The business intelligence market promised data democratization. Twenty years and $38 billion later, most business users still can't ask their database a question. Agentic Text-to-SQL doesn't complement BI — it replaces the layer where most of the cost and friction live.

The Market Setup: Two Curves, One Collision

Two markets are on a collision course, and the impact zone is the ad-hoc analytics layer — the most expensive, most frustrating, and most underserved segment of enterprise data infrastructure.

The BI market is large and slow. Global business intelligence software was valued at approximately $38 billion in 2026, growing at a CAGR of 8–9% toward an estimated $72–78 billion by 2034. It's a mature market. The vendors are entrenched — Microsoft Power BI, Tableau, Qlik, SAP BusinessObjects — and the architecture has remained fundamentally unchanged for a decade: ingest data into a warehouse, build dashboards, publish reports.

The agentic AI market is small and explosive. Valued at roughly $7 billion in 2025, it's projected to reach $47–57 billion by 2030–2031, growing at a CAGR of 42–46%. Multi-agent systems — the architecture behind production-grade Text-to-SQL — already command over 53% of the agentic AI market and are growing at 43.5% CAGR.

These two markets aren't adjacent. They're converging. And the point of convergence is the ad-hoc query — the single most common, most costly, and most poorly served interaction between a business user and their structured data.

The BI Promise vs. the BI Reality

The business intelligence industry was built on a compelling proposition: give every decision-maker access to the data they need, when they need it, without relying on IT.

Two decades in, that promise remains largely undelivered.

The self-service BI segment — the part of the market explicitly designed to let non-technical users analyze data independently — reached approximately $7.1 billion in 2025, growing at roughly 11.5% CAGR. That sounds healthy until you examine what "self-service" actually means in practice. Over 85% of enterprises now manage more than 100 terabytes of business data. Around 60% of enterprise employees interact with analytics tools at least once per week. Yet the large-enterprise segment still accounts for 68% of self-service BI revenue, while SMEs — the segment with the most to gain from data democratization — remain underserved.

Why? Because self-service BI still requires trained users. Building a Power BI dashboard isn't writing SQL by hand, but it's not asking a question in plain language either. It demands understanding of data models, DAX formulas, filter contexts, and visualization logic. The result: a narrower definition of "self-service" than the marketing suggests — one that serves power users and trained analysts, not the VP of Sales who needs a number before a 9 AM meeting.

The unserved demand sits in the gap between what BI dashboards answer (predefined questions, refreshed on schedule) and what decision-makers actually ask (ad-hoc questions, in natural language, right now).

The Ad-Hoc Query: Where the Money Burns

The ad-hoc analytics request is the silent cost center of every data-mature organization.

The pattern is universal: a business user has a question. The dashboard doesn't answer it. They file a ticket — or walk over to a data analyst's desk. The analyst writes a custom SQL query, validates it, formats the output, and returns it. Elapsed time: hours to days. Cost: a significant share of highly paid engineering capacity diverted from strategic work.

Data professionals spend a disproportionate share of their time on data preparation and ad-hoc requests rather than on analysis, modeling, and strategy. While the often-cited "80% of time on data prep" figure has been debated, industry surveys consistently show that data cleaning, organizing, and responding to bespoke requests consume 40–60% of analyst capacity. A 2025 Gartner Analytics Survey found that 47% of enterprises cited incomplete or poor-quality data as the single biggest blocker to timely reporting.

The economics are punishing. A mid-market company with a 5-person data team paying €80–120k per analyst all-in dedicates roughly 2–3 FTEs' worth of capacity to ad-hoc requests. That's €160–360k per year spent on answering questions that an agentic system could handle in seconds.

For enterprises, the numbers scale accordingly. A wealth management firm with 200 advisors, each generating 3–5 ad-hoc data requests per week, creates a demand pipeline of 600–1,000 bespoke queries weekly — a volume that overwhelms any BI team and guarantees multi-day response times.

This isn't a technology failure. It's an architecture failure. BI was designed to answer known questions repeatedly. The ad-hoc question — by definition unique, unplanned, and time-sensitive — was never the use case BI was built for. Dashboards answer yesterday's questions. Business users need to answer today's.

