Let Your Data Speak For Itself 

Robin-Grosset-Interview

Datatechvibe spoke to Robin Grosset, CTO at MindBridge about the data auditing, data wrangling challenges most enterprises face, the process behind selecting ML model and how important are APIs to ease ML and AI adoption in businesses.

“MindBridge’s philosophy is not to have a single model system. It means that our AI is built from multiple different machine learning, statistical and rules-based models. It’s called an ensemble AI,” explained Grosset.

Excerpts from the interview;

What are the key aspects one should acknowledge during data auditing?

Performing a data-driven audit is not necessarily the most conventional thing. You have to consider the audit in a slightly new way. It’s important to let the data speak for itself. For example, when you are thinking through the audit outcomes and your business expectations, having a data-driven audit using advanced technology like AI is what you need to let the data speak for itself.

This means that you need to leverage the data at an early stage in the audit activity, right from the expectation setting and the audit planning phase. This can often lead you to some of the more important areas in the audit.

A data-driven audit requires us to rethink this approach. It is important that your data speak for itself.

What are the common data wrangling challenges enterprises face, and how can teams approach it?

This is about setting expectations. There will be challenges regarding data wrangling and getting the data ready and correct.

Often, if you don’t leave it to the last minute, you will not be surprised, and everything goes much smoother. It’s important for firms to have a data strategy on how to connect with their clients. It sounds simple, but know what to ask for, and be precise. For example, at MindBridge AI, we guide our clients so their data ingestion process can be easier. We even go to the extent of helping our clients load and wrangle data because we recognise this is a challenge.

Have a data strategy and engage with your vendor to ensure you get the right information. Then, it will be easy sailing, and you won’t have people working long hours to try to figure out a data wrangling problem because it will all be managed. Data wrangling challenges are sometimes as much as a third of the energy people put into data-driven audits. It’s possible to be more efficient if you plan and have a strategy in place.

How does MindBridge select its ML model? What is the process behind it?

Our view is to create a robust AI system that can analyse financial transactions from different industries around the world. You need to think through the problem, and not rely on a single model. So MindBridge’s philosophy is not to have a single model system. This means that our AI is built from multiple different machine learning, statistical and rules-based models – an ensemble AI.

Explainable AI is a big part of our philosophy Our AI serves non-technical financial professionals. Our AI needs to explain what it found in plain language. We spend a lot of time in the model selection, thinking about how explainable it is. These models need to have an explanation which is going to make sense to people.

We also have our algorithms audited by a third-party expert in the AI system. Last year we worked with an organisation called University College London Consulting to audit our algorithms. It’s important for our clients to know that not only are we building insightful algorithms, but also that they work and they don’t need have to take our word for it. They can look at a review by a third-party professional to know that algorithms are well-designed and robust.

We combine different techniques for selecting the model. We have statistical, and rules-based and machine learning. We are always looking for novel approaches to a general ledger analysis, we are applying 32 different methods. We call them control points to pinpoint areas of interest in the financial data. Not a single model system but one that is constantly evolving. We have a lot of pride in the capability that we have created and it’s continuing to get better.

I don’t think that we can imagine every possible circumstance. If a fraudster knows how they’re going to be detected, they will avoid all of the rules. They will sidestep any rule they’re aware of that might get them detected.

You have to get ahead of the game and design things that will detect unusual scenarios you don’t anticipate. That’s where our favourite kind of machine learning comes in, we extensively use Unsupervised Machine Learning which can detect unusual and unanticipated scenarios. 

How important are APIs to ease ML and AI adoption in businesses?

Our product serves non-technical financial professionals who aren’t necessarily coders or statisticians. They want a product that doesn’t require you to write code, something that jumps out of the box and provides value and insights. The expectation at MindBridge is that you shouldn’t need to write a code or use APIs to receive value. There’s a flip side to this – APIs are amazingly valuable. Let’s say you are a firm and want to perform 500 or a 1,000 monthly audits. Data can then be processed as efficiently and quickly as possible so that your team is spending time on the right things.

With APIs you can get data from one side of a system to another quickly by leveraging  APIs to automate. So APIs can make it possible for organisations to push data into MindBridge, run their analysis, get their insights, and then also pull those insights out and have them go where they need to go. This works well for integrating with working-paper systems or to leverage report output from MindBridge that can be shared with their client.

Both are super important. I would say it’s at the point where you are industrialising, and going to full-scale adoption of something that becomes vital.

So, MindBridge has APIs for that scaling adoption purpose, which is a great perspective

How is data science improving financial transparency in banking and audit systems?

Data science can help us improve the transparency around information by helping a human auditor understand what’s happening in vast amounts of data. I was recently working with a client where we were looking at millions of transactions on a single screen. Most of these, 99.4 per cent of the cases, were things we don’t need to worry about it. It was just the 0.6 per cent that we should spend our time on. If you can build a technology that can look at all of the financial data, every single line of every single transaction, and then create these insights and summarise it in such a way that we can save people time – which is a huge value.

It’s not about having the machine replace the human in the loop. It’s about the machine taking on the arduous process of looking at everything so that the human counterpart does not have to. This builds better financial transparency because we know where the problems lie.

It’s making the lives of people who have to audit and review financial data much easier.

We are allowing them to get away from mundane, repetitive tasks, constantly reviewing things that are normal, to spend their time on the highest value areas — looking at the things that require this human level intellect to review and understand.

How can AI be leveraged to increase trust in public services?

The insights you find from financial data help organisations to make better management decisions. Organisations that leverage AI and analytics on their financial data are making better decisions. They are producing higher quality outcomes for everyone – not just the organisation itself but for the receivers of the services as well.

Analysing what’s going on in the data, leveraging that information to make better decisions, rather than  sampling randomly or just looking in the corner of the data.

If you randomly sample, you’re looking to verify the presence of controls. The control worked in this case; you sampled the data and verified that the control worked. But isn’t it better to look at all the data, and then be able to say where the control didn’t failed?

Don’t look for confirmation that it worked. Look for the evidence that it didn’t because that’s where the value lies, and where action is needed to improve it. I think AI can be leveraged to increase trust in public services.

To learn more about the speakers, register for Velocity https://ksa.velocityda.com/

If you liked reading this, you might like our other stories
The 9 Cs Of Cybersecurity For The Modern Business
Is Poor Data Quality Affecting Your Business?