Business & Data Understanding in AI/ML, Part 1

What is the nature of this tech revolution? Part of a larger series on Full-Stack AI/ML in Production.

Intro

What is the nature of this tech revolution? How can businesses leverage AI/ML for impact?

Hotly debated questions. Lots of opinions (...some of which might be full of unearned confidence 😂). Well, I've been in a front row seat for the past eight years working with nearly every AI/ML model type, so I'm going to go ahead and throw my hat into the ring.

The last era was defined by increasing access, this one will be defined by increasing relevance.

Brief History

The internet brought us access to information, then smart phones made information more immediately accessible. Yes, but also, we didn't just get access to information, we got access to options. In fact, this was core to Amazon's value prop in the beginning, and why he started by selling books specifically:

"There are more than 3 million books worldwide active and in print at any given time across all languages... So when you have that many items, you could literally build a store online that couldn't exist any other way."

source: Jeff Bezos, 1997 Interview

Amazon gave everyone access to books at a scale that wasn't previously possible. But the more important point for our discussion isn't that it was newly possible, it is that their focus was to give access to more options. Something newly possible is interesting, but what really gave Amazon value at that time was making options available where they weren't before.

But Today, WAY Too Many Options

The world we live in today has been terraformed by access to information and options. If you pitch someone on giving them more options for just about anything, they might just have a full blown mental breakdown. 😂

With all the options, there's been an increase in responsibilities. One person can do far more than they have been able to on their own. Expectations have increased, and people are collectively exhausted by it all.

Just tell me what I need to know

It's no surprise that the company that brought us the most relevant website results for a query went on to become one of the leading companies in AI/ML. Of course I'm talking about Google.

"The ultimate search engine would understand exactly what you mean and give back exactly what you want."

-- Source: Larry Page, 2006

The pain felt was no longer a lack of options, it was lack of relevant options. The biggest tech companies had a head start because of access to compute and data, they began to train models that could give users information and options relevant to them. The result was faster progress and lower cognitive load.

Where We Are Today

Today, the game has changed in two major ways

1. Larger models require less data to be useful
2. These models, and data sets, are being open sourced for more businesses to leverage at increasingly affordable costs.

This has opened things up for smaller businesses to use AI/ML to impact their businesses as well. There are many other downstream effects, but let's stay focused on business impact for now.

Ok, But How Does this Map to Business Value?

In future articles we will break down models in a more structural way to build intuition, but for this article let's just go one click deeper and think about machine learning model inputs, outputs, and outcomes.

Inputs

For inputs, we have a pile of data. It can be any type of data you can compute: structured data, unstructured text, images, audio, 3D virtual environments, video, 3D video, etc.

Outputs

After a model is trained, what is the output we want? Well, the question is what would make this pile of data useful? If we wanted to simply retrieve data for accuracy, we would use a database not a machine learning model. So what is our goal if not to retrieve data? The goal is to derive relevant outputs from the data.

We don't need models that give us more access to data we could access other ways (like a database), we need models that give us access to the aspects that are relevant to our needs.

Outcomes

This is where the real business value emerges. The ultimate goal of a machine learning model is to produce outcomes that drive meaningful actions or decisions.

These outcomes are where the relevance of the outputs is tested against the real-world scenarios a business faces. Here are some ways outcomes map to business value:

