“Autonomous” is a word that makes procurement leaders lean in and lean back at the same time. The promise is attractive: agents that handle repetitive sourcing work and free people for strategy. The risk is just as real: decisions made at speed, without the judgment, controls or audit trail a CPO must be able to defend.
The Hackett Group’s The Autonomous Sourcing Index suggests that the most successful organizations are not choosing between the two. They are designing for governed autonomy.
What autonomy looks like today
Despite the headlines, procurement is moving carefully. When respondents describe the level of autonomy in their sourcing, 25% say they do not use AI for it and 23% describe it as assisted. Another 25% describe it as guided, and 20% use AI with human approval. Only 7% say sourcing is mostly autonomous, and none say it is fully autonomous.
Fully autonomous sourcing remains near 0%. That is not a failure of ambition. It reflects how the research defines the field: autonomous sourcing applies purpose-driven AI agents across the end-to-end sourcing lifecycle, while people retain appropriate authority and oversight over consequential decisions.
The phrase “governed by design” appears in the report for a reason. It lists five features: human approvals, explicit guardrails, evidence-linked rationale, continuous monitoring and a full audit trail.
Risk-based autonomy, in practice
One procurement executive quoted in the research describes a principle that many peers will recognize: lower-value, repetitive sourcing can move toward greater autonomy within clear guardrails, while strategic and high-value decisions require appropriate human oversight.
This is a more useful lens than a single autonomy dial. Instead of asking “how autonomous should procurement be?”, ask category by category and decision by decision:
- What is the value and risk if this decision is wrong?
- Can the agent show the evidence behind its recommendation?
- Who has authority to approve, override or stop it?
- How will we know if its behavior drifts?
Another executive in the research makes the human side clear: agent recommendations still need checks and balances, and procurement leaders must review commercial exceptions, supplier realities and key terms before anything is executed.
This does not slow adoption down. It makes adoption defensible, which is what allows autonomy to expand over time.
The readiness question
Governance is only one part of readiness. The Index reports organizational readiness across four areas, and the results show how much work remains. On the sourcing-process readiness dimension, 17% are in the scaling tier; that does not mean 17% are ready across every readiness dimension.
Looking at each dimension in turn:
- Strategy and leadership: 37% nascent, 51% developing and 12% scaling
- Organizational readiness: 34% nascent, 52% developing and 14% scaling
- Sourcing process: 36% nascent, 47% developing and 17% scaling
- Skills and talent: 43% nascent, 45% developing and 12% scaling
Skills and talent are the weakest dimension, with 43% describing their maturity as early stage. The research notes that skills and talent development are lagging other organizational readiness capabilities. Interestingly, skills gaps are cited as a barrier by only 8% of respondents, which suggests many organizations have not yet recognized how much capability building lies ahead.
Why readiness pays off
The case for investing in readiness is strong. The Index compares organizations that score 5.0 or higher on a readiness dimension, meaning scaling or highly evolved, with the overall median. Respondents scoring 5.0 or higher on a readiness dimension show substantially higher reported rates of significant-or-better improvement than the overall median comparison; this is an association, not proof of causation:
- Sourcing productivity: 73% report significant or better improvement, 2.3 times the median
- Spend under management: 60%, or 2.4 times higher
- Spend savings: 53%, or 2.2 times higher
- Category, supplier and market intelligence: 47%, or 2.0 times higher
Readiness, then, is not a soft prerequisite. It is associated with materially better results.
A phased path that builds confidence
How do you build governed autonomy without stalling? The executives in the research favor a phased approach. One describes introducing one use case or business area at a time, validating behavior and outcomes, and then expanding, noting that a phased rollout limits disruption and builds confidence.
The maturity playbook offers a structure for this:
- Prove (nascent): Run one governed use case on a high-volume, low-risk activity, with trusted data, human approvals, guardrails and audit evidence. Assemble a cross-functional pilot team that includes the process owner, digital or IT, a CPO sponsor and a change lead.
- Standardize (developing): Standardize two or three proven workflows, connect core applications and data, set portfolio stage gates, train roles and retire off-system workarounds.
- Expand (scaling): Extend proven flows across categories, apply common controls and build outcome ownership through a process council, with the CPO, risk and business leaders accountable.
- Optimize (highly evolved): Operate an integrated agent ecosystem with governed autonomy, tie funding to verified outcomes and redesign roles around enterprise value.
Notice that accountability shifts as maturity grows, from a pilot team toward a process council and, eventually, the CPO, CFO and executive team. Governance scales with the program.
Choosing technology that supports governance
Governed autonomy is as much a platform question as a policy one. You need approvals, guardrails, rationale and audit trails built into the way agents work, rather than layered on afterward.
Zycus, the Index’s partner, describes its Merlin Agentic AI Platform as delivering outcomes under full governance and control, with agents covering intake, sourcing events, negotiation and analytics. Its Autonomous Negotiation Agent (ANA), for example, is designed to run supplier negotiations with limited human involvement, which is the kind of lower-value, repetitive activity where the risk-based approach above can apply.
As with any vendor, validate the specifics against your own risk thresholds and policies, and be wary of claims that skip human oversight for strategic decisions. The Hackett Group does not endorse any vendor or product depicted in its research.
The bottom line
The goal is not maximum autonomy. It is the right autonomy, matched to risk, built on ready data and people, and measured against outcomes. The organizations pulling ahead are doing the unglamorous work of governance and readiness first.
The full report goes deeper than this post, with the readiness tier comparison across all four dimensions, the complete maturity playbook with accountable roles and metrics at each stage and recommended next steps for benchmarking your own position.
Download The Autonomous Sourcing Index and design your path to governed autonomy.
For an independent evaluation of supplier value management platforms, see Zycus in The Forrester Wave™: Supplier Value Management Platforms, Q4 2026. This evaluation is separate from the Autonomous Sourcing Index.
For additional context on organizational AI adoption, read The Hackett Group Agentic AI in Procurement Adoption Index. This is a distinct research study.
For an adjacent governance perspective, read The Hackett Group’s The Shadow Agent Estate. It examines the organizational risks of unmanaged AI agents rather than the sourcing benchmark measured in this article.

