A chart can show that sales fell without explaining whether the cause was weaker demand, delayed shipments or a change in accounting. Business intelligence turns data into metrics, analysis and information people use to make decisions. Its value comes from helping users ask the right questions and trust the answers. AI changes how those questions are asked, but it does not remove the need for clear definitions and sound analysis.
The investment question
The investment question is whether BI suppliers retain control of the analytical workflow as users move from dashboards towards conversational assistants and embedded insights. If generating a chart becomes easy and inexpensive, value must come from something more durable than visual authoring alone.
Our view is that trusted metrics, distribution and integration into decisions become increasingly important. A platform with widely adopted business definitions can remain valuable even when its interface changes. However, cloud bundles, application-native analytics and AI assistants can weaken pricing power for products that mainly present information already governed and processed elsewhere.
How business intelligence works
BI tools connect to data sources, model relationships, define calculations and present results through reports, dashboards or embedded views. Some import and cache data; others query the source directly. These approaches trade off responsiveness, freshness, infrastructure demand and management complexity. Neither guarantees accurate analysis without appropriate modelling.
The semantic layer defines business meaning: measures, dimensions, relationships and calculation rules. Revenue can mean invoiced, recognised or collected revenue; active customers can be measured over different periods. The dbt Semantic Layer documentation illustrates how metric definitions can be shared across consuming tools. This separates some business logic from a single dashboard interface.
Visualisation then helps users interpret patterns and compare alternatives. Filters, aggregation and chart choices affect what users perceive. A correlation is not a causal explanation, and a dashboard cannot establish the effect of a business intervention without suitable evidence. Good BI supports analytical judgement rather than substituting attractive output for it.
Market structure and competitive advantage
| Competitive position | Main advantage | Main vulnerability |
|---|---|---|
| Enterprise suite BI | Procurement, identity and broad distribution | Complexity and potential limits outside its ecosystem |
| Specialist analytics platform | Exploration, usability and distinctive workflows | Bundled alternatives and price pressure |
| Embedded analytics | Information inside an existing application | Dependence on the application’s distribution and economics |
| Semantic and metrics infrastructure | Consistent definitions across interfaces | Competing native semantic layers and standards |
Microsoft Power BI, Salesforce’s Tableau and Google Looker sit within large technology ecosystems. Looker on Google Cloud illustrates the integration of managed analytics with cloud infrastructure. Independent products compete through particular analytical experiences, embedding capabilities or business workflows. Market share alone does not reveal how much of the customer’s value each layer captures.
Switching costs include report libraries, calculations, permissions, training and habitual use in management processes. Yet many organisations already operate multiple tools. A new department or application can choose a different interface while retaining the same warehouse, creating competition for incremental users without an immediate enterprise-wide migration.
Economics: paid seats are only one part of the model
BI pricing can combine creator and viewer seats, shared capacity, embedding, AI consumption and data-platform charges. A low licence price may be offset by implementation, refresh processing or extensive report maintenance. A premium product can be economical if it materially improves adoption or reduces the effort needed to deliver useful analysis.
For illustration, suppose a company pays £120,000 annually for 1,000 licensed users, but only 300 use the product regularly. Licence cost is £120 per licensed user and £400 per active user. If active use doubles with the same bill, the second figure falls to £200. These are illustrative figures, not vendor prices, and still say nothing about whether decisions improved.
AI complicates seat economics. One analyst may create more output, while many occasional users consume insights through an assistant rather than opening a BI application. Vendors may seek revenue through capacity, agents or usage. Customers will compare the total bill against analytical outcomes and the cost of alternative interfaces.
For suppliers, distribution efficiency matters. An existing enterprise relationship can lower selling friction, while independent vendors may need stronger differentiation to justify a separate procurement decision. Gross margin also depends on hosted compute and AI costs, particularly when intensive usage is included in a fixed subscription.
AI and hyperscalers: conversational access needs a trusted foundation
Natural-language interfaces can broaden access to analysis and help users explore follow-up questions. Microsoft’s Copilot for Power BI documentation describes assistance with reports, summaries and questions about data. Practical results depend on the underlying model, configuration and supported capabilities; generated output still needs to be evaluated against the intended business meaning.
The difficult part is often ambiguity. A question such as “Why did margin fall?” requires a definition of margin, a comparison period, relevant drivers and evidence about cause. An assistant should distinguish a computed decomposition from a causal conclusion. Showing the metric definition, filters and sources helps users assess the answer.
Hyperscalers can combine BI with warehouses, identity, collaboration software and model services. This gives them several routes to the user and lets them bundle capabilities across a broader contract. Independent tools can counter with stronger usability, cross-platform semantic consistency or deep support for specialised decisions that generic assistants handle poorly.
Current market debates — September 2026
The first debate is whether agentic analytics replaces dashboards or complements them. Salesforce positions Tableau Next around conversational and agent-supported analytics connected to its semantic layer and application environment. The opportunity is to move from observing a metric to taking informed action. The countercase is that repeatable dashboards remain more efficient for routine monitoring, and consequential actions require dependable context and appropriate controls.
The second debate is ownership of the semantic layer. Salesforce’s semantic-model guidance connects business definitions to conversational queries. BI, transformation and data-platform vendors all have reasons to own those definitions. Customers benefit when definitions are reusable, but must test whether different tools apply the same logic consistently.
The third debate is monetisation as authoring becomes easier. AI may reduce the hours required to build reports and increase the number of people exploring data. That can expand demand for governed analytics while reducing demand for some traditional creator workflows. A growing number of generated charts is therefore a poor substitute for evidence of paid, repeated business use.
Structural debates: interface, intelligence and accountability
An analytical interface can become easier to replace when metrics and permissions live elsewhere. Conversely, a BI platform deeply embedded in planning, sales reviews or operational decisions may retain substantial influence. The investment question is where the customer would feel the most disruption if one component disappeared.
There is also a tension between democratising analysis and multiplying inconsistent answers. Self-service works best when users can explore within clear definitions and understand when they have moved beyond them. Central modelling should provide dependable foundations without preventing legitimate local questions or hiding the assumptions behind a result.
Finally, analytics that triggers actions creates a different responsibility from analytics that merely displays a chart. An agent adjusting a sales campaign or inventory order must operate within explicit permissions and measurable feedback loops. Suppliers that connect insight to action can capture more value, but they must support traceability and correction when the recommendation is wrong.
What to watch
Watch active use, retention by role, report maintenance effort, semantic-model adoption and the time between a question and a defensible answer. Compare full costs including refreshes, query execution and AI. Customer success is stronger when analytics changes a repeatable decision process than when it simply increases dashboard counts.
For AI, measure answer accuracy on realistic business questions, consistency across repeated queries, transparency of definitions and conversion into useful action. The enduring opportunity is to make trusted analysis accessible where people already work. Suppliers that own that trusted relationship have a stronger case than those competing only on how quickly they can generate a visual.
Explore this sector
Data platforms & analytics — sector overview
Related sectors: Semiconductors & chipmaking · AI infrastructure & data centres
