Your AI is only as good as its data set
Anders Liu-Lindberg
Leading advisor to senior Finance and FP&A leaders on creating impact through business partnering | Interim | VP Finance | Business Finance
Teaser: The Artificial Intelligence you’re using to support FP&A isn’t clever, it’s just good at extracting things from data. How can you manage gaps in data and labeling to improve performance?
This is the newsletter "Here's the Future of FP&A and Business Partnering" which has 71,000+ subscribers. You can subscribe as well to receive new articles straight in your inbox every Monday and the occasional Saturday.
The latest episode of The Softer Side of Finance Podcast is live.?Catch it here!
Catch the latest episode of the?#FinanceMaster?Podcast here.
Listen to The Softer Side of Finance today at 15:30 CEST.?Sign up here.
Listen to the?#FinanceMaster?Show on Friday at 13:00 CEST.?Sign up here.
Brought to you always by?Business Partnering Institute.
For all the fuss over ChatGPT and the power of machine learning, it’s important to remember that AI isn’t actually clever. It’s just good at extracting things from data. This means that your AI is only as good as its data set. Managing its performance is a question of managing its data and its biases to improve the quality of the insights it can deliver.
Tales of poor data sets and opaque decision-making abound. One AI was found to consistently distinguish between pictures of a husky dog and a wolf. It was interrogated to see exactly what new insight was revealed. It turned out that the machine had learned to classify pictures with snow on as a husky, without as a wolf. It was very good at identifying snow.
Biases and gaps that exist in data sets can skew decision-making. Data that goes back into history will throw up attitudes and behaviors that may no longer be acceptable. Data that has not been disaggregated is a classic problem when sex and gender come into play. In 2019, the Apple Card ran into problems with its credit decision-making, offering lower lending limits to women, and there are a plethora of examples from medicines to road safety where decisions can disadvantage part of the population.
The importance of labeling data
Poorly labeled data can significantly impact the performance of AI systems, as the accuracy of machine learning models heavily depends on the quality of the data used to train them. When data is labeled incorrectly, the model will learn from incorrect information, resulting in inaccurate predictions or decisions. This can lead to serious consequences in various fields, such as healthcare, finance, and security.
For example, suppose an AI model is trained to detect cancer cells in medical images, but the data used to train the model is poorly labeled, with incorrect annotations or mislabeled images. In that case, the model might identify healthy tissue as cancerous or miss actual cancerous cells, leading to a misdiagnosis or incorrect treatment recommendation.
Similarly, if an AI system is used in a fraud detection system, incorrectly labeled data could lead to false positives or false negatives. For example, if a machine learning model is trained on a dataset that includes transactions labeled as fraudulent when they are not, the system will incorrectly flag legitimate transactions as fraudulent, leading to customer frustration and distrust.
Therefore, it is essential to ensure that data is correctly labeled before using it to train AI models. The process of labeling data should involve multiple individuals with domain expertise, and a system of checks and balances to ensure accuracy. Additionally, regular monitoring of the AI system's performance can identify any issues with the data, enabling timely correction and optimization.
Improving your AI’s performance
If the data set you’re using is too small, too aggregated, or poorly labeled, then your results will be guesswork - or worse. If the decision-making is opaque and the computer says no, can you be sure why?
At the AI team member’s performance review, the focus must be on closing gaps in data and making sure that decisions are being made based on accurately labeled information. The only way that AI can deliver quality strategic decision-making support is to be fed on the best raw materials you can provide.
Do you have any stories of good and bad data? Let us know in the comments.
This was the second article in our latest series about the use of artificial intelligence in FP&A. Subscribe to catch the upcoming articles and read the previous ones below.
Keep reading to catch our full previous series "It's time to fix management reporting" below.
Don't forget to check out our series "Building the FP&A team of the future". You can read all the articles below.
领英推荐
You can read a lot more articles about FP&A, Business Partnering, and Finance Transformation below. It all start's with “Introducing The Finance Transformation Nine Box”?where you set the ambition for your transformation. You should join the?Finance Business Partner Forum,?which is part of the Business Partnering Institute's online community.
