Your Sourcing Agents Are Only as Good as Your Data: The Barrier Holding Back Agentic AI

Picture of Zycus

Zycus

Published On: 11/17/2026

Group-1000005301.png

Listen to this blog

Your Sourcing Agents Are Only as Good as Your Data: The Barrier Holding Back Agentic AI
Watch the on-demand webinar

Listen to this blog

Ask a procurement leader what is stopping agentic AI from scaling and you might expect answers about technology maturity, budget or talent. The research says otherwise. 

The Hackett Group’s The Autonomous Sourcing Index asked senior sourcing leaders to identify their most significant barriers to adopting and scaling agentic AI. The top answer is not close to the others. 

The barrier ranking 

More than half of respondents, 53%, cite data quality or availability. Integration with enterprise applications comes second at 32%, and organizational change resistance third at 27%. Trust, explainability or reliability follows at 20%, then governance, risk or compliance concerns at 18%, unclear business case or ROI at 15% and funding or competing priorities at 13%. 

Now look at the bottom of the list. Skills or talent gaps are cited by 8%, and vendor or technology maturity by just 3%. 

Vendor or technology maturity is rarely named as the primary obstacle. Instead, the underlying technical and organizational foundations – especially data quality and enterprise integration – are the more frequently cited constraints. 

Platforms are ahead of the foundations 

The readiness data explains why. The Index assesses technology readiness along two lines: IT operations and platforms, and application integration. The two are not equally developed. 

On IT operations and platforms, 30% of organizations are nascent, 50% are developing and 20% are scaling. Application integration is weaker: 45% are nascent, 43% are developing and 12% are scaling. 

Sourcing data is weaker still. When respondents rate how prepared their sourcing data is for agentic AI, only 2% say fully prepared and 8% advanced. At the other end, 33% describe their data as developing, 32% as initial and 5% as not ready. Overall, just 10% rate sourcing data as advanced or fully prepared. 

The research summarizes the picture in one line: operations and platforms are further along, while integration and trusted sourcing data remain the key constraints on scaling agentic AI. 

Why data hits agents harder than it hit analytics 

Procurement has lived with imperfect data for decades. Spend cubes were cleaned, with effort, and reports were reconciled by hand. Why is the problem more urgent now? 

Because an agent does not just report on data, it acts on it. A dashboard with a duplicated supplier record shows a wrong number and someone notices. An agent that sources from a duplicated record or applies the wrong policy because category data is inconsistent, produces a wrong decision at speed. 

One procurement executive quoted in the research states the principle plainly: autonomy starts with clean, standardized master data, and even strong agents will return weak decisions when supplier and category inputs are fragmented or inconsistent. 

That is a useful standard to hold your own program against. Before asking whether an agent is capable, ask whether the data it will read is trustworthy. 

The integration gap 

Data quality is only half the problem. Agents need to move through the systems where sourcing actually happens intake, contract repositories, supplier management, ERP and analytics. 

The Index measures the degree of application and data integration around agentic AI. Only 10% of organizations report high integration. About 28% report moderate, 24% basic, 23% minimal and 15% none. In other words, roughly six in ten organizations are operating with basic integration or less. 

This helps explain one of the Index’s other findings: that organizations with connected, orchestrated architectures report better outcomes than those with standalone tools. If an agent cannot read from and write to your core systems, a human fills the gap, and value leaks out at every handoff. 

What organizations are doing about it 

The research also captures the mitigations leaders are applying. They are practical rather than exotic: 

The pattern is that data and integration work comes first and that the same agents can then compound. The report’s own advice is to fix integration and data first. 

A sequence that makes sense 

If you take one thing from the data, take the order of operations. A reasonable sequence looks like this: 

  1. Pick the data that matters for one use case. Rather than cleaning everything, identify the supplier, category and policy data one governed use case actually needs. The maturity playbook for early-stage organizations recommends exactly this: select a high-volume, low-risk activity and define trusted data, approvals and guardrails around it.
  2. Connect before you multiply. Integrate the core applications that use case touches, so inputs flow in and outputs flow out without manual re-keying.
  3. Instrument data quality. Treat data quality as something you measure continuously, not a one-time cleanup. At the scaling stage the playbook calls for instrumenting performance and data quality.
  4. Reuse the pattern. Once one workflow runs on trusted data with proper integration, standardize that pattern for the next.

Where platform design helps 

How a platform handles data and integration can make this sequence easier or harder. Zycus, the Index’s partner, builds its Merlin Agentic AI Platform on an integrated source-to-pay core with native intake and analytics agents, such as Merlin Intake and Merlin Analytics. The intent is that agents work from a shared transactional foundation rather than from disconnected extracts. 

Whichever route you take, test any platform against your own data conditions. Ask how it handles inconsistent supplier records, how it connects to your ERP and what evidence it keeps for each recommendation. The Index is vendor-neutral, and The Hackett Group does not endorse any product it depicts. 

The bottom line 

Most procurement organizations are not blocked by a shortage of clever agents. They are blocked by data that is not ready and systems that do not connect. That is good news in one respect: those are problems procurement and IT can solve with discipline, and solving them benefits every agent you deploy afterward. 

The full report goes further. It shows how data readiness, integration and skills compare across the four readiness dimensions, how ready organizations are about twice as likely to achieve significant outcomes, and what a prioritized gap map looks like. 

Download The Autonomous Sourcing Index and see how your data and integration readiness compare. 

To explore broader adoption considerations beyond sourcing, see The Hackett Group Agentic AI in Procurement Adoption Index. Its findings should not be conflated with this sourcing-specific benchmark. 

For further reading on procurement leadership challenges involving fragmented technology and data, see the Forrester report on the new CPO’s first 90 days in the AI era. Its analysis is distinct from the sourcing-readiness findings cited here. 

Build or Buy Your Procurement AI?

Build or Buy Your Procurement AI?

Share:

Zycus
Zycus delivers Autonomous Procurement, powered by Agentic AI, on the Merlin Agentic Platform. Its Intake-to-Outcomes approach connects intake, sourcing, negotiation, contracts, supplier management, procure-to-pay and accounts payable in one source-to-pay suite, with products such as Merlin Intake, Merlin Autonomous Sourcing and Merlin Autonomous Negotiation Agents. Recognized by Gartner, Forrester and IDC, Zycus helps enterprises move procurement from transactions to outcomes.

Analyst Reports on Agentic AI

Subscribe to Blogs!

Get the latest blogs, insights, tips and exclusive content delivered to you inbox, Join Now

Subscribe by email

Opens your email app, addressed to crm@zycus.com.

Recommended blogs 

Intake to Outcome demo

Please share all the required details by sending an email to crm@zycus.com. Our team will review your request and assist you further.