Here's How To Test If Your AI Solution Will Be A Success

Anders Liu-Lindberg and Thomas Schultz

Part 7 in the “RPA and AI in Finance series, which examines the role that robotic process automation and artificial intelligence can play in finance operations.

Does this sound like you? You’re a tech geek and on top of all the latest developments in the market. When it comes to artificial intelligence (AI), you’re like a fish in the water and know exactly what solutions you can leverage to achieve significant benefits in your finance function. In brief. You ROCK AI! Any takers? Probably not. Most of us don’t have a good idea about how AI solutions could deliver benefits to finance. What’s the best way to figure it out? Test it! That’s exactly what we’ll teach you to do in this article about bringing AI into the business and the finance function. Here’s the third step in our five-step model for succeeding with AI.

Do a feasibility study on the back of a napkin

A terrific way to see if you can grasp the full extent of a problem is to try to solve it on the back of a napkin. If it can’t be described in simple terms with simple solutions, you likely don’t understand it well enough yet. This is also the challenge with AI.

Challenge: The AI part of AI projects is the “easy” part. That makes it very easy to make a small prototype that gives promising results. However, the road from a simple prototype on curated data to a real-life, operationally implemented solution is often long – and you can’t predict beforehand what you will be facing moving from prototype to implemented solution.

Fix: Do the “Thomas Schultz” 10-step AI-test on a napkin:

  • Step 1: Can you identify an “object of interest”? That’s the object you want to be able to predict something about: a credit card transaction, a customer, a bank account, a bank transfer, etc.
  • Step 2: Does an improved analysis or more correct handling of the object of interest have large, positive consequences business-wise? Or are a lot of resources used on manually handling the object of interest?
  • Step 3: Is it easy to define a meaningful and simple categorization of the object of interest that makes sense in business terms? Credit card transaction (fraud score), customer (high/low) long-term value, bank transfer (money laundering yes/no), etc.
  • Step 4: Pretend that you already (miraculously) have the AI solution in place and can predict exactly what you want it to do. How will you use the new knowledge? What will you do that is different from what you do today? (A clear answer is needed.)
  • Step 5: Training data for labeling: AI-models needs training. Do you have access to historic data about your object of interest and the needed classification? Do you have a record, e.g., of fraudulent credit card transactions, loss-generating customers, money-laundering cases? If not, can you generate some training data?
  • Step 6: Training data attributes: For your object of interest, do you have data on ALL the attributes that might explain why something happened historically? Do you know the right things about your customers to build a good AI model (based on knowledge about loss- and profit-generating customers)? If shoe size is the most important loss/profit predictor and you don’t have any data on shoe size, it’s game over.
  • Step 7: Did you do your business case on a napkin? Does it promise business success if the AI project is a success?
  • Step 8: Are you sure you need AI to solve the problem? Is it really that complex of a problem?
  • Step 9: Are you sure you can’t just download some piece of software, such as image recognition or speech-to-text translation, that will do the job? There are many problems that these applications have solved already (by AI). Just download and use!
  • Step 10: Are you sure you are legally allowed to do what you want to do in terms of GDPR, privacy, and ethics? Joining a multitude of data from many diverse sources is often a prerequisite for AI projects.

Does it feel more tangible now to evaluate whether AI will be good for your business and finance function? At least you now have a structure to follow that will organize all the thoughts you’re currently struggling with. As you can imagine from the sketchy examples we’ve already given, there’s a significant potential for the finance function in AI.

Are you ready to do some testing?

Yes, we still have more steps to go in our model, but now you should be ready to test if AI could be the answer to some of your business challenges. Just identify an object of interest and get started. It is not that hard.

If you have examples of how AI has helped your finance function solve business challenges, it would be great if you could share them – not just the final solution, but also how you went about building the business case, the software, and the organizational change efforts. More real use cases mean even more AI in finance.

Please email us: Anders Liu-Lindberg and Thomas Schultz.

This article originally appeared on LinkedIn and is republished by permission.

Anders Liu-Lindberg

About Anders Liu-Lindberg

Anders Liu-Lindberg is the head of the Global Finance Program Management Office at Maersk and has more than 10 years of experience working with finance at Maersk, both in Denmark and abroad. Anders is also the co-founder of the Business Partnering Institute and owner of the largest group dedicated to finance business partnering on LinkedIn, with close to 5,000 members. His main goal at Maersk is to create a world-class finance function not least when it comes to business partnering. He is the co-author of the book “Skab Værdi Som Finansiel Forretningspartner” and a long-time finance blogger with 20,000+ followers.

Thomas Schultz

About Thomas Schultz

Thomas Schultz is the CCO of Enversion, an AI-experts company based in Denmark. Thomas leads the efforts of Enversion to dramatically change the engine room of the finance function.