Join the Community

22,077
Expert opinions
44,027
Total members
419
New members (last 30 days)
204
New opinions (last 30 days)
28,695
Total comments

Get going on Machine Learning: from excitement to readiness and justification

The Internet is bursting with opinions on artificial intelligence (AI) and machine learning (ML). An intellectual war is raging amongst academicians, ML evangelists, data scientists, and big data gurus. At the same time, much of the business community is still working out exactly what these emerging technologies are and how to exploit them for competitive advantage.  

What are AI and ML?

Simply put, artificial intelligence (AI) enables smart systems to act like humans – in other words, with intelligent behaviour. In contrast, machine learning (ML) is a subset of AI that enables machines to act without being explicitly programmed; ML is the underlying capability that supports AI-enabled smart machines.

We are all already interacting with – and benefitting from – AI and ML. Crowdsourced traffic app, Waze, Uber’s ride optimisation functionality, and the facial recognition and social networking functionality of Facebook has AI/ML at it's core. Amazon uses neural networks (AI) to generate product recommendations, and the core technology powering Alexa is AI.

Even the banking industry, progressively is using AI and ML to transform service experiences and deliver greater value to customers. HSBC using AI-enabled fraud detection has been a much talked about story among many others.

Innovators now understand that the time to learn about and contemplate AI and ML is over. The pressure is on to drive real business innovation using AI and ML. But this will require shifting from a “can’t do, won’t work for us” mentality to a “let’s try and make it happen” way of thinking. At the same time, business leaders need to take time to learn about these technologies before they sign off on AI and ML initiatives.

So how can we get started with ML?

To jump-start AI and ML initiatives for your organisation, answering the following three questions will help:

Question 1: How do I ensure strategic business relevance?

ML initiatives need to have strategic business relevance. Ask yourself if your ML initiative meets the following three criteria:

  • It has a clear purpose.  Specify how algorithmic knowledge from AI and ML will drive your business strategy or accomplish specific business objectives. If the ML promise is to “know things before they happen,” then all ML applications must be aligned to future strategic drivers such as customer growth and risk management. Companies need to focus on use cases and specific business problems that will transform business drivers and board-level level management information.
  • It should be scalable.  Single-use case applications can be limited in scope and business relevance. To unleash the power of data and ML, plan ahead for scalability, as ideally, ML solutions will be relevant for multiple use cases across the business.
  • The initiative will deliver tangible and significant benefits. With effective implementation and upfront planning, ML can impact multiple business drivers such as risk and compliance, revenue, and cost control – even when the implementation starts as a simple use case. (For example, the benefits of a complaints management ML initiative may initially focus on risk and cost benefits. But over time, ML may expand the value proposition by identifying new customer clusters and customer engagement segments to drive revenue.)

Question 2: How do I prepare for an ML project (inception to technology implementation)?

Next, ensure organisational readiness by looking at it from multiple perspectives:

  • Pick the right project lead:  Selecting a lead for an ML programme requires careful resourcing and due diligence so you find someone with delivery capabilities, an innovation and risk-taking mindset, and strong senior stakeholder engagement skills. Other key skills needed include technical awareness (such as various maths, coding languages, technologies) and soft skills such as problem solving, communications, and storytelling. 
  • Ensure data readiness: Data preparation will be key to a successful project. For banks, data comes in many formats – from paper and media files (audio, video) to real-time natural language (voice, accents, tonality) and unstructured data – and the quality often may be questionable. In addition, with new GDPR requirements, ML programs will require careful planning and preparation from a data perspective. As a result, your enterprise and data architects may need to upskill and have a comprehensive enterprise-wide data perspective.
  • Determine algorithmic appropriateness: Figure out the best algorithms to apply to the vast amounts of data, in order to understand and predict behaviours of customers, employees, and more. This will require thinking through:
    • Your objectives: Are you predicting something based on a defined data set with a clear purpose such as fraud detection, customer segmentation, and next-best product? Or are you exploring the data to gain insights that could bring something new to the business (for example, through the use of transactional clustering to understand the behaviours of similar customers).
    • The business application: Is the ML output going to drive real-time decisions in the front line? Or will it feed an operational process (such as interpreting incoming help desk calls and accurately routing them to the right person for problem resolution within a single call)?
    • The best technique: Both supervised (classification and regression) and unsupervised (clustering) objectives work with different underlying algorithmic models, from discriminant analysis to decision trees and neural networks. Seek advice and expertise to determine which one works best for your ML initiative.
  • Assess infrastructural readiness: Your technical architecture may need to evolve into a scalable data platform that supports structured and unstructured data and simplifies integration with multiple applications, so people can securely access right insights at the right time. This architecture will need to be supported by the necessary networks and computing power along with robust test/dev and production capabilities in a cloud infrastructure.  
  • Ensure appropriate process selection: In most cases, processes that are large and complex are the most appropriate candidates for ML initiatives. To find the best candidates, review your end-to-end processes for complexity and value to the business, as this will help ensure a quick return on investment.

Question 3: How do I structure a business case for ML?

To prioritise your ML initiative within your company’s change portfolio, ML teams need to articulate measurable benefits. Executives and board members want to know how ML will deliver new revenue, higher NPS scores, greater employee satisfaction, and more. If you can demonstrate an opportunity for clear savings (for example, by reducing regulatory fines or fraud costs), it will strengthen your argument.

As you structure your business case, consider the following components:

  • The business design effort – to redesign the operating model (people, processes, technology) and determine how ML will change the way the business operates
  • Infrastructural costs – to ensure readiness for machine learning tools (applications to data platforms to hosting capabilities)
  • Data preparation requirements – to ensure data quality via ingestion, preparation, and normalisation (particularly when data is unstructured, multi format, or big data), as well as to plan ahead for the cost of labour-intensive tasks such as sorting and preparing physical artefacts (such as paper forms).
  • Solution design – from determining the best algorithm to solve the business problem to understanding how the solution will develop over time
  • Build effort – to develop the data model and go through the trial and error and training and re-training process needed to arrive at an acceptable level of accuracy
  • Change effort – focusing on training and communication to embed new ways of working across HR and for employees interacting with HR across the business
  • Ongoing maintenance costs – to ensure the machine (i.e., the algorithm and technique) performance, usability, and fitness for purpose. Over time, the ML system may need re-training (for example, if you add new data types) and updates to meet business requirements and take advantage of new technology innovations.

Together, these components will drive the ROI debate and project sign off.

Organisations exploring ML will no doubt want to embrace the new possibilities it affords. As you get started, focus on two priorities first: 1) identifying potential business problems where machine learning becomes a powerful enabler and 2) developing a proof of concept that delivers clear business outcomes.

AI and ML will change the way we live and go about our business. And the first ones to embrace and exploit its potential will be the winners. Clearly the ML winds are gaining momentum. How long can you afford to wait?

----------------

To learn more, join us for our webinar with Finextra on accelerating machine learning scheduled for September 12 @3pm GMT. Register here

 

External

This content is provided by an external author without editing by Finextra. It expresses the views and opinions of the author.

Join the Community

22,077
Expert opinions
44,027
Total members
419
New members (last 30 days)
204
New opinions (last 30 days)
28,695
Total comments

Now Hiring