The Ultimate AI Checklist – Part 2

Building a successful AI prototype and beyond

Hi there,

Today brings you Part 2 of the Ultimate AI Adoption Checklist – (click here if you missed Part 1).

To recap: The goal of this checklist is to help you successfully launch your first AI use case - today with a special focus on prototyping and beyond.

There's also a little personal note in this newsletter.

More on that below!

🚨 I'm thrilled to be speaking at DSC Europe, 20 - 24 Nov in Belgrade, which has an incredible lineup of speakers including Cassie Kozyrkov, Jepson Taylor, and Tarry Singh. If you want to join, use promo code DSC-AI4BI for a 20% discount when you sign up!

On a personal note

This newsletter has grown beyond its original topic of "AI for Business Intelligence".

When I started the first issue over a year ago, ChatGPT didn't exist and Generative AI was a geeky thing. But now things have changed a lot. This has led me to rethink the main theme of this newsletter.

Going forward, this newsletter will be called AI for Business Growth and will explain how AI - especially for, but not limited to, data analytics - can help you grow your business.

I hope you like the new design:

The actual content won't change much - I'll still be sharing practical tips, frameworks, and AI use cases. The new title should reflect that better.

Thank you for being an avid reader and I hope you'll enjoy AI for Business Growth as much as you enjoyed AI for BI Rocks!

What do you think? Hit reply and let me know!

Now, on to the checklist!

The Ultimate AI Adoption Checklist - Part 2

7. Scope a Prototype

[ ] Check if you can build a prototype for your use case

A prototype is a small, early version of your AI use case. It’s typically not integrated into your existing system landscape. But it’s user facing. The goal is to test if your idea is actually as valuable as it sounded on paper.

Example: If your goal is to build a customer churn predictor for your CRM, your prototype could be a simple prediction model that gives you churn probabilities for some customer records in an Excel spreadsheet. What would you do with them?

[ ] Set strict time and financial bounds

If your use case doesn’t allow to be built in a reasonable timeframe or budget, you should reconsider scoping it down - especially if it’s your first try. Make it an Atomic AI use case.

What is reasonable? It depends. From my experience, most companies go well with the 20-20 rule: Don’t spend more than 20 working days and not more than $20k to ship your first iteration.

8. Aim for Augmentation

[ ] Keep a human in the loop per default

In the “old days” of AI before ChatGPT, augmentation often meant integrating AI into a process and then having humans on the vendor or technical side try to improve it while the system was running. Especially with Generative AI, we realized that it’s often better to have AI work alongside humans by design.

Example: Instead of aiming for an AI system that automatically predicts churn for all your customers to trigger some automated actions, consider an AI system that assists customer service agents looking at a client in a CRM.

[ ] Design AI to provide actionable insights or assistance

If your goal is not to automate, but to augment in the first place, design your AI to offer insights that users can act upon, making their tasks easier and more informed.

Example: Instead of just flagging whether a customer will churn, the AI could suggest reasons for churn and highlight specific countermeasures in real time.

[ ] Test AI-human interaction for effectiveness

Once the AI system is in place, try to find a way to monitor how it interacts with human users. How can you verify that it's adding value and not causing confusion or extra work? Simple metrics will give you an idea: How often are people using the service? Are they sticking with it?

  9. Iterate Often

This picture says it all:

[ ] Gather user feedback post-implementations

Hard truth: the first implementation of your AI use case will probably suck. There’s to much complexity and uncertainty involved.

Consider the “v1” of your use case rather the beginning as the end of your journey.

Having an agile mindset is great, but avoid the 'infinite loop' of iteration.

Frame it as 'constant improvement', not constant 'work in progress.'

To do this, you need two things.

  • A clear goal: This always comes down to the original pain point you want to address, your destination (what & why).

  • User feedback: People interacting with the AI system will be your primary data source for improvement. This feedback informs the path to achieving your goal (how).

