This will also help with the emerging era of going beyond identity matching especially with third-party data assets, given the likely challenges in the future.
8. New Sources of Passive Behavioural Data: Passive internet-of-things consumer data that has not been leveraged for broader applications at scale will gain prominence but with consumer consent and exchange for agreed applications. This will vary from energy meter datasets to smart machine data (such as coffee machines, washing machines) which will drive new sources of consumer engagement with higher targeted reach and lower cost. Harnessing these datasets will require an ecosystem of data collaboration which may not exist today and the integration of different types of ML models (energy data models and consumer propensity models).
As the above ML evolutions gain more relevance and drive business impact in organisations, the boundaries will be pushed continually powered by our imagination and advancements in ML capabilities.
At the heart of these evolutions is the ability to learn and predict the consumer decision-making process. Without having adequate knowledge of the consumer decision-making process, can a data scientist alone drive such evolutions and applications? Hence, the key task for organizations is to achieve the knowledge convergence of data scientists and consumer insights experts. While data scientists need to think in terms of the consumer decision-making process, the consumer insights experts need to be conversant with the abilities of ML.