Natural-language BI and the analyst's new role: why the "democratisation" story is only half true

Something has shifted in how organisations talk about business intelligencebusiness intelligenceTechnologies and processes that turn raw data into actionable insights via reporting, dashboards and analysis, so teams can decide based on facts rather than intuition.View full definition →. Since the mid-2020s, every major BIBITechnologies and processes that turn raw data into actionable insights via reporting, dashboards and analysis, so teams can decide based on facts rather than intuition.View full definition → vendor, from Microsoft with Copilot in Power BI to Salesforce with Einstein Analytics, to Tableau's Ask Data feature and ThoughtSpot's core product proposition, has centred its pitch on one idea: anyone should be able to query data by typing a question in plain English. No SQLSQLSales Qualified Lead: a prospect the sales team has validated as ready for direct outreach and a proposal, having passed clear qualification criteria.View full definition →, no waiting for an analyst, no ticket queue. The promise is seductive, and the vendor investment behind it is real. But the conclusions most organisations are drawing from this trend deserve more scrutiny than they are getting.

The consensus view, stated fairly

The dominant position in CDO circles goes roughly like this. Natural-language interfaces (NLI) lower the barrier to data access so dramatically that traditional analyst roles become redundant at the margins. Business users self-serve. Analysts, freed from repetitive report requests, shift toward higher-value work: modelling, data productdata productA data asset managed like a product, with an owner, defined users, guaranteed quality, and measurable business value.View full definition → development, strategic advisory. The analyst headcount question becomes a talent quality question rather than a volume question. According to Gartner projections published in prior years, a significant share of analytics queries were expected to be generated via NLI by the mid-2020s, a figure vendors cite frequently in their sales materials. The logic chain is clean: less friction, more access, better decisions, leaner analyst teams.

There is genuine substance here. ThoughtSpot's design philosophy has proven that non-technical users can extract meaningful insight from large datasets without SQL knowledge, given adequate data governancedata governanceData governance is the set of policies, roles, and processes that ensure data is accurate, secure, well-defined, and used responsibly across an organization.View full definition → and well-modelled semantic layers. Microsoft's integration of Copilot into the Power BI workflow does reduce time-to-chart for straightforward questions. These are not trivial gains.

Where the consensus breaks down

The problem is that the consensus conflates access with understanding, and treats the analyst's value as residing in query execution rather than in judgment.

Start with the access-versus-understanding problem. When a regional sales director asks a NLI tool "why did revenue drop in Q2?", the tool can surface a chart and perhaps a contributory factor. What it cannot do reliably is distinguish between a seasonal pattern, a genuine demand shift, a one-time customer churncustomer churnChurn rate is the percentage of customers or revenue lost over a period. It measures how fast a business loses its existing customer base.View full definition → event, and a data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.View full definition → artifact caused by a delayed batch load. A business user without statistical grounding will read the output and conclude something. That conclusion may be wrong. The analyst's role was never primarily to write the SQL query; it was to know when the answer the query returned was misleading.

This is not a hypothetical. The academic literature on data literacy is consistent on this point. Research from the MIT Sloan Management Review has documented repeatedly that organisations with high data tool adoption but low data literacy investment produce more confident wrong decisions, not fewer. The tool lowers the friction of getting an answer; it does nothing to lower the probability that the answer is misinterpreted.

Second, the semantic layer problem is routinely underestimated. NLI tools work well when the underlying data model is clean, consistently defined, and aligned with the vocabulary business users actually use. The average enterprise data environment is none of these things. Metrics like "active customer", "recognised revenue", and "conversion" mean different things in different systems, sometimes within the same company. Building the semantic layer that makes NLI tools actually useful is a substantial analytical and organisational effort. It requires experienced analysts. Organisations that have cut analytical capacity in anticipation of NLI self-sufficiency often find themselves unable to build the very foundation those tools require.

Third, there is a vendor incentive problem worth naming directly. Salesforce, Microsoft, ThoughtSpot and their peers have a direct commercial interest in the "eliminate the bottleneck" narrative. When ThoughtSpot (a vendor selling NLI BI software) publishes research showing that self-service analytics drives business outcomes, that data should be weighted accordingly and cross-referenced against independent sources. The same applies to Microsoft Copilot adoption figures published in Microsoft's own case studies. This does not make the data false, but it does make it partial.

Finally, the "analysts shift to higher-value work" assumption is doing a lot of lifting in the consensus narrative. In practice, this transition is neither automatic nor guaranteed. Moving an analyst from report production to strategic modelling requires retraining, changed team structures, and managers who know how to task and evaluate that kind of work. Many organisations lack all three. The analyst who spent five years building Power BI dashboards is not automatically equipped to run scenario analysis for the CFO. Assuming the upgrade happens on its own is wishful.

What a sharp CDO should actually do

The CDO who has absorbed the consensus uncritically tends to make one of two errors: either cutting analyst headcount prematurely because the tools "handle it now", or deploying NLI tools without investing in the semantic layer and governance that would make them work, then being surprised when adoption plateaus and data quality complaints increase.

A more defensible position starts with a clear-eyed audit of where analyst time actually goes. In most organisations, the majority of analyst effort falls into three buckets: fielding ad hoc data requests, maintaining and updating existing reports, and fixing data quality issues. NLI tools can address the first bucket meaningfully, and arguably some of the second. They do nothing for the third, and may actually expand it by surfacing previously invisible inconsistencies.

The investment logic that follows: redirect analyst capacity freed by NLI tools toward semantic layer construction and data quality remediation, not toward headcount reduction. Build a small team of what some organisations are starting to call "analytical translators", people who understand both business context and data model design well enough to configure NLI tools for non-technical users and validate the outputs those users generate.

Maintain the expectation that business users who consume NLI outputs receive some baseline data literacy training. Not SQL, not statistics, but enough critical reading to ask whether a chart's axis is truncated, whether the time period shown matches the business question, and when to call the analyst rather than act on the dashboard.

The NLI shift is real and worth acting on. But the value it creates accrues to organisations that invest in the infrastructure and judgment layer underneath it, not to those that treat it as a replacement for analytical capability. CDOs who understand that distinction will build something durable; those who don't will spend 2027 explaining why their self-service analytics programme quietly failed.