Business intelligence exits to support decisions and AI exists to replicate human intelligence through machine learning. AI BI magic happens when coupling the right people, prioritization, and processes around data, information management. AI + Analytics works, like any initiative when you have everyone on the team rowing in the same direction. This article was written to highlight some core concepts that help with alignment for AI BI initiatives.
What BI Should Answer
Where are we? – Leading indicators (metrics and KPIs) summarize the current or previous state. These metrics and KPIs are calculated and summarized and mixed together based your the goals / objectives. These assets are designed for frequency and wide distribution. Dashboards for infrequent use by small numbers of users are reserved for executive and board-level users.
How did we get here? – Lagging indicators and trends over time and segments (dimensionality) help explain what is changing over time. The reality is trends and basic descriptive statistics explain “what” but not “why.”
Predict What “should happen next” – Basic statistical methods and machine learning algorithms have been paired with BI over the last 10 years.
What AI Should Answer?
What should we do to achieve our desired outcome? This has long been the promise of AI, long before ChatGPT: provide an objective, provide data and compute, and AI returns an answer. This is the domain where the competition to AI is gut, experience, and interpretation. Most data-driven organizations are still in the BI world, where data influences a human decision. Machine learning was reserved for high impact use cases because the cost, time, and expertise required made it so.
| Last 15 Years = Data Driven Aspiration | Now = Agentic Data Driven Aspiration |
| human is the recipient of a data / analytics | AI agent is recipient of data / analytics |
| descriptive statistics or predictive score provides predictions | descriptive statistics / predictive score provides predictions |
| reasoning, analysis, consensus occurs | AI reasoning interprets, iterates, recommendation are formulated |
| a decision is made and action occurs | AI or human decision and action occurs |
AI models have evolved dramatically with the explosive adoption of LLMs. Industry analysts still report that most BI AI initiatives have a high failure rate. AI assisted analytics can be far more nuanced, so here is what you should look out for:
Why AI BI Initiatives Fail
- Implementation of tech in search of a problem and problems without material impact
- Lack of data governance, fidelity and as a result low trust
- Mismatched expectations between “art of the possible” and reality organizational readiness
- Knowledge management and metrics alignment
Nuanced Details that Matter for Planning AI BI
AI is going to touch every facet of how information is originated, delivered, and distributed inside an enterprise. Vendors like Databricks and Microsoft offer end-to-end platforms that facilitate the full BI and AI stack. Even the top AI companies in the world can’t solve your business / process challenges without consultants (they call them forward deployed engineers). No matter how well the tech tools are marketed, the devil’s in the details when you need to solving business problems with AI BI.
Deterministic vs Probabilistic
Deterministic: same inputs plus same rules equals the same output every time.
Probabilistic: predicts the most likely output based on learned patterns and uncertainty.
When LLMs first arrived, they were unreliable at math because they predicted answers from patterns rather than calculating them. Today, advances in reasoning, tool use, and code execution allow LLMs to offload calculations to deterministic systems, which has dramatically improved performance on math and statistics tasks. The model is still probabilistic, but it increasingly orchestrates deterministic tools beneath it.
This is where we see AI BI and the need for business and data semantics being critical.
Monitoring, Exploring, Storytelling
Somewhere along the way, these three disciplines were blended into BI dashboards. With correct data and well-defined semantics, AI can be a very capable analyst for applying descriptive and predictive statistics across all three modes.
Monitoring is the dashboards, reports, and scheduled PowerPoints already circulating. AI is well-suited to producing these assets, and there is real innovation happening in this area across the AI BI landscape.
Exploring is where you ask questions business questions seeking answers using data. Excel is still the most widely used data exploration tool. Tableau is arguably second, unless you include Python and R. Exploration is also widely used for data wrangling tasks.
Storytelling is where disciplines overlap. Every PowerPoint is a curated story. The art and science of narrating a story with data is a special skill and AI quite effective at telling these stories.
Understand these 3 data presentation disciplines will help with tool selection because you need all 3.
Data Needs Meaning. Meaning Needs Context.
A semantic layer defines your metrics. Revenue = SUM(amount), every time, for everyone.
An ontology defines your business. Customer owns Opportunity. Opportunity belongs to a Segment. Segment drives Revenue.
BI tools need the first. AI agents need both.
As AI moves from predicting answers to orchestrating deterministic tools beneath them, the semantic layer becomes the calculation engine and the ontology becomes the reasoning and context map. Neither replaces the other. Together, they’re what separates governed AI analytics from a confident guess.
AI BI Governance
AI agents can generate insights, dashboards, and analytics artifacts faster than any business teams can digest them and technology teams can manage them. At 20-50 of anything, you reach an inflection point where managing them without structure becomes a new challenge and pain point.
- Distribution and curation remains mostly unsolved. Search, discovery, and exploration of analytics assets has been a weak link in enterprise BI for a long time. Searching across chats, folders, repositories for the brilliant AI insight or artifact is causing enterprises a lot of pain. We are always hunting for better delivery experiences; for most clients, the answer is embedding analytics where users already work.
- Data governance is the unglamorous but necessary structure that ensures AI initiatives can reach production with data that is correct, consistent, and appropriately scoped. A significant number of AI projects are stalling there. Access, trust, and security are are all bottlenecks.
- Semantic disconnect is the other failure mode rarely discussed and something we understand intimately. It’s our specialty and focus at DataTools Pro. LLMs can be gradually misdirected over time through inconsistent business terminology. Those misdirected signals produce incorrect results that poison your results. Incorrect results erode trust. Data can be perfect and the semantic layer can be well-executed; the breakdown happens when business semantics shift and the model does not keep pace.
Learn more AI BI Challenges
What does succesful AI BI look like?
Organizations that succeed with both BI and AI are are managing with some form of governance. Knowing what your data means, who owns it, and whether it can be trusted, and what AI vendors are doing with it is critical. Feel free to check out some of our recent learnings and solutions for AI BI or Contact Us so we can learn about your AI BI initiative.