1. Drive Growth: Uncovering Business Insights from Metrics
   • Revenue Optimization: AI models can analyze customer behavior, sales data, and market trends to identify opportunities for optimizing revenue. This might include adjusting pricing strategies, focusing on high-margin products or services, and targeting the most profitable customer segments.
   • Customer Retention and Acquisition: AI-driven insights help businesses better understand customer needs and preferences, leading to more effective marketing strategies, higher retention rates, and lower acquisition costs. This drives growth by maximizing customer lifetime value.
   • Product and Service Innovation: Machine learning can reveal patterns in business metrics that highlight opportunities for innovation, whether in product development, service enhancements, or new business models. This fosters growth by ensuring the business stays ahead of market trends and demands.
2. Lowering Cost of Goods Sold (COGS)
   • Optimized Supply Chain Management: Machine learning models can analyze supply chain data to identify inefficiencies, forecast demand more accurately, and optimize inventory levels. By reducing waste, minimizing storage costs, and enhancing supplier negotiations, these models directly contribute to lowering COGS.
   • Predictive Maintenance: In manufacturing, AI-powered predictive maintenance can anticipate equipment failures before they happen, reducing downtime and extending machinery life. This leads to lower production costs and ultimately reduces COGS.
3. Reducing Cost of Services Rendered (COS)
   • Automation of Routine Tasks: Machine learning models can automate repetitive tasks such as data entry, customer service interactions, and parts of service delivery. This automation lowers the labor costs associated with providing services, thereby reducing the overall cost of services rendered.
   • Optimized Resource Allocation: AI models can predict service demand and optimize the allocation of resources, ensuring that the right amount of labor and materials are available when needed. This efficiency minimizes unnecessary expenses and contributes to cost savings in service delivery.
4. Lowering Operating Expenses (OpEx)
   • Operational Efficiency: By identifying patterns and trends in operational data, machine learning models help streamline processes, optimize resource usage, and reduce waste. This leads to significant reductions in operating expenses, making operations more cost-effective.
   • Enhanced Decision-Making: Machine learning provides actionable insights that enable more informed decision-making, reducing costs associated with inefficient or suboptimal decisions. This leads to lower overall operating expenses by avoiding unnecessary expenditures.

Most people will read the above and think, But how though? Well, as you can imagine all of the above are WAY too much info to cover in this post. I will be covering topics listed above in greater detail in future posts as well. Until then, I would encourage you to use this post as a launching point to talk with ChatGPT or similar to understand it more detail.

Sequence Business Impact

There is a nice logical sequence to leveraging AI/ML within businesses if you look closely at the impact it can create. It's best to start small and sequence for impact. Consider this sequence:

1. Measure first.
   1. Do you know your most important KPIs for growth? Take the time to bring your data together in an quickly digestible visual way to keep an eye on the big picture.
   2. This means collecting data from various sources like Google Analytics, Shopify data, Stripe data, Klayvio, etc. Then aggregating and displaying this data in one place to align leadership. If your just getting started, take a look at Looker Studio by Google. A lot of teams are forwarding their data to BigQuery as an affordable data lake option, then querying this data for visualization within Looker Studio.
   3. I strongly recommend defining your growth flywheel, and making it visible within a dashboard. If you move from funnels to flywheels, you can take advantage of compounding growth. Take it from Hubspot.
   4. After collecting your data from various sources, your opportunities to leverage AI/ML will open up.
2. LLM to chat with your KPIs and metrics
   1. Here's a great example by Google. Honestly, this video is pretty epic.
   2. There's a lot you can do here, but I think among the most useful is to identify your companies main constraint to growth. This is why it is useful to define your flywheel as it can help you get clear on the most important next KPI to improve. The LLM may also have strategic suggestions to help improving the KPI. Remove the constraint and take advantage of the compounding growth over time.
3. Lower COGS
4. Lower Cost of Services Rendered (COS)
   1. Consider creating a process map to measure value added time vs non value added time when fulfilling on work for a client, or manufacturing a product. This will give you some specific targets to focus on for leveraging AI to improve profit margins.
5. Lower Operating Costs

Conclusion

In summary, the relevance-driven outcomes of machine learning models translate directly into business value by uncovering insights that drive growth, lowering COGS, reducing the cost of services rendered, and cutting operating expenses. The sequence I outlined above shows how AI/ML can drive revenue growth, increase profit margin per unit sold, and lower operating expenses to increase overall profit.

As we explore specific models and techniques in future discussions, we’ll continue to examine how aligning these outcomes with key business metrics can maximize their impact on a businesses bottom line.

Next up in this series is Business & Data Understanding in AI/ML Part 2. I will be focused on answering the question: Is my data set ready for machine learning model development? Subscribe to see this post when it's ready.