Deloitte: Scale autonomous intelligence with decision-grade data

Deloitte says enterprises must scale what it calls “autonomous intelligence” by combining agentic AI with decision-grade data, verifiable agent identity and enterprise governance so AI agents can execute transactions across internal systems. The guidance was published ahead of the AI & Big Data Expo North America where Deloitte will present further details.

Prakul Sharma, principal and AI & Insights Practice Leader at Deloitte Consulting LLP, described autonomous intelligence as the next stage after assisted and artificial intelligence: “Autonomous intelligence pursues an outcome by reasoning over a goal, invoking tools and data, and adapting as conditions change, with humans setting guardrails.” He framed agentic AI as the bridge from conversational GenAI to systems that act within defined boundaries.

Deloitte recommends starting with a decision audit to identify value chains where slow or inconsistent decisions limit results. The audit maps who holds data and authority, where handoffs fail and where judgement is applied, revealing governance and data gaps that can block scaling beyond pilots.

The firm uses procurement as a concrete example. An agentic application can compare supply inventory with live vendor pricing in an enterprise resource planning system and automatically issue purchase orders within pre-set financial limits, pausing for human approval when deviations occur. For such execution the agent must have a verifiable identity in the ERP, access pricing data that is current enough to be contractually binding, and operate inside approval thresholds cleared by legal and compliance teams.

Deloitte distinguishes “decision-grade” data from reporting-grade data. Decision-grade data requires timestamped freshness, traceable provenance and access controls that confirm an agent is authorized to read and act on records. Reporting-grade data, typically aggregated in nightly batches and stripped of lineage, is suitable for human review but unsafe for autonomous execution. The firm advises integrating agents with event stores and databases designed to manage both structured and unstructured information.

The report identifies three common failure modes when moving from pilot to production: upstream data friction, production gaps and governance debt. Pilots can work with curated datasets and manual workarounds, but scaling reveals missing identity verification, absent continuous evaluations, incomplete audit trails and unforecasted API costs.

Deloitte also highlights financial risks. Agentic workflows often require multiple model calls and retrieval-augmented generation to reduce hallucinations, which can raise variable compute and API expenses. The firm recommends modelling those variable costs and building financial monitoring into the platform from the start.

Operationalizing autonomous agents requires integration with enterprise identity providers and cloud-native security controls across hybrid cloud environments. Deloitte warns that teams that bypass standard security and compliance controls during pilots can create governance debt that prevents production deployment. The firm encourages building identity, continuous evaluation and financial controls into initial platform design so the same foundations can be reused for subsequent deployments.