BI Works. So Why Do Only 16% Use It Regularly?

Last week we wrote about what it takes to make AI analytics reliable: Anthropic’s data team got from 21% to 95% accuracy by building a system around the model — curated semantics, per-domain reference docs, evals, and a standing maintenance regime.

This week: why even good BI reaches surprisingly few people.

BARC’s BI & Analytics Survey 26, based on more than 1,000 users, reports 97% faster reporting or planning, 96% improved data quality, and 94% better decisions.

In large enterprises, only 16% of employees use BI regularly.

BI is not failing. It is stopping at a boundary. It creates value for the people who can reach it, but that value does not spread very far. The survey does not prove why — that is the hedge worth keeping. Our thesis is that part of the adoption gap is a translation gap: BI expects a business question to arrive already translated into the system’s language.

The wrong question is “Does BI work?”

For the people who know which dashboard to open, which metric to trust, and how to navigate the model, BI often works well. The harder question is what happens to the person whose question was not anticipated by a dashboard.

They usually have three paths:

• adapt the question to the report that already exists;
• learn enough of the data model or SQL to answer it themselves;
• ask an analyst to translate the business question into the available definitions and data.

Those are not simply interface problems. They are access patterns — and each assumes knowledge that many business users do not have.

“Adoption” hides several different failures

Low regular use can reflect training, incentives, discoverability, data quality, or workflow. It can also reflect a translation problem: the business question arrives in ordinary language, while governed meaning lives in metrics, models, reports, and local conventions.

That last problem is easy to miss because analysts absorb it. They know that “active customer” has three definitions, that Finance uses net revenue, or that the Northeast report excludes a particular channel. They turn a loose question into an executable one.

When someone outside that circle asks a new question, the difficulty is not merely reaching the warehouse. It is reaching the right meaning.

Dashboards are answers to questions someone has already asked. The adoption problem begins with the question nobody anticipated.

Conversational access is not enough

Putting a chat box in front of the warehouse lowers the effort required to ask. It does not, by itself, settle what “active customer” means, which comparison period applies, or which governed definition belongs in this decision. It can make an unplanned question easy to ask while leaving the important choices buried in generated SQL.

Conversational BI needs a review point between the question and the query: ground the analysis in the business context already available, show the intended meaning, ask when a material choice remains unresolved, and retain a trace of what ran.

That is a better evaluation target than “Can the model produce SQL?”

The question I would ask instead

If regular BI use is low, do not start by assuming the answer is another dashboard or another training program. Follow one unplanned business question from the person asking it to the definition, interpretation, query, and answer.

Where does the person get stuck? Which choices are invisible? When does the work become a ticket? That path tells you whether you have an interface problem, a workflow problem, or a meaning problem.

The survey does not tell us why each person sits outside the 16%. But it points to a market the industry has not served well: people with real business questions who cannot translate them into the data stack’s language without help.

That may be the largest unserved market in analytics — not companies without BI, but people inside companies that already bought it.

The next adoption curve will not come from another place to look. It will come from giving an unplanned question a governed path to an answer.

---

Previously: The Engine Is Not the System · See the interpretation before a query runs: how it works

Source: BARC BI & Analytics Survey 26, based on responses from more than 1,000 BI users.