Great planners always plan with contingencies?(the latest article in my series about the ten commandments of planning and forecasting)
Can't help falling in love FP&A?(the last article in my series about charting the course for a successful FP&A career)
Can we trust the machine for financial planning and analysis??(the last article in my series "Planning (as we know it) is dead")
The secret sauce of FP&A transformation?(the last article in the series "FP&A Transformation Talks")
How Finance should use its seat at the table?(the last article in the series "The Unfair Advantage of Finance")
The ultimate guide to decision-making for finance professionals?(the last article in a series about the decision-making process and how Finance should impact it)
The Mindset Change Checklist For Finance Professionals?(the last article in a series about the mindset change that finance and accounting professionals should make to become business partners)
It's Time To Decide If You Want To Be A Business Partner?(the last article in a series about the personality traits of business partners)
All Successful Business Partners Are "Leaders"?(the last article in the series about our new capability model)
Should We Keep Talking About Business Partnering??(part of a 17-article series where we deep-dive into the WHY, WHAT, and HOW of business partnering by putting it on a formula)
Everyone Can Adopt A Business Partnering Mindset?(part of a six-article series about FP&A Business Partnering)
From Business Partner To Working Within The Business?(part of an article series where I interview finance professionals about their careers in FP&A and Business Partnering)
Is Your Product Optimized For Value Creation??(part of a toolbox series where we look at what tools FP&A professionals should leverage to drive value creation)
How Business Partners Turn Analysis To Insight?(part of a case study series where I interview business partners about how they drive value creation using real cases)
Anders Liu-Lindberg?is the co-founder and a partner at?Business Partnering Institute?and the owner of the largest?group dedicated to Finance Business Partnering?on LinkedIn with more than 11,000 members. I have ten years of experience as a business partner at the global transport and logistics company?Maersk. I am the co-author of the book “Create Value as a Finance Business Partner” and a?long-time Finance Blogger?on LinkedIn with 185,000+ followers and 245,000+ subscribers to my blog. I am also an advisory board member at?Born Capital?where I help identify and grow the next big thing in?#CFOTech.?Finally, I'm a member of the board of directors at?PACE - Profitability Analytics Center of Excellence?where I support the development of new analytics frameworks that can improve profitability in companies around the world.
| CFO | Finance Business Partner | Planeación Estratégica | Modelación Financiera | Valoración de inversiones | Estrategia Financiera | Power BI | Educación | Oil & Gas |
1 年Anders Liu-Lindberg you have highlighted a critical point to keep in mind if you are planning to relay on #AI in your daily work. I could say that lots of companies don’t have their historical data well-structured in order to make basic analytics (don’t know whether in your country is the same). Now, with the appearance of Tools like Chat GPT (and all its new versions) companies should put in place data management strategy as soon as possible, to get the full potential of #AI. Also, all professional must acquire the technical knowledge to understand whether the output makes sense.?
CEO @JumpUp Global | Finance Career Coach | ??I help aspiring finance professionals in becoming CFO Ready Leaders | ??120-Days CFO JumpStarter Program | ?? Certified MBTI?Practitioner
1 年Agree ?? Anders. We really need to start rationalizing the hype of ChatGPT. It is NOT a superpower and it is far far from replacing human. It is definitely a strong tool in generating massive data but it is still up to us to choose which data is applicable and without the human's actual experience and knowledge in the particular sector, it would be challenging for ChatGPT or AI in connecting the dots and let alone delivering the final product.
NJ Cannabis Manufacturing License Holder??| Fully Operational??| I can help you launch your brand in NJ??
1 年Please scream this louder for the folks in the back! The fact is that any machine is only as good as its data and it calculating "algorithm". Decision making LP's have been around for some time and anyone who has used them knows they need a human touch to vet them. If you relied purely on what the algo said, you'd be in major trouble.
Advance Your Finance/Data Career ?? with English Communication Skills ?? | Specialist English Communication Skills Coach
1 年It's good to be sobered of all the hype around ground-breaking AI tools, Anders. All useful tools need a sturdy platform on which to operate. The way I think about AI is as a magnification effect — if your processes are well organised, they magnify the process and improve efficiency; conversely, if your fundamental processes and data has flaws, it magnifies those and creates more problems.
Senior Manager at Bayat Rayan
1 年This a just in time reminder for those who believe myths about AI