AI isn’t the future—it’s the present. A recent study reports that almost 80% of respondents use AI in at least one business function, with about half of those organizations already expanding usage to three lines of business. That expansion is happening at an incredible rate. The genie is out of the bottle, and we can see the landscape of every industry changing around us.
Earlier this year, I joined Amplitude senior campaign manager for a to discuss how data teams can use AI to dramatically expand access to data and personalize the process of turning insights into action.
The age of AI is an opportunity for unprecedented efficiency and innovation, but it can’t be deployed haphazardly. To be successful, you need to establish thoughtful guardrails. Here are 5 of the highlights:
1. Establish clear boundaries
The uses of AI are limitless. But if your team wants to be effective in your AI implementation, one of the most important questions to answer is how not to use it. Classic, common-sense ethics are a good place to start this conversation.
For example, our team at includes several Pulitzer-winning journalism brands. In that capacity, AI can be extremely useful at pulling analytics, reducing operational costs, and personalizing news distribution. But it seemed clear to us early that there was an ethical line to draw around AI replacing human journalists as writers.
To us, there is an irreplaceable human component in journalism. Integrity is important to readers and we knew we needed to ensure that our AI innovation didn’t also have adverse effects. Removing an empathetic person from the writing process would damage a critical component of our products: trust. So we drew a hard line. We made it clear to our journalists (and their audiences) that we would never use AI to replace a local reporter—only in a way that would empower them.
Your team can do this too. Start by asking if anything is out of bounds. Consider how AI could accidentally use data in a detrimental way. Limit your scope to just the workflows and tasks that are consistent with your ethics.
2. Start new AI initiatives with a clear focus
After your team has ruled out some AI applications you won’t do, it’s time to set up clear rules for the projects you actually will take on. This is as easy as establishing a small, focused set of questions that help you narrow the path of your AI journey to something that clearly moves the business forward.
There’s no right number or right type of question here. Just make sure that you’re starting your AI projects in a way that has direction and maps to metrics you’re already tracking, so you can measure impact.
If you want some help, here are the four questions that our team asks to make sure our AI ideas are correctly focused before we kick them off:
- How can we expand our reach and engagement?
- How can we encourage readers to stay and subscribe?
- How can AI interact with our data to get us information so we can make better decisions?
- How can we use AI to improve operational efficiency?
3. Personalize customer experiences with data-driven insights
Every business is looking for a way to use AI to make stronger connections with customers and prospects. With so much data available about user attributes and activities, smart data leaders are finding ways to develop unique experiences based on different sets of user characteristics.
For our team, that personalization journey started with an audit of our data stack—we were using as a data warehouse and Amplitude for , , and . We wanted to review all the available data, identify simple steps we could take to differentiate types of users, and offer those groups different things. Looking at the data we had available, we decided to start with content recommendations.
Rather than blasting every visitor with the same recommended pieces of content, we decided to build an engine that would consider unique factors like what other articles a user had read and what was popular with similar readers. We started with baseline metrics that we already tracked (subscriptions, retention, newsletter signups, session duration, etc.) and ran experiments with our recommendation engine to see if we could improve those numbers.
We didn’t need to start collecting new data or plug in new technologies. We didn’t need to build new dashboards to track results. We just plugged in data about what made the customers unique and let AI cater existing features to those characteristics. We analyzed results and kept experimenting until we had content models that suited several different types of readers—free visitors, paid subscribers, newsletter members, and a lot more.
Right away, we saw increases in session duration and engagement. Readers were seeing content that interested them without feeling spammed by impersonal promotions. We started with common sense and kept enhancing after that. Our big secret is that we didn’t overthink it.
4. Democratize AI tools to empower internal teams
People are the reason for your success. Your AI initiative won’t go anywhere unless it offers value to the people on your team. The challenge is finding a way to bring AI to your colleagues rather than requiring them to force their work into an AI-powered environment. The less you require people to change, the better.
Too many teams today have a gatekeeping problem when it comes to data. Colleagues feel like they need to be a data scientist to use the tools. It’s the data team’s responsibility to destroy that myth. The way they’ll succeed is by building data products that anyone can use on their own to accelerate everyday work.
An easy way to make AI tools more approachable is to work with colleagues to uncover their most common time-consuming tasks. If there’s any way to use data to accelerate that process, work with them to build it. The same goes for answering common questions with data: if there’s data that a person or a team frequently needs, find a way for AI to assist with that request.
Beyond simple access, a lot of first-time AI users need encouragement. Leaders should celebrate small wins across all lines of business equally here. If AI can help anyone save a few minutes on a task they perform multiple times a day, those savings will add up quickly. The more comfortable your team gets with managing AI basics today, the more comfortable they’ll be experimenting with new projects in the future.
5. Navigate challenges and mitigate risks with a strategic AI roadmap
AI is so powerful and so complex that it’s easy to bite off too much. It’s extremely common for companies to hurry to implement AI tactics without having a strategy to tie those tactics to critical business goals. That’s an easy way to make a mistake.
For any data team looking to implement AI, let me give you a piece of advice: AI doesn’t have to be flashy. Start with an AI strategy that meets you where you are now and expand in small steps. Think about what you want to achieve with AI (satisfied customers, stronger products, etc.), not the flashy tactics you’ll use to get there.
The biggest early wins you’ll see are often just accelerating routine, repetitive, and time-consuming tasks. You can still have bigger, splashier projects on your roadmap for later, but any initial wins won’t come from net new creations, they’ll come from doing the same work more efficiently.
Ready to build a strategy around AI deployment? This covers all of Amanda's best tips along with stories about what has worked well from experiments at McClatchy. Grab whatever drink helps you focus and check it out!