James Maxfield, Chief Product Officer at Duco.
It’s an exciting time for technology. There are some seriously powerful AI tools on the market, solving real issues that capital markets firms have been battling with for years, if not decades. Undoubtedly, this is just the start.
Yet at the same time there’s a noticeable trend to try and push AI as the solution to every problem. This is clearly too much to expect, but the allure of having the latest tools to put to work against your biggest pain points is a strong one.
This is particularly an issue when the solution you’re being sold, no matter how cutting-edge it may be, will only treat the symptoms of a problem, instead of addressing the root cause.
In the realm of data, this means you end up with a ‘solution’ that cleans up the exceptions caused by bad data, but does nothing to prevent them coming back. And given that bad data is a huge source of operational inefficiency, cost, risk and opacity, these kinds of AI solutions only serve to bury the problem further.
So here’s a public service announcement: sometimes you actually don’t need AI.
A.I. why?
We had this experience with a new client. When they were looking for a solution they came to us with a list of requirements for what they wanted AI to be able to do. Two of their top requirements were using AI to quickly build rules and improve match rates.
Our response? We can deliver those outcomes, but you don’t need AI.
Let’s take a look at both in more detail.
Quickly building rules
This is a classic problem for firms running hard-coded legacy technology. The systems are so technical that only developers in the IT team have the skills and knowledge to build processes or even make small changes to existing ones.
So our client’s request was essentially can we replace developers with AI?
The thing is, Duco clients have been building processes quickly and without coding for years using our proprietary Natural Rules Language (NRL). This no-code capability means that business users can select from plain English commands in order to build data transformation and validation rules.
Is it quick? Clients running best practice operating models are able to make changes in production in under 30 minutes and build new reconciliations in under 5 days. One client built 700 reconciliations in one year.
And, on top of this, because these rules are built by the business user and not AI, they’re easily explainable and their plain English nature means they’re understandable by anyone; other team members, internal auditors or regulators.
So as well as the speed the client was looking for, they got total transparency too.
Improve match rates
The five core data challenges of variety, change, scale, lifecycle and control stymie traditional reconciliation solutions.
The rigid way in which they work means that data often goes unmatched because of minor discrepancies. It doesn’t matter if a human can see at a glance that it’s the same data – the system can’t.
AI is able to quickly review these exceptions and determine the root causes of exceptions, making them quicker to address.
But…what’s really the problem here? That exception management is tedious and repetitive? Or that the reconciliation system upstream isn’t doing its job properly?
Think about it like plumbing: if you have a leaky ceiling, do you buy an expensive robot bucket that can whizz around catching all the drops, or do you get someone in to install better pipes?
We were able to show the client that our proprietary matching engine produces much higher match rates. It does this because it compares data differently to other solutions and therefore spots connections other systems can’t.
Combined with the no-code ability we talked about above, it’s possible to rapidly iterate on your processes to add new rules and transformations. This will push match rates even higher.
As a result, compared to other matching engines, Duco clients are enjoying benefits such as 40% fewer cash reconciliation breaks, or match rates between 90 and 99% versus 60-70% on an older, on-premise solution.
The issues with false exceptions we’ve outlined above are avoidable. There is a role for AI in exception management, but if you’re using it to improve match rates then it’s a very expensive band-aid for an avoidable problem.
The solution to this has to be applied to the problem (the low match rates) and not to the symptoms (the false exceptions).
Thinking about value
I went to a conference hosted by Google recently and the conversations around AI were very interesting. While there’s a lot of demand for AI from the market, in fact most firms there shared that they or their clients weren’t actually buying it. Those that did were left a bit disappointed by the results.
Google themselves shared some key insights as to why this is – most of the time they’re asked by their clients to build an AI solution for a problem where AI just isn’t necessary.
The real takeaway from the event was that there’s a lack of value here. People are building features rather than solutions. This is a classic technology problem in the industry – we saw this with the dot.com bubble, we saw this with various financial products. It’s all very well saying ‘our technology can do X, Y and Z’, but what does that actually mean for a client?
Take document processing as an example of where AI clearly adds value. These types of data are unstructured – the data could be anywhere and doesn’t have to follow conventions. Automating document processing requires technology that can understand what it is looking at and what to do with that information. Clearly something only AI can do.
And the value for firms is just as clear: they currently have teams buried under document processing work. AI helps clear backlogs and improve straight-through-processing (STP) rates. This results in quicker responses to clients and faster processing of things like validation of OTC confirmations. It makes businesses more responsive and efficient and frees up employees to spend more time on value-adding activities, not admin.
The results are clearly measurable in terms of how many hours are saved each month, how much higher the STP rate for documents is and the time spent manually intervening where the AI isn’t confident in its prediction.
Compare that to the exceptions management example. Sure, you may save time and money by needing fewer Operations or Finance team members to manually sift through exceptions. But if those exceptions are caused by an inefficient solution upstream, that saving is on a cost you shouldn’t have been paying in the first place!
Using the right tool for the job
A fundamental rule of any technology is: use it for what it’s built for. Otherwise you’ll just be disappointed. AI exception management can be a useful tool in your data management arsenal. But if it’s there to compensate for weakness in other platforms then you’ve paid to solve the problem that you’re already paying to create.
AI absolutely has a place in data automation. There are things it can automate that other technology can’t – such as adapting to the many different varieties and requirements of documents that need processing. But buying AI for AI’s sake doesn’t move you closer to the agile, efficient and transparent operations that you need to remain competitive and future proof.
For that, it’s always better to focus on solving problems rather than treating symptoms.
And if AI tools fall into the former category, then by all means embrace them and enjoy the genuine benefits they can bring.
Subscribe to our newsletter