[ ] Analyze AI performance against success metrics

Every AI solution should have defined success metrics. Compare your performance regularly. These can be technical metrics (such as accuracy) or business metrics (e.g. time saved, output volume, etc.).

Example: For churn prediction, success metrics might include the accuracy of the predictions, or the amount of revenue saved by preventing churn.

[ ] Make necessary refinements - fast

The key here is to move fast. Faster iteration = faster learning.

You want to strive for improvement, not perfection.

You don’t have to act like Big Tech shipping a new version every hour, but try to keep your iteration cycles short. A typical sprint could range between 2-4 weeks, allowing you to ship a new improvement every 1-2 months.

10. Engage Key Players

[ ] Identify stakeholders across departments

AI solutions typically impact multiple parts of a business. It's important to identify ALL stakeholders across different departments - and NOT treat them all the same way. Use a tool like the stakeholder matrix to engage people.

Example: In the churn use case, stakeholders might include obvious teams like sales (act on the churn scores), marketing (design retention campaigns), or customer success (help customers). But it could also involve other departments like legal or compliance (for processing customer data).

[ ] Organize meetings to align as a team

It's usually a good idea to have a kick-off meeting to get key stakeholders on the same page. Depending on your use case, you can skip this step for the prototyping phase. But never skip it for the pilot or production phases - these meetings will be your platform for ensuring alignment, support, and collaboration.

[ ] Establish a feedback loop

You don't have to call every stakeholder each week - make sure you know who plays what role and treat them as such.

11. Assess Risks and Ethics

[ ] Identify risks and ethical challenges

Every innovation carries risks. The question is: Can you manage them?

Look at risk from an technical, and organizational perspective. What might go wrong and what would be the impact? Consider legal implications, especially if your AI handles sensitive data. Ethical challenges can emerge from biases in AI predictions or how AI might affect customer relations.

Example: For customer churn prediction, if the data used to train the model has unintentional biases (e.g., it predominantly represents a particular demographic), the predictions might be skewed and not universally applicable. This can lead to ethical issues if certain customer groups are treated differently based on flawed predictions.

[ ] Develop strategies to mitigate identified risks

Once risks are identified, you need to plan how to manage them. One way to derive the risk strategy is to look at risk impact vs risk probability:

The goal is to be proactive, not reactive, in managing potential pitfalls.

12. Plan Training Sessions

[ ] Determine training needs based on user roles

Team members will use the AI tool in different ways. Some may use it every day, while others may only need a basic understanding. By understanding the needs of each role, you can provide the appropriate training for everyone.

Example: Sales representatives might need detailed training on how to interpret churn predictions and take proactive measures, while management might only need an overview of how predictions impact overall business strategy.

[ ] Organize hands-on workshops or tutorials

Active learning is better than passive instruction. Holding meetings where power users can mingle and share ideas are as important as demo sessions to onboard new users. Giving people a place to ask questions, and receive immediate feedback can greatly increase their confidence and skill level.

[ ] Provide resources (manuals, FAQs) for self-paced learning

Everyone learns at different speeds, and having resources available helps team members go back to topics they find difficult. Manuals give step-by-step instructions, while FAQs answer common questions.

Key Takeaways

This concludes Part 2 of the Ultimate AI Adoption Checklist.

To download the full list in Word/PDF format, click here:

Feel free to adapt the list to your own needs.

For any other questions, feel free to reach out!

I'm always happy to discuss how AI can augment and grow your business.

See you next week - at AI For Business Growth!


PS: If you found this newsletter useful, please leave a feedback! It would mean so much to me! ❤️

Want to grow your business with AI? Here are 3 ways I can help:

  1. Book a meeting: Let's find out how I can help you over a coffee chat (use code FREEFLOW to book the call for free).

  2. Read my book: Improve your AI/ML skills and apply them to real-world use cases with AI-Powered Business Intelligence (O'Reilly).

  3. Follow me: I regularly share free content on LinkedIn and X.