What do you do if your non-technical stakeholders need to understand Machine Learning insights?

Leveraging the AIM (Audience-Intended Messaging) framework is a great way to communicate with non-technical stakeholders. It helps you get into the shoes of your audience and have better messaging!

Quite often, you are presenting to a business driven audience. I would say it helps to re-phrase a lot of the insights and ideas into potential business impact or outcomes depending on the stage of the projects

Use visualization techniques, to present insights to customers. List out inferences in simple language to avoid confusions. Do not use mathematical or technical words in presentations.

It is important to rebrand machine learning as high school maths. We have all done Statistics in school so bringing back that knowledge important. So doing a quick tour of Standard deviations, Distributions, Mean, Mode will enable the non technical stakeholders to feel empowered. It is something we all have learned in school. Now that they recap their learnings, we need to simplify the ML insights as some form of statistical insights. For example, what is an anomaly? It is something that can be found in a Histogram. In addition, I will also recommend the non technical stakeholders to read my book, "Algorithm Diaries" which explains ML algorithms as simple stories and analogies :)

Okay, this is a personal one. Don't be the stats and ML expert that nobody wants to discuss with because they're afraid of being overwhelmed with complex math concepts. Three pieces of advice I can share from my experience: Explain complex concepts in simple terms so that everyone can participate in the conversation, regardless of their background. Listen to the product, business, and process stakeholders; they may have a lot more knowledge about the specific domain. Question whether this is really an ML problem, or if it can be solved in a simpler way. You can often determine this from the problem framing alone. This perspective opens up a lot of opportunities to transform processes, products, and services within any business.

Opacity is major issue for AI/ML adaptability and it is important to bring explainability. There are several ways to bring the same. Explainability has become so important that there is a term called XAI ( explainable AI). What has worked for many folks - listing the variables used. The weightage for them and then how funneling happens and then what is the patterns , for this workflow tools as well as data visualization tools works best. Even excel can do the job. Also showing what if analysis with certain variables can give a major boost in understanding of the model and outcomes associated with it.

Use data visualization techniques such as charts, graphs, and infographics to present Machine Learning insights in a visually appealing and easy-to-understand format. Visual representations can help stakeholders grasp complex patterns and trends more effectively.

Using visual aid can help non tech stakeholders visualize what's happening underneath. Using charts, confusion matrix, histograms, or graphs can help them understand better. If possible, you can show the progress of algorithm via graph at each step. For example, you can plot iterations for gradient descent or Normal function in case of anomaly detection. Using visuals will help them understand the work of algorithm in a better way.

Be sure that the way you show the concepts can be visually explained, a good rule of thumb is that you try and practice with someone that is not familiarized so you can see you are not overusing terms.

In my experience, I have found that using visual aids has been extremely helpful in explaining which features are driving the predictions or outcomes of a machine learning model. The visuals have allowed me to better understand why the model has included certain variables, enabling me to perform a high-level sanity check based on the business context. Visualizations have proven to be a powerful tool for data storytelling, as they can make complex concepts and model insights much more accessible and understandable. They have been invaluable in helping me communicate the key takeaways from the machine learning analysis to stakeholders and decision-makers.

Real-world examples help spread the insights to people. Imagine sifting through endless news articles. ML in news apps filters the chaos! It analyzes what you read, clicks you make, and even location (weather news for your city!), then recommends stories tailored to you. This personalizes your news feed, exposing you to a range of interests while keeping you informed. It's like having a news curator, but powered by machine learning. Here, we can point to the different problems of how curation to your preferences can lead to biases, letting the user know that ML isn't a silver bullet.

The first lecture we heard in the class was how ML replicates neurons in a brain. Given that explanation has helped many people to understand much of ML algorithms, a real world example can be inferred as the best tool for people outside the field to understand the ML insights. In addition, this understanding would allow the stakeholders to further explore the potential of the project.