The Replacement Layer: What Agentic Text-to-SQL Actually Displaces

The thesis isn't that agentic Text-to-SQL replaces all of BI. BI platforms will continue to serve their core function: recurring dashboards, scheduled reports, executive KPI views, and embedded analytics in operational applications.

What Text-to-SQL replaces is the ad-hoc query layer — the part of the BI ecosystem where cost is highest, latency is worst, and user satisfaction is lowest.

Here's the functional displacement map:

The addressable displacement is concentrated in the two highest-cost categories: ad-hoc requests routed to analysts and bespoke SQL written by data engineers. Together, these represent the majority of operational BI spend in most organizations — and they're precisely the categories where agentic Text-to-SQL delivers 95%+ time-to-answer reduction.

Sizing the Displacement: A Conservative Estimate

How large is the ad-hoc query layer as a market?

There's no standalone analyst report that sizes "ad-hoc enterprise analytics" as a category — it's buried inside the broader BI and data engineering markets. But we can triangulate:

The self-service BI market — the closest proxy — is valued at $7.1 billion in 2025, growing at 11.5% CAGR. Self-service BI exists specifically because the ad-hoc layer was too expensive to service with human analysts. But it only partially solves the problem (trained users, predefined data models, visual interface required).

The agentic AI in enterprise workflows market — a new category tracked by Technavio — is projected to grow by $5.8 billion from 2026–2030 at 47.4% CAGR. A significant portion of this growth maps to data access and query automation.

The data engineering labor cost that Text-to-SQL displaces is harder to quantify at macro level, but consider: the global BI market employs an estimated 500,000+ data analysts and engineers whose work includes ad-hoc query fulfillment. At an average fully loaded cost of $100–150k, the annual labor spend on ad-hoc analytics alone likely runs in the tens of billions — dwarfing the software spend.

A conservative framing: if agentic Text-to-SQL captures even 15–20% of the self-service BI market plus 10% of the ad-hoc data engineering labor cost over the next 5 years, the addressable market exceeds $5 billion annually. That's before accounting for greenfield demand — organizations that never had BI at all because the cost and complexity barriers were too high.

The SME Greenfield: Where BI Never Arrived

This is the dimension that market-sizing exercises consistently undercount.

The BI market is dominated by large enterprises: they account for 61–68% of revenue depending on the segment. SMEs — companies with 50–500 employees — are forecast to grow fastest (15%+ CAGR in self-service BI, 43.5% CAGR in agentic AI), but they start from a small base.

The reason is structural. Traditional BI requires infrastructure: a data warehouse, ETL pipelines, a BI platform, and trained users to operate it. The total cost of ownership for a mid-market BI deployment — including licenses, implementation, and the analysts to run it — easily exceeds €100–200k in the first year alone. For a 45-person e-commerce company or a 300-person manufacturer, that's a prohibitive investment relative to the value delivered.

Agentic Text-to-SQL collapses this cost structure. The system connects directly to the existing database — the ERP, the CRM, the operational data store. There's no data warehouse to build. No ETL pipeline to maintain. No dashboard to design. No analyst to hire.

The business user types a question. The system generates governed SQL. The answer comes back in seconds.

For the German Mittelstand — the backbone of the European economy, comprising roughly 3.5 million SMEs — this isn't an incremental improvement. It's the first time many of these companies can access their own operational data without an Excel export and a weekend of manual work.

The greenfield isn't a niche. It's the majority of the market by company count, currently generating near-zero BI software revenue.

The Architecture That Makes Replacement Possible

Replacing the ad-hoc query layer isn't a matter of plugging an LLM into a database and hoping for the best. The reason BI hasn't been disrupted earlier is that accuracy requirements in enterprise data access are non-negotiable.

The architecture that enables displacement requires five components working in concert:

Multi-Agent Task Decomposition breaks complex natural-language questions into logical steps before any SQL is generated. This prevents the failure mode of monolithic LLM queries — syntactically valid SQL that answers the wrong question.

Table-Augmented Generation (TAG) queries the actual database schema before writing SQL, ensuring the system maps business concepts to real table and column names — not hallucinated ones.

