Why BI Will Be Like AI

Why AI Will Be Like BI

❤️ Before Artificial Intelligence was all the rage
💛 Business Intelligence was THE big challenge for most companies
💚 To solve for AI, consider three key lessons from BI

1. It starts with Aggregation

Realising a “Single Customer View” is a common pain point. Whether you have just launched your MVP or have several mature revenue streams, you likely have fragmentation of data across analytics, application databases, and CRMs you use for talking to customers.

Failing to map and cleanse this data across sources results in BI that is messy and inconclusive. Now, it will also cause “garbage in, garbage out” with AI as you try the capabilities your current tools promise out of the box!

Once the “wow” factor fades and you’re worried about hallucinations and poor content at scale, consider the source and whether you’ve really given your AI-enabled tools the most relevant data.

2. AI, BI, & CI: Oh My!

Continuous Integration (CI) is a concept usually applied to getting code into production smoothly with automated tests and deployment.

This exists in the modern data stack with tools like Fivetran or Airflow automating aggregation regularly – ideally with further transformations (e.g. DBT) in an “Extract, Load, then Transform” (ELT) pattern.

These tools did not come overnight, but have evolved to solve this common pain point. This allows organisations to automate aggregation with MUCH less reactive engineering work when something changes (but sadly, still non-zero!)

Any scalable use of AI should consider the same approach at a foundational level, ideally using the great ELT infrastructure you already have!

3. Make it Actionable

BI is only as good as the decisions that it enables. The same will be true for AI, with “content” in place of “decisions”. Remember, Generative AI at the end of the day is just content!

The low hanging fruit could be in the CRMs and admin tools you’re already using. Boost your growth by generating individually personalised emails for each customer rather than segmented campaigns. Optimise your support costs by having high-context co-pilots drafting your chat responses. CRMs will claim to solve this for you, but will likely hit a wall without more aggregated data.

If you’re still struggling with BI before you consider AI, the good news is you can solve for both by fixing your horizontal aggregation. Then, you can focus on the evolving ecosystem for getting this ‘cleansed’ data into the right LLMs to power your in-house tools.

My prediction: the emergent AI ELT stack will look a lot like the current BI ELT stack, with concepts like vector warehouses, prompt engineering, RAG fulfilling the “transform” stage.

Who will build the DBT for Large Language Models that empower businesses to curate data across their entire business?

I’d love to hear your experience on this journey as I develop my own value proposition in this space! Have you found synergy in AI vs BI? Get in touch if you're stuck!