I perceive Machine Learning (ML) as a novel software solution applicable in cases where traditional algorithm-based applications fail due to the complexity of paths the logic would entail. I often illustrate this concept to customers using the example of distributing containers on a ship, considering various factors such as: - Loading with balance to prevent sinking. - Optimizing the load/unload time to minimize the time spent at docks. - And so forth. This illustration helps customers grasp the idea that creating an algorithm to solve such a problem is practically impossible. Instead, humans learn patterns to tackle it, and AI functions similarly by learning from these patterns

While drawing parallels between the ML jargons and everyday term, using real world examples makes them more relatable. By using storytelling to convey the processing of ML model, you can make process memorable and relatable. You might want to use relevant case studies along with visual aids to help them understand the working of ML algorithm. Make sure to not use acronyms.

Grounding ML insights in real-world examples bridges the gap between abstract concepts and practical applications. 1)Contextual Understanding:Real-world examples provide context,helping stakeholders grasp how ML insights relate to everyday scenarios or business operations. 2)Relevance Demonstration:Examples illustrate the practical applications of ML insights,showing stakeholders how they can address specific challenges or capitalize on opportunities. 3)Concrete Illustration:Examples makes abstract ML concepts tangible,enabling stakeholders to visualize how algorithms analyze data and generate insights. 4)Problem-Solution Alignment:Real-world examples demonstrate how ML can solve real problems, aligning ML insights with stakeholders' goals.

Talk about outcomes, from the point of view of your audience. I have recently worked on a solution to accelerate the resolution of support tickets. At its core was all about classifying "urgent" vs "non urgent" tickets. We first discussed our findings with end users using a confusion matrix, arguing the importance of "recall" vs "precision". They didn't seem particularly interested. I then changed tactic and told a story that started with their current challenge (it takes a long time to go through all the tickets, wasting time on non-urgent ones) and show how, if we got the recall right, they would have been able to focus on a much smaller inbox and get through all the urgent issues faster. That got them engaged!

Often, my tech colleagues and I fail to highlight the benefits of a well-architected and high-permforming solution, whereas the business people primarily prioritizes tangible business value over technical aspects. Then, we frequently observe that technical professionals translate these benefits into abstract business advantages, rather than providing concrete values. I suggest that when selling AI models, we also incorporate domain-specific performance indicators. For instance, certain retailers prioritize the conversion rate, which measures the ratio of people who make purchases to those who visit the store. By focusing on the key performance indicators (KPIs) that resonate with your customers, you enhance your selling opportunities.

Emphasize the impact of ML insights on business outcomes. Non-technical stakeholders want to know how ML will affect bottom line, risk management, or customer satisfaction. Discuss how predictive analytics can forecast sales trends or how natural language processing can enhance customer service. Linking ML capabilities directly to strategic goals helps stakeholders see its relevance to their roles and responsibilities.

I shared a case study where ML predictive analytics significantly boosted sales by forecasting upcoming trends in customer demand, allowing the company to optimize inventory and capitalize on market opportunities. Additionally, I highlighted how ML-powered natural language processing improved customer service by quickly analyzing and responding to customer inquiries, resulting in higher satisfaction rates and increased retention. As I linked these ML capabilities directly to strategic goals such as revenue growth and customer satisfaction, the stakeholders became more engaged, realizing the relevance of ML to their roles and responsibilities.

Google Colab simplifies creating and sharing these demos. Anyone can create and share Python notebooks with just a few clicks, regardless of their technical background. This accessibility allows non-technical stakeholders to interact with these notebooks and adjust models' features, seeing how these changes influence values and output graphs in real-time. By engaging with these demos, abstract machine learning concepts become concrete and easier to grasp. Start practicing right now. Open Google Drive, right-click, and select 'Google Colaboratory'. Follow any tutorial by copying a few lines of code into your new notebook. Within minutes, you'll be experimenting with ML models, adjusting parameters, and observing the outcomes in real-time.

