In a decade that promised ‘Back to the Future’ style pizza ovens and flying cars, Panasonic unveiled a concept smart fridge that could anticipate and predict consumer behaviours and then reorder basic items based on the speed of consumption. It seems we are still some way off this becoming an everyday reality for consumers but could machine learning be the key to delivering on the promise of ‘smart’ living and consumption?
The ‘just-in-time’ supply chain model employed by supermarkets was tested, and at times broken, during the COVID-19 pandemic. Panic buying and stockpiling led to shortages of a range of goods with the lack of toilet paper making the headlines in many countries. This got me wondering whether emerging technology could deliver the capability to predict the items customers will likely require based on a number of factors, such as the time since the last order, previously ordered products and taste profiles?
Where smart fridges promise reordering capabilities for chilled items only, it is not too much of a stretch to imagine, in the near future, a world where all manner of products and goods will be delivered on a ‘just-in-time’ model to your front door step, rather than the local supermarket.
Moore Machine Learning
With 90% of the world’s data having been harvested over the past two years alone¹, it is clear that the accumulation of data is only going to accelerate. The increase in data generation has encouraged tech firms, and traditional companies alike, to shift their focus from manufacturing chipsets, smartphones, and appliances, to leveraging data to understand and anticipate consumer needs.
The companies who have engineered approaches to predict future trends by leveraging historical data are swiftly becoming the most valuable organisations in the world. I am sure this trend will continue; data is the newest commodity on the block.
The proliferation of machine learning technology in recent years has been possible because of the exponential growth in computing power (Moore’s law) combined with the shift to the cloud computing. The huge growth in processing power and storage capabilities have encouraged a dramatic increase in the harvesting and processing of metadata. If data is something basic such as what product an individual orders, metadata can be explained as the related data that describes the data such as:
- Date of purchase;
- IP address; and
- Payment method used.
Evidently, each data point can generate a near-infinite degree of metadata but viewing these in isolation may not be hugely insightful. How about if we were to combine multiple data points with other sources of information, such as weather forecasts, seasonal trends, and other consumer purchasing history? This approach could provide the foundation for a consumer behavioural engine and this is where machine learning becomes important.
Machine learning technology is still in its infancy. Nonetheless, companies are discovering that determining meaning from data provides a significant commercial edge.
The ability to process huge amounts of data (and metadata) to answer seemingly simple questions such as seasonal supply and demand for a given item makes it possible to identify products that individuals want and are likely to run out of in the near future. It is already technically feasible to achieve this for a small number of dependent variables. But technology cannot yet cope with the same huge number of variables that that humans currently take into account in decision-making.
Machine learning is taking huge strides towards “conscious thought” and could soon deliver the power required to develop a consumer “brain” to make decisions on behalf of individuals. In the next few years, innovative warehouse hubs, such as Ocado, could possess enough consumer behavioural data to order and deliver virtually any item to the doorstep without any human interaction. Machine Learning is here stay and those who do not embrace the opportunities it provides will be left behind with the, once smart, kitchen appliances.
Machine learning and R&D; Why you need to be careful and how we can help you
While machine Learning is considered to be a relatively new field of science and technology, assessing and qualifying Machine Learning projects for R&D tax purposes is rather intricate. Not all machine learning related activities directly address a technological uncertainty and so can fall outside the bounds of a Research and Development project, for tax purposes. Examples of ineligible activity may include the assessment of existing off-the-shelf libraries, fine tuning of models, aggregation of data, and the learning approach that is the most optimal.
We are specialists in assisting with these distinctions and have a team of 75 individuals comprising engineers as well as auditors and tax specialists. We were one of the first UK accountancy practices to deliver benefits such as engineer-to-engineer discussions because we were the first to employ multi-disciplined engineers. This has resulted in better understanding of the R&D tax definitions and an established ‘HMRC agreed’ process.
For more advice and guidance on the tax treatment of the novel technology you are engineering please contact Ash Rawson.
1.IBM.com - Big Data: Big Challenge or Big Opportunity