The BI maturity model that actually predicts ROI

Every BI vendor has a maturity model that ends, conveniently, at exactly the product they’re selling. We needed one that actually predicted whether a client engagement would generate ROI — across a five-stage path from chaos to compounding. Here it is, with the gate at each stage.

The model below is not theoretical. It came out of a retrospective across the last few years of engagements: stages a client cluster moved between, what unlocked the next stage, and where ROI started compounding vs. flat-lining. The clearest pattern: ROI is not a function of how many dashboards exist. It is a function of which stage the program is in.

Stage 1 — Reactive

Reporting is on-demand and human-driven. Excel everywhere, definitions drift, every leadership question takes a person two days to answer. The gate to stage 2 is owning a defined KPI hierarchy — not building anything, just agreeing on the metrics.

Stage 1 assessment criteria

• No agreed KPI list across leadership
• Reports produced manually on request, often by analysts in Excel
• Every senior meeting includes an argument about numbers
• No named owner for any individual metric
• Budget for BI is reactive — funded one ad-hoc request at a time

Stage 2 — Repeatable

Standard reports exist, refreshed on a cadence, with named owners. Numbers are mostly consistent. The gate to stage 3 is a governed semantic layer — Power BI tabular model, dbt-defined marts, something that prevents the same number being calculated three different ways.

Stage 2 assessment criteria

• A KPI list exists and is broadly agreed
• Standard reports are refreshed on a schedule
• Named owners exist for the headline metrics
• Reconciliation to finance happens, but inconsistently
• Semantic layer is partial or absent — measures live in individual reports

Fastest path from stage 2 to stage 3

This is the highest-leverage transition in the entire model. The play that works: pick the top revenue domain, build a governed semantic layer for it (tabular model + measure library + RLS + lineage), migrate the three highest-impact reports onto it, and reconcile every metric to finance. Eight to twelve weeks of focused work. Skip the temptation to build coverage across domains — depth in one beats breadth across five at this stage.

Stage 3 — Governed

Semantic layer exists, RLS is in place, dashboards reconcile to finance. ROI starts to show up here because leadership trusts the numbers. The gate to stage 4 is self-serve — analysts and operators can answer their own questions without filing a ticket.

Stage 3 assessment criteria

• Governed semantic model for the top one or two business domains
• Row-Level Security and sensitivity labels in production
• Reconciliation to finance is a routine cadence, not a project
• Measure library is shared across reports
• Lineage is documented and visible to analysts

Stage 4 — Self-serve

Business users build their own reports against a curated dataset. The data team builds the substrate, not every report. The gate to stage 5 is predictive — using the foundation to answer forward-looking questions, not just historical ones.

Stage 4 assessment criteria

• Business analysts can build their own reports against curated datasets
• The data team is producing models and measures, not individual reports
• Workspace governance includes a clear promotion path from sandbox to certified
• Self-serve usage is monitored — the team can name the most active builders
• A documented retirement path exists for unused reports

Stage 5 — Predictive / AI-native

Forecasting, scenario modeling, and conversational AI on top of the governed layer. The semantic model is also the substrate for the AI surface. This is where the demos finally match production.

Stage 5 assessment criteria

• Forecasting and scenario models in production use the governed semantic layer
• Conversational BI is exposed to a controlled audience with audit logging
• Anomaly detection and proactive alerting are wired into operational workflows
• The data team is shipping product, not service tickets
• Leadership can attribute decisions to data outputs with confidence

Budget allocation by stage

Budget at each stage should match the bottleneck of that stage, not the appearance of progress. We use rough rules: at stage 1, spend on KPI definition and metric ownership, not on tools. At stage 2, spend on the semantic layer and reconciliation work that unlocks stage 3. At stage 3, spend on self-serve enablement, governance, and analyst training. At stage 4, spend on the data product mindset and the AI substrate. At stage 5, spend on the application of forecasting and AI to real operational decisions.

Budget rules per stage

• Stage 1: spend on people (analysts, definitional work), not on platform
• Stage 2: spend on the semantic layer and reconciliation discipline
• Stage 3: spend on enablement and self-serve tooling
• Stage 4: spend on the substrate for AI and predictive workloads
• Stage 5: spend on integrations between AI outputs and operational workflows

What this means for budgets

The highest-ROI quarter for almost every BI program is the one that moves stage 2 to stage 3 — building the semantic layer and governance model. It’s also the least exciting one to budget for. Skipping it to chase stage 5 is the most common, and most expensive, mistake in mid-market BI.

The one-line takeaway

Pick your maturity stage honestly, fund the bottleneck that unlocks the next one, and resist the gravitational pull of skipping ahead to whatever stage the latest vendor demo represents. The program that compounds is the one that earns each stage before it moves on.