Data Driven - Our Blog

We've all seen the hype surrounding AI – promises of revolutionizing the way we work. But when it comes to applying these technologies to our own data, it's easy to get caught up in the glitz and forget the nuts and bolts that truly drive success. This year alone, we've tackled over a dozen AI projects using the Explore Assistant within Looker, and the journey has been a masterclass in realizing the power of AI while also understanding its limitations.

The industry is at a juncture where LLMs (Large Language Models) are getting smarter, but we're still miles away from a "point-and-click" solution for your own first-party data. This is where leveraging the power of a semantic model like LookML can become your secret weapon. Think of it as having a conversation with your company data – and LookML becomes your translator, turning queries into actionable insights.

So how do you make your AI project a success? Think beyond the initial spark and consider these factors that will guide your project's success:

1. Foundation First: AI Exposes, It Doesn't Fix

Imagine a beautifully designed kitchen, but the plumbing beneath the floor is in disarray. That's how your AI project might turn out if you're not setting a solid foundation. AI is great at exposing inconsistencies and data gaps. The best way to avoid these hurdles is to invest in good data hygiene practices, reevaluate your current architecture, data models, and internal processes.

Even if you're not diving into AI immediately, planning for the future to stay ahead of the technology curve, especially as rapid changes continue, is crucial.

2. Context is King: Feed your AI the Right Information

Your LLM needs a comprehensive understanding of your data to function effectively. This means providing it with the right context, which includes details like field names, descriptions, and data types. Imagine you're trying to understand a complex sentence without knowing the meaning of certain words. That's what happens to your AI if it doesn't have enough context.

For example, if you want to understand how marketing campaigns affect customer sales, your data should include details like:

Specific dates: created_date, updated_date, campaign_start_date, etc.
Clear field names: Instead of generic names like "date," use specific names like "campaignStartDate" or "customerAcquisitionDate."
Descriptive field definitions: Explain what each field represents and the units of measurement (e.g., "Date the customer was acquired," "Total sales revenue in USD").

Remember, AI isn't magic – it's simply making educated guesses based on the information you provide. That's why every bit of context you give it helps improve its accuracy.

3. Keep Refining your Data Model

We’ve gone this far with analogies so let’s keep going. Think of your AI model like a new car. It runs smoothly out of the dealership, but to keep it performing at its best, you need regular maintenance. That's exactly what continuous model refinement is. It's like giving your AI regular oil changes by consistently feeding it new information and real-world questions.

Every new business question you add and the corresponding LookML query you include is crucial. Looker's Explore URL sharing feature acts as a detailed log, recording each field, filter, visualization setting, the reasoning behind it in the generated query. This helps the model "learn" from every interaction, ensuring that it runs more efficiently and provides the most accurate answers possible. The more information you feed it, the better your AI's performance will become over time.

4. Talk it Out… Like Onboarding a New Hire

Another valuable way to give context to your AI is to simply record and transcribe a detailed explanation of your business logic, like you were onboarding a new hire. Think of it as providing a giant prompt for the AI. It might seem a bit tedious, but it allows your LLM to learn the intricate details of your specific workflows and business processes, resulting in a far more knowledgeable assistant.

5. Change Management: Embrace the AI Journey

Don’t jump into AI unprepared! You'll need an internal champion – someone who deeply understands the business and has a comfortable understanding of basic Python and LookML (enough to manage example uploads and updates). They’ll help bridge the gap between technical needs and the overall business vision. It might also mean involving a dedicated team depending on company size and resources.

Once your team is in place, consider how to integrate AI into existing workflows. Think about rolling out initial use cases and testing the waters with early adopters who are willing to embrace this new approach to data exploration.

The more people you involve in the journey, the more enthusiasm they’ll have for evangelizing the benefits. Make sure you also have training sessions and support mechanisms in place for ongoing onboarding and continuous improvement.

And finally, always remember to track! Monitor your existing ticket requests for ad-hoc data requests and compare those to the new needs met by your AI. This will help determine the true long-term value of your investment.

Conclusion

The future of AI in business intelligence is full of potential. By taking the steps outlined in this blog — building a strong data foundation, providing rich context, and actively refining your AI — you can unlock its transformative power. Don't just wait for AI to change your business, be proactive and shape how it works for you.