Before Jumping into AI, Conquer Your Data & Analytics
Why is it that of all the thousands of artificial intelligence (AI) use cases being considered today, only 5% are in production?
Because too many companies jump into AI without planning. They jump straight from identifying a business requirement to data modeling. They skip the steps in between that are vital to rolling out a successful AI solution.
AI, of course, is hot. Companies hear that their peers or competitors are doing something in this space and they say, “Hey, let’s do this particular AI project.” They rush in without making sure AI can deliver on the business needs and before knowing that the company has the infrastructure and data to support it.
Some AI projects fail because they don’t receive buy-in from end-users who fear the technology is here to replace them. Others fail because companies don’t anticipate what they are going to do with the mountains of insights and analysis that AI uncovers. They start their AI projects without asking how to deploy it in production.
Leaping without looking
Consider a bank, for example, that creates an AI app to automatically detect fraud or automatically predict when borrowers are about to default on their mortgages. These are sensible use cases, right? These are useful things to know if you’re a bank, right? Making the business case for the AI app is likely straightforward.
But what if this bank does not have the infrastructure, staff, budget and other resources needed to do anything about the warnings that the AI app generates? In other words, what happens when the bank learns on a Tuesday before lunch that 124 of its borrowers are about to default the next day? This is why the majority of AI use cases are not in production—they haven’t been thought through.
First, conquer your data
As we say in the business, you can’t have AI without IA. In other words, you can’t have artificial intelligence without information architecture. So, before you jump into an AI project, you must conquer your data and analytics.
That’s what one company did. Once they conquered their data, they started cutting costs, increasing sales and delivering better customer service. If you want to emulate their success, follow these five steps to use your data to create a leaner commerce company.
The company in question shall remain anonymous. They invent, develop, and sell innovative kitchen products for Canadian and global markets. They sell their products to chefs and consumers through Amazon and their own online store. And they wholesale their products through brick and mortar retailers across the country.
The challenge this company had come in the form of last-minute surprises from its largest customer. Walmart would order tens of thousands of a given SKU and would then, days before taking delivery, change their order.
This is what it looked like.
- Walmart orders 40,000 can openers
- Company instructs their manufacturer in China to fulfill the order
- Shipment arrives from China three months later
- Three days before the company ships the can openers to Walmart, Walmart reduces the order by 25%, from 40,000 units to just 30,000
As you can imagine, the company now had the challenge of finding a home for their excess inventory of 10,000 can openers. They solved the problem by leasing warehouse space. But over the space of hundreds of orders spread over many years, the company gradually developed an $8 million nightmare. Excess inventory, excess warehouse space and the overhead costs that go with them were harming the company’s bottom line and hindering its ability to take advantage of opportunities.
The fundamental problem the company faced was a lack of insights into their data. “What really keeps me up at night is not being able to understand my inventory and what’s sitting in my warehouse and what’s not,” is how their general manager put it.
He would walk the warehouse floor with a clipboard that supposedly told him everything he needed to know. But that information was based on spreadsheets and manual entries and a manual pulling of data from one database or another. He was never 100% confident in the data he used to make strategic business decisions.
The company needed a way to become leaner and more agile. So, they hired a firm that specializes in data analytics and AI. Here are the steps they took. They’re the steps you can take to conquer your data and analytics before launching your AI initiative.
Step 1: Name your business objective
Gather your key decision-makers and discover the business objectives that will impact your business. Aim to come up with at least four objectives.
Then dig a little deeper to understand if you have the data and infrastructure you need to achieve these business objectives. Be realistic. For example, someone at the table might say, “We need an AI-powered drone that flies around our warehouse and scans all our inventory.” But the chances are that this represents a higher-risk project because you don’t have any of the infrastructure in place to make it happen.
A more realistic business objective looks something like this: “We have our data in our ERP. Yes, it’s in multiple tables, but if we can combine these tables, and if we can get use AI to collate the data and organize it in a way that makes sense for our business, and present it to us in meaningful ways, we can achieve our business objective.”
Step 2: Build a business case
Take each business objective and plot it on a grid. Make the Y-axis the anticipated value and the X-axis the anticipated risk.
In the top-right quadrant, you’ll find a cluster of the business objectives to focus on. Decide which objective or objectives you want to move forward with by building a business case for each one.
The easiest way to do this is to quantify the cost of the problem and the cost of the solution. For example, if the problem costs $8 million and the fix costs $300,000, the ROI is likely worth it and getting executive buy-in should not be an issue.
Step 3: Build a proof of concept
Build a proof of concept for solving your challenge. This includes getting the data and putting it into a visualization tool to see how the proof of concept is going to deliver the results you need.
Then pass this proof of concept past your executive team and get their approval to move forward with the project. They may have some objections. They might decide that the project is too much work, for example. Or that you will hit too many roadblocks. Or they might even decide that your company does not have the data you thought you had to make the project successful.
Most importantly, they might discover, before you have invested vast sums in development, that your company lacks the resources to act upon the insights that your proposed AI solution delivers.
Your proof of concept is a vital step. It demonstrates how what you intend to build solves your business objective. And it’s the stage where your executive team decides to move forward or not. Don’t skip this step.
Step 4: Develop the solution
The next steps are straightforward. You build what you designed in your proof of concept. This involves:
- gathering your team
- creating a budget
- creating a schedule with milestones and deliverables
- developing a project plan
- finalizing the design
- building the AI solution
- buying hardware
- testing the AI solution with users
- testing the final data to make sure it is accurate
Step 5: Launch the AI solution
Once the development stage is over, you roll out the solution. This step includes training users, conducting ongoing quality assurance testing and troubleshooting bugs.
The company that took these steps shrank their $8 million problem down to a $1 million problem. They got rid of their excess warehouse space and became leaner. Today, their AI solution gives them a better understanding of what is in inventory and what isn’t.
They have a better handle on what customers have ordered in the past, in what quantities, and when, which helps the company anticipate demand and create accurate forecasts on what to order to keep pace with demand. And they now, thanks to a well-thought-out AI solution, maintain optimum levels of inventory (not too much, which is costly; not too little, which hinders sale and hurts the customer experience).
If you operate a business that could benefit from the insights, automation, and speed that AI promises, avoid the common mistake of thinking you know the tool to use. And also avoid the common blunder of taking IT’s lead and jumping straight to a proof of concept.
If you want to deploy AI to solve a business problem, start by identifying the business objective you’re trying to solve. Create a business case. Build proof of concept. Get executive buy-in. Then build the solution and roll it out. You’ll be successful, sooner.