Adding machine learning to the contact centre
There has never been a better time to start using machine learning. In the last year, the momentum behind data science and its connected fields (artificial intelligence, big data, data analytics) has grown from strength to strength. We are now seeing a record number of companies developing AI-based solutions for a wide range of applications, from self-driving cars and personal assistants to photo editors and chatbots.
Perhaps the biggest catalysts spurring the rise of AI have been the democratisation of machine learning and the affordability of the cloud. Up until two or three years ago though, the situation was dramatically different: companies looking to add artificial intelligence to their products needed to make a sizeable financial investment and recruit a large team of in-house data scientists.
Now, the emergence of the cloud has made it easier for anyone to develop and deploy AI models in a cost-efficient manner. One company making the case for AI is Google who recently announced that its Cloud Machine Learning service would be available to other businesses in public beta. Before the announcement went live, Ocado was among the few companies that could try out the Google Cloud Machine Learning platform in private alpha. By using Google’s machine learning service, we were able to train our models at a much faster rate, and integrate them with the already widely-used Google Cloud platform.
In addition to the Cloud Machine Learning service, Google has played an integral part in the development of TensorFlow – an open source library for machine intelligence. A major feature of TensorFlow is its flexible architecture which allows developers to deploy models across desktops, servers, or even mobile devices using a single API. This makes it easier for smaller teams or indie developers to experiment with machine learning on a laptop device or even a tablet before scaling up to more expensive hardware – all without making any changes to the original code base.
Ocado Technology used TensorFlow to enhance its contact centre with a machine learning algorithm that would tag and categorise customer emails according to their priority. Unlike other machine learning projects, we began this venture to solve a real world problem, rather than as a purely scientific endeavour.
We used our learnings from adopting the Google Cloud Platform for customer analytics to form the project structure and reduce the delay between the initial proof of concept to the end product.
The other unique aspect of this project was the level of involvement from various parts of the business, from the head of service delivery to our CTO. On a personal note, I can definitely state that Ocado is perhaps one of the few companies where these cutting-edge collaborations involve not just different departments but also everyone from the CTO to our graduate engineers.
We decided that the product owner for the contact centre should manage the project, working closely with the data science team and the software development team. This joined up thinking approach helped in many ways:
- The data learning curve was reduced as data producers were aligned
- The data science team had quick access to the operational team so the validation of models could happen quickly
By defining a clear path to run the project in parallel upfront, we were able to create the final product very quickly, including defining the required metrics. Secondly, the data curation team was engaged early, therefore helping protect customer sensitive data and defining the storage requirements but without losing the meaning or value of the dataset needed for training the neural net.
Finally, adopting a phased and iterative approach allowed the business to measure the value of the project at each stage and decide whether to progress further, rather than ‘pay big’ upfront.
Ocado Technology has a proud tradition of using artificial intelligence to improve many of its existing systems and platforms. We have been working closely with Google and Amazon AWS to create innovative solutions that combine the best practices of AI with our own technical know-how and experience in the grocery retail market.
If you’re looking for the opportunity to shape the future of online retail, I definitely recommend having a look at our job openings today – there are some really exciting roles available in our data science team.
Dan Nelson, Head of Data