Johannesburg, 25 Apr 2018
2018 is becoming the year where everyone is putting artificial intelligence on their key strategy list. Because AI is a nebulous term, for many, having an "AI strategy" is demonstrated by a plan to roll-out a chatbot to their customers. Don't get me wrong, chatbot technology is maturing fast and can bring great value when applied in the correct way, but many organisations are missing something big here.
Data science and machine learning, the tools and technology often associated with AI, are more mature in many ways and can transform a business at a more fundamental level. There is an underutilised goldmine of data in every business that may be hiding the answers to some important questions. I'm talking about the unstructured and logging stores generated by the sales process, your mobile app, Web site and call centre, every minute of every day. At best, this data ends up in a graph on a specific dashboard, but most often it's simply stored and archived.
Organisations have started using machine learning as part of a data science programme to put this data to work.
Some proven ways of using the predictive power of historic data include:
* Predicting when systems and parts will fail, enabling proactive maintenance;
* Identifying fraudulent transactions and clients;
* Predicting the success of a product or item and highlighting which factors make particular products popular or unpopular;
* Predicting the likelihood that a discount will help close a deal or which customers are thinking about abandoning your brand; and
* AI-driven logistics scheduling.
So, how does putting your data to work become ingrained in your strategy? It's about putting data at the heart of every decision. Management's role needs to change from answering questions, to crafting the right questions. Two organisations that have undergone this change are Google and Uber, where pricing, product details, logistics and marketing decisions are all driven by data and almost entirely automated using that data. Here, management's role is to ask the right questions of the data.
In 2006, Google's then-CEO Eric Schmidt said: "We run the company by questions, not by answers... and that stimulates conversation. Out of the conversation comes innovation. Innovation is not something that I just wake up one day and say 'I want to innovate'. I think you get a better innovative culture if you ask it as a question."
This is very different to approaches of the past, where management's over-reliance on experience resulted in a lack of innovation and group think. Yes, data was used, but it was often cherry-picked and assembled to support an existing decision leading to confirmation bias rather than real findings.
A key component of the new data-driven business is the data scientist. Guided and focused by the big questions defined by management, the data scientist's objective is to navigate the many pitfalls associated with extracting truth from raw, unprocessed storage. Statistics and machine learning are full of pitfalls.
As data is the fuel that powers a machine learning strategy, the machine's performance will depend on the quality and quantity of the fuel. Taking a data-first approach to decision making therefore has a huge risk.
Without a critical analysis of the data, it will tell you untruths leading to rickety, ineffective strategies. Bias is data's biggest affliction. Perhaps your sales team only captures leads from specific market segments; this will result in a bias which could lead to flawed conclusions if this data isn't normalised.
Another common pitfall is confusing correlation with causation. A correlation needs to be investigated, often in the real world, to determine what the causation factor is. An over-simplified example would be to assume that fire trucks are causing fires as they always seem to be where the fires are: the mistake here is obvious, yet businesses will often make similarly bad choices based on not clearly understanding the correlation/causation pitfall. Ultimately, the data scientist needs to prove the predictability of the data and test the accuracy of predictions made by the trained AI models. This approach of testing evidence and proving the reliability of outcomes and business impact is what puts the "science" in data science.
These new data-driven organisations are leading the way in innovation. Becoming a business that is able to predict outcomes based on past information is an iterative process involving the following steps:
1. Refine the questions: What are the big questions that drive our business? Even the ones we think we know. What are the important questions that we aren't asking?
2. Evaluate your raw data: What data do we have that could shed light on the questions? What additional data can we obtain? What actions need to be taken to ensure higher data quality in future?
3. Process the data and refine the AI models: Process the data to understand the correlations and unravel the hidden features. Build machine learning models to make predictions and evaluate the historic predictability of the models.
4. Run real-world experiments: To untangle correlation/causation confusion, predictions and correlations need to be tested against the real world in controlled ways that don't put the business at any serious risk. Real-world experimentation, often in the form of A-B tests, are a key ingredient to becoming a data-driven organisation and the craft of data science.
5. Take the findings and repeat...
Machine learning has the ability to bring predictability and optimisation to organisations, which was simply not possible in the past. It enables a business to have predictive maintenance schedules, minimising logistics and minimising failures, with less staff. Customers can have an entirely personalised experience that will feel more human and less like advertising. As with all revolutionary technology, it comes with huge organisational risk if rolled out inappropriately. But, ultimately, businesses not utilising their data to drive their strategy simply won't be able to compete.
About the author
Craig Heckrath is an enterprise architect comprising a software development history of more than 20 years.
Currently, Heckrath is the delivery lead for Mint Intelligent Insights, specialising in artificial intelligence (machine learning, NLP, chatbots, DNN, AI platforms). He has architected high-profile projects across sectors including: financial, healthcare, public, media, telecommunications and transport. His guiding drive is to keep technology simple, usable and meaningful.