Tools such as Jupyter notebook, Google Collab or websites where you can visualise the working of model can be used for interactive demos. You can present relevant YouTube videos to non tech stakeholders. Alternatively, you might want to present a demo to them based on your presentation or experience. Several online labs features hand-on experience to various ML models, wherein users can change parameters and visualise the working of the algorithm. Such simulations ease the complexity.

Create a presentation that will explain everything in a very simple manner. Provide real world examples and ensure that it is relatable to the projects you are working on. After providing a simplistic overview- in depth training is important so that all stakeholders have a grasp on the concepts and can also explain to others what they have learned.

Forget lectures, let's get hands-on! Instead of dryly explaining how a machine "learns," create an interactive demo where stakeholders can play with the ML model themselves. Imagine a simple app that predicts customer churn based on data points you enter. Just like playing with a new gadget, this lets stakeholders see ML in action, transforming abstract concepts into real-time results. It's a fun and effective way to demystify ML and show its practical application in making informed decisions.

Seeing is believing. That might work for some cases especially if the management is not believing that sth. works. This might be even more powerful than lengthy theoretical explanations. If management however is knowledgeable and has bought into that ML and Ai works, then it is not so relevant. I such cases the business case is much more important.

First understand if your stakeholders are interested in understanding ML in a way that helps them thrive in their position. If they are, then half of your work is done. It is valuable to provide this knowledge yourself or someone trustworthy within your team or enterprise. This will help build their trust on you as a reliable ally. Be their partner on their journey than just a person who gives them directions on how to build their ML arsenal.

The best way to help them understand Machine Learning is to recommend ML short courses that are designed to help everyone to learn the ideas involved in it. As an introvert and someone who's busy already at current work, sometimes this is the best option. Mention these recommendations after you try your own explanation, through story telling.

I can't stress enough how good is deeplearning.ai's "AI for Everyone" course by Andrew Ng. Its a short course and results in a better understanding of ML and its impact. Along with this following LinkedIn and other social platforms helps you keep yourself updated.

I have conducted workshops with my client to address the specific gaps in their understanding to continue/start the project. We simplify a lot of concepts with analogy but also warn them that what we are sharing is not true deep understanding and only limited to what is needed for the success of the project

Any technology has its advantages and limitations. ML is no different. Depending on the ML application I would present the non technical stakeholders 3 aspects 1. what ML can do better. You present metrics on where AI performs better than humans/previous tech: XX% accuracy, XX response time , XX% error rate etc. 2. present ML limitations. ML can go wrong in 'certain conditions'. Your job is to 'define' as you can these conditions. The goal is from a business perspective to define processes for mitigation 3. trend on near future improvements/costs. ML is rapidly evolving so whenever a solution is adopted it is important to give a sense how much a certain solution can improve over time (hence reducing limitations as point 2 or costs).

The key is to communicate the information in a clear, concise, and relatable manner without using technical jargon. We need to focus on business impact by explaining how the machine learning model or insights align with business goals or objectives. We need to discuss the impact on key performance indicators (KPIs), efficiency improvements, cost savings, or revenue generation. It would be great if we can frame our discussion onto case-studies before and after the applications of machine learning models and correlate the case-study with business model on similar grounds without ignoring the differences. Most importantly, be transparent about the limitations of the ML model and the data. This helps set realistic expectations and builds trust.

1. Before presenting any insights, know who your stakeholders are, their level of technical understanding, and their specific interests and concerns. 2. Focus on the "So What?": Instead of bombarding stakeholders with technical details, highlight the practical implications of the insights. Explain how the findings can impact decision-making, improve processes, or drive business outcomes. 3. Provide context: Help stakeholders understand the broader context of the insights by explaining the problem or question the ML model was designed to address, as well as any limitations or uncertainties associated with the findings.

Try to define what’s in it for them. When stakeholders see a WiiFM that’s aligned to their priorities they are open to learn anything, not just Machine Learning.