Here's What You Should Know to Launch Your First AI Pilot Project

McKinsey expects AI to contribute around $13 trillion to the world economy by 2030. Given this massive potential, businesses are keenly looking toward AI adoption. However, switching gears to machine learning is a tall challenge for organizations with legacy issues. That is why their initial concern is to weigh all the risks of AI adoption before making a big move. In such a scenario, an AI pilot project makes a lot more sense to executives who are enthusiastic but cautious of what is coming next.

As a business leader, if you are looking to gauge the threats and opportunities that AI presents for your company, this guide will help you make the right decision.

Here is a list of essential things to consider while starting the first AI project for your company.

Here’s a quick view of things we will go through in this article -

• Start Small
• Choose An Appropriate Business Case
• Choosing The Right Team
• Choosing The Right Approach
• Acquiring Data
• Right Tools
• Communicating The Business Context
• Conclusion

1. Start Smal

Turning your organization around into an AI-driven business is a tempting yet risky affair. There is no guarantee of how all the variables will play out, and you do not want to stake it all in one go. Hence, the best advice that AI professionals will give you is to start small.

Why?

If you fail, it is not detrimental to the whole organization.

If you succeed, you gain confidence & room to venture into newer projects.

For example, instead of automating the whole logistics department of your company, you could start by automating a robot to pass some packages to your inventory guy. When automation on simple tasks gets iterated repeatedly, that is when you get AI-based solutions for more complex tasks.

Another aspect of choosing a small project is to ensure that it has a shorter turnaround time. Ideally, the gestation period should not be more than a year; Anything fewer works. However, it is still essential that you give enough time for the solution to produce results. The key is to find the sweet spot.

Advice #1 – Think small; start smaller.

2. Choose An Appropriate Business Case

"An appropriate business case" is a vague phrase if not described. Every business has different use cases for using AI. However, there are a few generic features to keep an eye on.

The first feature is to choose a business case that is small and simple.

Another feature can be a clear automation opportunity. For example, A simple manual task that is repetitive and inefficient currently presents a good automation opportunity.

Additionally, you should ensure that the AI adoption would lead to some actual value creation. An AI project that delivers no value will eventually fail to create enough enthusiasm in the organization to move ahead with more such projects

Hence, your business case must not be so trivial or impractical that it struggles to contribute to the organization in the long run.

You can measure that created value in terms of the following –

• Cost reduction – Machine learning can contribute by reducing any overheads in the current implementation of a task. For example, a smart AI check on machinery's health can lead to fewer device failures and accidents.
• Faster Execution – AI can prove helpful by making a process more efficient than its current form. For example, a quick allocation of resources can lead to prompt task execution.
• Increased Sales – Machine learning can facilitate better and more personalized recommendations that increase up-sales and cross-sales. For example, Amazon sells more products with the help of a smart recommendation system.
• A new product offering – Sometimes, AI adoption might lead to a completely new product offering that was not possible before. For example, an investment firm that starts offering a personalized portfolio creation based on your background and preferences uses AI.

Then, another thing to ensure is that your pilot project is relevant to your industry. You are starting a pilot project inside your company. It means that the project should somehow further the company's vision. It should be scalable to contribute at the industry level.

It shouldn’t be so unrelated that it stands out as a business on its own. Like, if you are an agri-tech business, it would make more sense for your AI engineer to develop a smart agri-tool. Instead, if you start developing an AI-driven smartwatch that keeps your employees healthy, that’s a separate business altogether.

Advice #2 – Start with simple tasks, create value, & keep it relevant.

3. Choosing The Right Team

Choosing a team for your project might come after you have decided on the project. It should be a concurrent process ideally. Knowing your team's capability serves you better while picking up a project from multiple options.

As you are building a project, a balanced engineering team is better suited than a team of only data science professionals. In the end, it is the engineering team’s job to build coherent and usable products. Thus, choose people who have engineered products before.

Similarly, the leader should be a person who can easily liaison between the business and the tech side of the pilot project.

If you have the budget, it is not necessary to have your in-house team do all the job. You can collaborate with external organizations that specialize in AI exclusively. It would be a faster approach to onboard an expert AI engineer to help you out rather than training your own team members to be experts.

Finally, everyone should be just as excited about the project as you are. There is no AI culture in your organization as you are just running a pilot. You need your team members to root for you and stick with you until the desired result achieved. Resilience is the key to success.

Advice #3 – Choose an engineering team that is passionate about AI, collaborate with experts, and choose leaders and members wisely.