The Semantic Layer encodes business logic — fiscal calendars, KPI definitions, currency conversions, entity hierarchies — so the system calculates the way the business thinks, not the way the database is structured.

Governed SQL Execution runs every query in a read-only, sandboxed, auditable environment. Full logging. No data exfiltration. On-premise deployment where required. This is the capability that unlocks regulated industries — banking, pharma, insurance — where cloud-based BI alternatives are often ruled out by compliance requirements.

Adaptive Abstention ensures the system fails safely. When confidence is low or the query is ambiguous, it asks for clarification rather than returning fabricated numbers. In enterprise contexts, a 90% accuracy rate isn't "pretty good" — it's a system that's wrong one out of every ten times a decision-maker relies on it.

This architecture doesn't look like BI. It doesn't look like a chatbot. It looks like a governed query terminal — and that's exactly the form factor the ad-hoc layer needs.

The Regulatory Tailwind: Data Sovereignty as Market Accelerant

One structural advantage that agentic Text-to-SQL holds over cloud-centric BI platforms is the data sovereignty story — and in Europe, this is becoming a primary purchasing criterion.

Hybrid and on-premises agentic AI deployments are growing at 44.6% CAGR, outpacing cloud-only deployments. The EU AI Act, GDPR, the Cyber Resilience Act (requiring vulnerability reporting by September 2026), and sector-specific regulations (BaFin for banking, EMA/BfArM for pharma) are creating a structural premium for solutions that keep data within the enterprise perimeter.

For regulated mid-market companies — regional banks, pharmaceutical companies, insurance firms — this isn't a preference. It's a requirement. Cloud-based BI platforms that route queries through external servers are increasingly disqualified by compliance teams before evaluation even begins.

An agentic Text-to-SQL system that runs on-premise, generates SQL against the local database, logs every query for audit, and never transmits data externally isn't just a feature advantage. In the European regulated mid-market, it's a market-access requirement that most BI incumbents can't meet.

The Five-Year Outlook: Coexistence, Then Absorption

The displacement won't happen overnight. BI platforms have deep integration into enterprise workflows, established procurement relationships, and a decade of sunk cost in dashboard development. But the trajectory is clear:

2025–2026: Coexistence. Agentic Text-to-SQL deploys alongside existing BI stacks, initially handling overflow ad-hoc requests that analysts can't service fast enough. Early adopters in financial services, manufacturing, and pharma validate accuracy and auditability.

2027–2028: Layer Replacement. As confidence in governed Text-to-SQL accuracy grows (>99% target), organizations begin routing all ad-hoc data requests through the agentic layer. Data analyst roles shift from ad-hoc query fulfillment to data modeling, governance, and strategic analysis. BI budgets reallocate from per-seat analytics licenses to agentic infrastructure.

2029–2030: Greenfield Expansion. The SME segment — 3.5 million Mittelstand companies in Germany alone — begins adopting agentic Text-to-SQL as their primary data access layer, bypassing BI entirely. The total addressable market for data access tools expands beyond what current BI market sizing captures.

Gartner's prediction that 40% of enterprise applications will include task-specific AI agents by 2026 aligns with this trajectory. The data access agent — the one that sits between a business user's question and a structured database — is the most natural and highest-ROI deployment of agentic AI in the enterprise.

The Investment Thesis in One Paragraph

The $38 billion BI market grows at 8%. The agentic AI market grows at 44%. The intersection — the ad-hoc query layer where business users need instant, auditable answers from structured databases — is where value migrates. The winners won't be the largest BI platforms or the most powerful LLMs. They'll be the systems that combine governed SQL generation, semantic business logic, on-premise deployment, and adaptive abstention into an architecture that's accurate enough to trust, fast enough to use in a live meeting, and sovereign enough to deploy in a regulated European mid-market. That's the replacement layer. And it's arriving now.

Parlance is an agentic Text-to-SQL platform built for data-sovereign enterprises. On-premise. Auditable. No data leaves your infrastructure.

Niklas Hegemann & Sebastian Schorsch

Co-Founder of Parlance AI — Agentic Text-to-SQL for data-sovereign enterprises. Founded in Frankfurt am Main, 2026.