Introducing AI: Roll it out right

You need to build in quality controls as part of your AI deployment. That doesn’t mean redoing everything that your AI does. But it might mean sample checking.  

It might mean starting small, in low risk situations, to build up the confidence level before you use it more widely. 

In the case of documents, for example, you might use it on a small number, where you can check all or most of the results – before you use it on a million, which you couldn’t realistically check. 

If you then started to use it on a new type of document, or you asked it to do something quite different, you might repeat that process, just to make sure things are still OK. 

And with some types of AI, you’ll want an ongoing system of monitoring – perhaps with spot checks or periodic tests – to make sure their continuing development isn’t introducing distortions into the output.  

Find out more in our latest Bandwidth AI series. 

Different international businesses work in different ways. But many, as far as possible, like global solutions to their problems. 

They might be happy to do something on a regional basis – so maybe do something in the Americas that they’re not doing in EMEA. But many won’t want to bring those differences down to individual country level. 

They don’t want to have lots of different rules in different territories. So they’ll often go for the lowest common dominator, and find a solution that works in multiple jurisdictions – even if better solutions might be available for individual jurisdictions. 

And for AI, at the moment, that lowest common denominator might be fairly unambitious. Having said that, the nature of AI is that it creates opportunities. 

You could switch elements of it on or off in certain places, or have it within some applications and not in others. But even then, in the real world, compliance teams may well have concerns about rolling out certain functionality in some territories and not others. 

A lot of people have been interacting with AI for a long time, perhaps without realising it. A typical chatbot, for instance, is just a basic form of AI. 

We’re are now at the stage where some people readily accept advice on investments and other decisions from AI. For others there’s definitely a trust issue. 

But over time, people will become more accustomed to the idea and more comfortable with it – just as they’ve got used to other technological change like mobile phones, online banking and e-commerce. 

That won’t necessarily be a linear, uniformly smooth progression. It could be disrupted by high-profile mis-selling, personal data breaches or many other possible events. 

Public sentiment can shift, as we’ve seen in the long-running story of where banks locate their call centres and how they staff them. But overall, it’s clearly the direction of travel.