Never before have AI developments captivated the mainstream media and public’s attention in this way, with 2022 and 2023 likely to be remembered as the years that such tech became conventional. And for a number of years, Recordsure has used the most important technologies that underpin the most well-known products like ChatGPT to provide a valuable, secure utility to financial service firms.
To learn more about how Recordsure’s artificially intelligent speech analytics tool ConversationReviewAI can help process risks in customer conversations safely and securely, watch the latest in our five-part mini-series. Together, Recordsure’s Senior Product Manager, Olivia Fahy and Chief Product and Commercial Officer, Garry Evans look ‘under the hood’ and explain how when AI is harnessed in the right way it can provide revolutionary solutions for financial service firms.
Part 3: How speech analytics can identify themes and process risks in customer conversations
Olivia Fahy
ConversationReviewAI uses sophisticated speech analytics to identify themes and process risks in conversations. Our Classify model is unique among speech analytics vendors, as the only product that visually organises conversations thematically using AI trained on use case specific conversations by our in-house subject matter experts.
Essentially, ConversationReviewAI helps find the important parts of conversations that align to your firm’s review checklist. Then your reviewers can listen to the parts of calls that are relevant, or easily jump back to the parts of calls they want to relisten to and find the key sections of focus.
Our smart segmentation will signpost you to those sections, - and that provides a 50% time saving per review. Reviewers can complete their review in half the time and do double the amount of reviews just by using our Classify module.
The second element of the product is Assure. Assure uses AI to identify whether processes were followed, and scores calls based on their level of process adherence, highlighting those that need human review. This part of the product gives oversight and a score for every single case that has been ingested into the system whilst providing MI that can feed into central risk reporting. And it’s through this that the need for random sampling is eliminated because review teams can be directed straight to high risk calls. By using Assure, firms can see an increase in their effectiveness from around 5% (if only 5% of the cases you’re checking are high-risk) to more like 95% - because the majority of time is then spent checking the cases that really matter.
Garry Evans
And as you said, it's always the question of whether an organisation has used speech analytics before. So, how is ConversationReviewAI different to other speech analytics software?
It's a very good question and one we get asked regularly. Other speech analytics tools on the market will generally rely on keyword or phrase spotting. This means that they find five false positives for every true positive, so for everything that you're looking for, it’ll find five things that you aren't. So, in the context of QA, that can create more work rather than less. For instance, systems that rely on keyword and phrase spotting would flag lots of these cases you need to look out for - but five out of six of them are absolutely fine. In essence it’s creating more work for your QA teams rather than making them more efficient.
Unfortunately, a lot of people that try using speech analytics tools for QA processes have experiences like that - and then don't believe that speech analytics is a useful tool that can enhance QA. But that isn't true. You just need to be using the right speech analytics tools. Keyword spotting can be really useful for in call type analytics, which are normally the built into systems. So, in that scenario, the lack of sophistication means that the accuracy is at less than 20% because of those high numbers of false positives.
But ConversationReviewAI does things differently by building neural networks to create our AI models. Without wanting to get super technical, our machine learning models are used to make decisions like the human brain. They're trained on thousands of hours of high-quality data from financial services conversations. We have an in-house AI training team with linguistics analysts and speech recognition scientists, and we teach the models the context around the language that is used, rather than just telling them to search for the word, or similar words to, 'mortgage' for instance.
That means we can target the conversation’s context of the language. When compared to that simple keyword matching that's used across the AI and speech analytics industry, it means that actually we can apply review frameworks more accurately and more consistently than humans every time. And as a result, our accuracy is around the 96% mark, so much, much higher than the 20% accuracy that's the general standard on the market.
I guess that's what makes it fit for purpose. If you need extra resource to do it, firms won’t to do it. But actually, if you make it as accurate as that then you’ll have a reduction in the amount of resource plus greater oversight of your portfolio. And I can see how that works really well for call centre conversations. But what about ones that aren't typically recorded like face-to-face or video calls? How can you use this system to evidence those conversations in the realms of the Consumer Duty?
As if by magic we actually have an app for that! In the context of the Consumer Duty, if conversations are happening face to face - financial advice type conversations – that aren’t being recorded, then it's something that does really need to start happening. Not just as evidence of the consumer outcome element of the conversation, but also because it's supportive for the advisor.
Should a complaint be made, or even for training and development purposes, it's really useful to have a record of those conversations. Our app called Capture has patented technology within it, which lets users securely record face to face and video meetings. It then creates a high quality, authoritative and reliable record of those interactions that you wouldn't otherwise have oversight of.
Compliance teams that we work with have described it as revolutionary because it allows them to ‘be in the room’ with the advisor every time. If you're currently spending time sending supervisors out to observe conversations that are happening once a quarter or whatever the cadence is, it saves money because you don't have to be sending people all over the country to observe the conversations.
But it also means you can get broader oversight because you can record every single conversation - then you've got an authoritative record should it ever be questioned.
Check back here for part four of the series, ‘ Futureproofing for Consumer Duty: Transform your Quality Assurance to innovate beyond random sampling’. Plus, catch up on part one and part two here.
