AI has been a buzzword for the last few years, but increasingly I meet people who don’t know what it can do for them, or how best to use it – yet they have been tasked with establishing AI to drive efficiency in their business. At this point there is a risk of falling for the AI platform marketing pitch which, if you don’t know where to start, invariably results in selecting a similar use case to the case study presented in the marketing collateral, as opposed to specifically establishing how AI could add value to the business.
To evaluate how AI can drive value within your organisation, you first have to take a step back and define a selection of problems or points of friction that cause delays, impact customer service, increase costs, or cause inefficiency. If you then identify the business value that derives from addressing these points, you can rank them and start to identify the AI tools that will enable you to create high value use cases for deployment. Any decent AI company should offer an indication of the effort required to deliver the AI solution for each identified use case. With the business value identified and the technical effort quantified, you will be in a position to design your AI roadmap and prioritise the most valuable opportunities.
The key to effectively delivering the first AI use case on your roadmap is to narrow the scope and set clear parameters for what success will look like. This foundational use case enables you to quickly demonstrate to the business that AI will drive value, which will help you to get buy-in from the senior team. It will also enable you to quickly scale your AI solution and expand the scope. Implementing AI in this manner also allows you to establish within your company that you’ve understood how best to use AI and show that you are planning ahead for other opportunities that may arise.
Delivering AI solutions follows the standard delivery framework of Define, Design, Develop and Deliver, except for an additional and crucial step that must be taken – Determine.
AI technologies require training, validation and stabilisation prior to being released into production.
For example, if you create an AI solution to identify fraud in customer transactions, you need to supervise it initially to ensure that it correctly identifies fraud and for the right reasons. One way of doing this is to set a confidence threshold, for example if the system is less than 70 per cent certain that the transaction is fraudulent then it hands it over to an agent to review and validate. The agent can then feedback the correct result to the system so that it will know how to assess that information next time and increase confidence above the set threshold.
Once moved to production, the management of AI depends on the tools being used, but AI is no longer the black box it was once perceived to be. My recommendation would be to select an AI system that provides an audit trail so that you can review reports and adjust the system as needed in order to provide ongoing training as and when outliers occur. This will ensure that the confidence threshold is met and assist general trouble shooting.
The prospect of evaluating, delivering and maintaining an AI system may seem overwhelming when you’re being directed to implement AI in your business, but if you identify the pain points that AI could address, then you can quickly assess how best to drive value and demonstrate to your business the deliverable capabilities of Artificial Intelligence.