4. Choosing The Right Approach

Running a pilot project means that you are on a short leash. You must test the idea fast without wasting the time and resources of the company.

The Lean approach to product development can be especially helpful when you are looking to minimize resource wastage.

• Ideally, you should have your finished project prototype within 4 – 8 weeks. The time can be lower than this if the project is easier to execute. The goal here isn’t to make the perfect product in one go. Rather, it is to validate the proposed solution.
• Another 2-3 weeks can be employed for testing the prototype extensively before adopting it in the operational chain of the organization.
• A good strategy would be to automate simple tasks like data updates, online model training, and code updates. It should take another 3-4 weeks of your time.

Choosing the right approach for your AI project also depends on choosing a suitable algorithm for the problem that you are solving. Initially, you can pick up anything that works for the problem. A suitable choice would be an algorithm that learns faster with fewer data.

Still, as you are not directly aiming for the best right away, a lot depends on data availability and the data cleaning and organizing ability of your AI professionals. If one algorithm is more resource-intensive than the other, the project team can choose to save resources by slightly compromising accuracy.

Advice #4 – Go for lean development, and choose the right approach, no need for a top-notch AI model right away.

5. Acquiring Data

Now that you are switching to becoming a data-driven organization, you will need to acquire quality data to solve your problems using machine learning. Be informed that every problem does not need tens of thousands of data rows. Even a thousand rows of data can be enough to get you started.

The initial goal is not to develop a state-of-the-art model. It is to train a model that gets by just enough. Once your pilot is at that stage, you can adopt continuous model improvement via more data and hyperparameter tuning.

If your organization has not been collecting data in the past, one of the initial tasks for your project team is to set up a system of data collection that gets you started. Having a clear set of project goals will inform the team in deciding the data points they must collect.

You might set up an ad-hoc system of data collection initially. It needs to be accurate, nonetheless. Once your solution is validated with enough confidence in its value creation, the project team can move forward with more robust systems of data collection.

Advice #5 – Data is essential, establish a system of data collection.

6. Right Tools

No matter the craft, every professional in every field needs tools that suit the job. AI professionals are no different. After deciding on the project scope and goals, the next thing on your list is the tools and technologies that you are going to employ to meet your goal.

You should keep the following in mind while selecting tools & technologies for your pilot project.

• They should have the capabilities that your project needs.
• There should be a rich knowledge base around such tools.
• It will serve you better if there’s a vibrant community of developers working in and contributing to the tech of your choice.
• The tools/technologies should be mature enough to have any critical bugs removed.
• Expert AI professionals in those technologies should be readily available.
• They should be robust and provide for easy migration in case of any unforeseen situation.
• They should be well within your resource constraints.

Considering these requirements, technologies like Python, R, AWS, Microsoft Azure, TensorFlow, PyTorch, etc. have become quite popular with AI engineers .

Advice #6 – Choose tools & tech according to your business needs and tech capabilities.

7. Communicating The Business Context

The one thing that engineering teams are not good at is mainstreaming the business context without being explicitly told to do so. And it’s not their fault. They are more inclined to build world-class products that give you an edge over the competition.

Thus, add someone in-between tech & business, it becomes the responsibility of the project leader to communicate the business context to the engineering team.

Let them know the organization's goal for the project. What is the product vision? What is the North-star of this project? When they know the strategic priorities, it will enable them to make tech decisions that are more aligned with the business context.

Similarly, inform them of the current problem that you are trying to solve using machine learning. Sometimes, there are better and cheaper solutions than using AI right away. Your tech team can act as a filter that ensures that the problem you are looking to solve actually needs the ML approach.

Communicate the performance metrics and KPIs that you will use to gauge the success of the product. Inform them regarding the acceptable metrics that the business cannot do without. Once they are aware of the expected outcome, they can better equip themselves to achieve that. They will also give you feedback regarding what can be done with the current data and what more data points, if any, you need to collect.

Better communication between the tech and the business team ensures that the end product is aligned with the larger vision of the organization.

Advice #7 – Give your tech team the business context to help them build a desirable product.

Way Ahead

AI has revolutionized several industries and will continue to do so in the foreseeable future. Any business that lags behind the AI game will not survive the future for long. Hence, it is prudent for businesses to chart an AI adoption strategy. A pilot project can be the right start to test waters before deciding to sail in. The right place to start is to look at current operational inefficiencies and rope in specialists to see if machine learning can help.