AI for business, all you need to know Part 2
How do machine learning and automation relate to AI?
Machine learning is a major part of most of the AI used in business today. This practice relates to the concept of using large sets of data to identify patterns, connections and trends, without specifically programming any rules. The algorithms train themselves with the data and can even be rewarded or punished if outputs aren’t optimal. A typical example is visual recognition, where AI models are trained with thousands of pictures of similar items (such as coffee cups), so that completely new coffee cups that have not been tagged before can be recognised.
Automation, on the other hand, does not necessarily need to have any AI involved if the process to be automated is well defined. IT teams have worked over the last few decades to automate millions of tasks, such as moving data from one system to another, with traditional scripts that work brilliantly – until they fail. Intelligent automation is starting to appear in businesses with the advent of RPA (Robotic Process Automation) which does a great job of simplifying and accelerating automation that can’t be programmed easily, and many companies have still got opportunities for automating a massive range of activities even without AI.
Is AI threatening jobs? Why?
Currently, I don’t think so. Poor business models, bad customer service and wrong leadership decisions are threatening jobs far more than AI.
Of course, there are always going to be companies that will try to make use of technologies such as AI to reduce headcount, but organisations that take advantage of the possibilities to amplify human performance with AI are always going to be the long-term winners. Paul Daugherty, the main author of ‘Human + Machine: Reimagining Work in the Age of AI’ tells the story of an innovative company called Stitch Fix that recommends and delivers an outfit personally chosen for each customer every month. AI algorithms trawl through thousands of clothing items and automatically produce short lists based on each customers’ unique preferences, but human stylists make the final choice for the best fit. Stitch Fix currently employs 2800 stylists – a great example of jobs that wouldn’t have existed without AI to augment them.
What are the steps to take when implementing AI in the business?
This is relevant to all technology initiatives – first decide on the goals – what you want to achieve, then the roles – what are the groups of things that different people are going to need to do to achieve those goals, and then the souls – who internally or externally could fill those roles now. Goals, roles, then souls.
You will need executive sponsorship, someone championing the project from the top. Budget allocation and technology choices that align to your standard architecture and stack.
Then the meaty stuff – picking the top 5 impact versus feasibility questions that need to be solved. Over time, you will need a methodology that allows you to identify the next priorities, if you don’t already have this for standard projects.
The biggest problem that organisations seem to have when implementing AI is often the deployment of the answer into a production scenario. It’s one thing to find an insight that could change your business from a data science project. It’s something else to turn that insight into an AI-powered process and an operational KPI that supports it, and then get people to amplify their results by using it. So, you need a process to bring organisational change management into the mix (as ever). Of course, AI can help with change management – if you can digitally gather as much data as possible on people using the new AI solutions, you can use AI models to pinpoint exactly where protesters or champions are forming, and feed that back into the change loop.
What are the benefits of using an existing solution vs building from scratch?
This is a question that is relevant to pretty much all technology choices – you always have to choose between build, outsource or buy – whether it’s an operational system or a set of dashboards. The benefits and risks are all related to your inhouse capability and capacity with regard to development, and your profile in regard to timing and purchasing. The different issue with AI projects is that there is a massive shortage of skills, both at developer and leadership levels worldwide, because it has become effective (and popular) so very recently. This is an added dimension to making this choice with a problem that could be solved by buying or building a traditional software solution. However, of course, there are solutions coming out now that use AI to help with the development and deployment of AI solutions! From Altron Karabina’s point of view, we would recommend outsourcing or buying small solutions that have a quick ROI at this stage, to help evaluate where AI can help and understand what is possible before building an internal competency.
What practical advice would you give teams that are embarking on an AI journey?
I think the one word in the question that is most appropriate is ‘teams’. It is very difficult to imagine solutions that are going to be effectively implemented without team roles and coordination, as with any major technology initiative. Top executive sponsorship, as ever, is vital, and the expectation that there will be failures, as with anything agile, is required. The main suggestion would be to have a set of five ‘sharp questions’ that can be solved by AI, prioritised by impact and feasibility, and use them as pilot projects – such as ‘how can we reduce our customer churn from 10% a year to 6% by identifying early interventions’. Lessons that are learnt both developing and operationalising those projects will be vital once you start looking at building an AI platform and methodology going forward.
Contact us for more information on leveraging AI for your business.