Recently, Napier and Fintrail teamed up to run a choose-your-own-adventure session, aimed at informing compliance professionals of the steps and pitfalls in AI implementation for AML.
Dr Janet Bastiman, our Chief Data Scientist, and James Nurse, Managing Director at Fintrail, co-hosted the session to guide participants through the implementation process and explaining why carefully considering the various steps – and their order – are crucial to nailing your approach.
Watch the replay below and read on for our summary of the event and some of the most important advice our experts gave.
For the purpose of the webinar, Janet and James set the stage with a use case scenario:
“You’re a FinCrime professional looking to enhance your company’s transaction monitoring capabilities using AI – but how do you get there?”
Just some of the challenges to financial crime compliance include:
- Increasing number and complexity of regulations
- Rising pressure on compliance teams to deliver
- Existing processes are insufficient for demand
- Organisations run the risk of hefty fines for noncompliance, so there is pressure from management
- Once AI has emerged as the ideal solution, an implementation process needs to be carefully selected and navigated
There are three steps to the AI implementation process:
- Identifying AI as the resolution to your current problems
- Implementing AI
- Go live!
Sounds easy right? But in reality, there are a few more steps in between to consider and it’s key that certain of these are completed in the right order. Luckily, Janet and James were on hand to help fill in the gaps in the AI implementation process.
For the choose-your-own adventure aspect of the event, attendees were given two choices at each stage of the implementation process. There are countless ways this AI journey could have gone, but below are the choices the attendees went for on the day (and what they didn’t choose at each stage):
- Risk assessment over market analysis
- Identify relevant data over operating model
- Data assurance over vendor selection
- Regulation assessment over maturity analysis
- Training the team over implementation
- Transitional planning and governance over implementation
- Model assurance over assessment of old system
At the end of the session, James and Janet unveiled what they would consider to be the ideal path to implementation of AI, which you can view below or as a pdf here.
Failing to prepare is preparing to fail
It’s important to allocate sufficient time and effort to the initial steps in the implementation journey, starting with the all-important maturity assessment. If your organisation and its data aren’t yet mature enough to implement AI, then this first step can help you make the call to bench the project until the time is right.
Assessing and understanding your company risk is another preparation step that’s not to be ignored. Before beginning your AI journey, it’s important to know the financial crime risks of your company, and what data and procedures you might need to manage or mitigate those risks.
Data hygiene is make-or-break
Especially if your business is moving to AI for the very first time, you may find that your data is disparate, unorganised and, in short, not ready. There may even be issues with accessing the information where there are legacy systems in place, so there’s a lot of investigation and digging to be done. The good news is that if you’ve done a risk assessment, you know what you’re looking for in your data and can start to consider how best to organise it.
Another key consideration is whether you have the right volume of data, as AI is powered by data. If your company is quite young and doesn’t have a lot of historical data, you may find you don’t yet have enough to power an AI solution.
Going live is not the end of your AI journey
While going live is technically the last step for AI implementation, it’s not the end of your AI journey as you’re responsible now for the system you’ve now set up. Ongoing quality assurance is absolutely necessary to continue to get the best out of your AI system and for keeping up with regulatory changes, as your system will need adjusting accordingly to continue to fulfil requirements.