✨ New ✨ The Digital Services Act: A fireside chat covering all angles Watch it here → ×

What Accuracy Should I Expect From Artificial Intelligence Moderation?


    The future is AI-powered!

    That is also true in the field of content moderation. For years we have been waiting for Machine Learning to advance enough for us to use it to manage the ever-increasing volumes of user-generated content in need of moderation. Now we have reached the point in the evolution of AI, where it can take very accurate moderation decisions on most content. At least if it has been trained well on quality data.

    The magic of AI moderation

    At times, however, the fact that AI is so capable leads to unreasonable expectations.

    We often encounter site owners eager to aim at the magical 100% accuracy number and while we always strive to meet and exceed client expectations, this is a place where we have to advice caution and manage expectations to get the best possible results.

    Because AI moderation can be magical when applied right!

    Is 100% accuracy on AI-moderated content impossible?

    No, of course not. If you narrow the scope enough in automation levels anything is possible. But narrowing the scope means automating less content. And if you narrow it enough, the whole benefit of automation falls away.  A 100% accuracy rate won’t help much if you can only automate a fraction of your content moderation.

    Wait! Does that mean I should accept that mistakes are made, and unwanted content is allowed to slip through and go live on my site?

    Choose your current accuracy as a starting point. Then improve!

    Well, even if you are managing your content manually today, you most likely already do. A 100% accuracy level even for manually moderated content is very rare and would require such in-depth analysis of each content piece that it is neither scalable, nor economically sound unless you are dealing with incredibly high cost items.

    That is why our general advice is to set an accuracy start-goal for AI moderation, that is in line with what you are currently achieving through manual processes.

    The good news is though, that when you work with tailored machine learning models and use it in tandem with manual moderation, like we recommend in most setups, your accuracy levels will improve over time. What we often see is that with the models we build and the tweaks we perform and as the AI learns from manually taken decisions, the accuracy level quickly rises as high as 99%, even when automating up to 80% of all content.

    Pretty impressive numbers, wouldn’t you say? And they prove that tailored, AI moderation will really help you up user experience (through faster time to site) all the while increasing the cost-effectiveness of your operations.

    Be realistic and keep the end goal in mind. Accuracy is just a means to an end.

    If plotting both accuracy and automation on a graph, you’d hit a sweet spot where the two align. High accuracy is effectively the more important result, but in an online environment where customers and users expect to see instant results, being able to moderate content quickly is key. This is why balance is so important. And so is being realistic about the starting point! If you are not, the discussion will quickly get stuck on accuracy numbers rather than what will help your business and your customers the most.

    Curious about how automated and manual moderation can work for your business? Drop us a note and we will be back in touch shortly.

    This is Besedo

    Global, full-service leader in content moderation

    We provide automated and manual moderation for online marketplaces, online dating, sharing economy, gaming, communities and social media.

    Form background