Summary: Done properly, applied artificial intelligence (AI) can enhance the user experience across your product – providing value for your users and your organisation. However, you need to allow your users to make the final judgement, spend time training your model, build in feedback loops and be transparent about how you’re using AI.
How to Apply AI
There are lots of different conversations going at the moment about artificial intelligence. Some are more abstract than others. The sharp end of research will continue to be out of product managers’ reach for a while yet – we’ll leave that to DeepMind. However there are numerous tools at our disposal to put AI to work now – and a few watch-outs for solving customer problems with these capabilities.
Start With the Problem, not the Solution
As all good product managers know, we need to identify the issues our users face and then find ways in which we can add value by solving them. It’s tempting to be led by the technology not the problems we’re trying to solve when working with new tools like AI. So avoid this mindset and be aware that not every problem can be solved with AI.
Any AI use Cases Have to Contribute to Your Overall Product Vision
It’s worth setting up some areas where you think AI can help to improve your product in order to investigate them. These areas must support your overall product vision – whether that’s making your users’ lives easier, removing hurdles within your product’s workflow and so on
Typical uses of AI at Diogo’s company, international media group Schibsted, have been to:
- Auto-fill / suggest data for users
- Identify steps that a user can skip
- Help with non-intuitive tasks such as categorisation
- Recognise real-world objects to save data input
The User is the Teacher and has the Final say
In any use case where you automatically suggest data to a user – such as categorisation of an item – you must allow them to “correct it” if they want to. You then must send this suggestion back to the AI tool so that it can keep learning and improving. You must track the number of these changes that users make, so you can measure the improvement of your feature. As it gets better, the feature becomes more effective and adds more value to your users and product.
Compare Human and AI Accuracies to Generate Buy-in
When implementing a new AI-powered feature, you need to test and monitor its accuracy. The best way to do this is to compare it against humans doing the same task. You may be surprised by how low the human success rates are for some tasks such as categorisation – and hence how good (or bad) your AI tool must be in order to still be worth investing in.
Don’t Forget Your Training Data
In order for your AI-powered feature to work, you need to train the algorithm on which it is based. This means you need data for the problem you’re trying to solve, with the correct answers attached. The more of this you have, the more accurate your model will be and the more value it will add to your product.
Decide how to Deal With Inaccuracy
Your AI won’t get the answer right every time. Fortunately though, it will give you a confidence metric with every suggestion. You can use this to decide what you want to show when it goes below a certain threshold. This can be a message to the user, allowing a choice or removing the feature altogether.
Be Transparent With Your Users
If you are using AI to change or influence the user experience, you have to let them know about it. If you don’t, they will get confused and stop trusting your product – causing issues across the board. Use copy like “auto-suggested XYZ” rather than just putting the information on a form.