BCG's procurement study: barriers come before autonomous buying

External-source research synthesis, not an OfferSignal client result. Based on the retained October 2026 research and source-specific dates.

A procurement assistant can find a supplier or assemble a request for quotation long before a business is ready to let it choose, negotiate or place an order. The difficult questions are whether its records are current, whether sensitive drawings can be used, how legacy systems connect, and who remains accountable when the recommendation is wrong.

What BCG studied

BCG's 21 July 2026 article describes a survey of more than 200 chief information officers, procurement leaders and specialised technology-procurement buyers across North America, Europe and Asia-Pacific, supplemented by interviews with procurement and technology leaders, vendors and global system integrators. The six use cases were vendor discovery, request-for-proposal generation, contract analysis, spending analysis, risk monitoring and supplier-performance management. BCG reports internal value—productivity, cycle time, manual effort and compliance—appearing before external supplier and commercial value, with production deployments reporting stronger outcomes than pilots. These are survey and interview findings, not a controlled experiment or a manufacturing sample. [1]

The barrier is governance as much as technology

Trust in autonomous decision-making was cited by 71% of respondents, security and intellectual-property risk by 66%, and regulatory uncertainty by 57%. Accountability for agent actions reached 53%, auditability 48% and the absence of structured key performance indicators 42%. Nearly half cited difficulty integrating with legacy enterprise resource planning and procure-to-pay systems; 47% cited integration complexity. About one-third cited data-preparation or system-modernisation cost, four in ten the challenge of running transformation alongside business as usual, and one-third an ineffective operating model. A global system integrator's 12-month implementation example is an interview illustration, not a benchmark. [1]

A clear majority retain human control, particularly organisations with less than three years of artificial-intelligence experience. BCG also describes an unnamed automotive manufacturer that moved from tender-assistant productivity gains to negotiation uplift and earnings-before-interest-and-tax contribution worth more than ten times the internal value after redesigning the full procurement lifecycle. No baseline, denominator, currency, period, attribution method or named customer is supplied, so this is an illustrative interview claim rather than reusable return on investment. [1]

Why this matters for bespoke manufacturing

The mechanism is plausible for a custom manufacturer that searches archived drawings and project records, then matches a current request against supplier capability, equipment, revision, material, tolerance, certification, capacity and lead time. But BCG did not test that workflow. In a no-stock-keeping-unit environment, a wrong revision can produce the wrong supplier match; missing tolerances can produce an invalid request for quotation; stale capacity can create an unfulfillable promise; and weak permissions can expose customer drawings. Data cleaning alone does not guarantee a valid model recommendation.

The wider European evidence points in the same direction without proving a procurement result. OECD reports manufacturing barriers in 2024 including lack of expertise above 7.5%, data availability or quality at 5.0%, equipment or software incompatibility at 4.8%, legal-consequence concerns at 4.9% and data-protection or privacy concerns at 4.4%. Eurostat reports that 19.95% of EU enterprises with at least 10 employees used at least one artificial-intelligence technology in 2025; among enterprises that considered AI but did not use it, lack of expertise was reported by 70.89%, unclear legal consequences by 52.52% and privacy concerns by 48.83%. These denominators and constructs differ from BCG's. [3] [5]

McKinsey's approximately 500-organisation 2026 trust survey offers directional corroboration: security and risk were the top barrier to scaling agentic AI, nearly 60% cited knowledge or training gaps, and about 30% reached maturity level three or higher in strategy, governance and agentic controls. [4]

An illustrative governed pilot

For a manufacturer, begin with supplier discovery, request-for-quotation drafting, contract analysis or risk monitoring. Preserve source and revision, permissions, units, currencies, certificates and exception reasons. The assistant can prepare evidence and highlight missing fields; authorised procurement, engineering and finance owners remain responsible for supplier selection, technical acceptance, price, order, certification release and payment.

Test retrieval of the authoritative revision, invalid or stale data, integration handoffs, unsupported claims, review time and correction rate. Measure qualified requests and downstream commercial outcomes separately from drafts and clicks. In Europe, the Commission says the AI Act became generally applicable on 2 August 2026, with certain high-risk obligations later depending on use; applicable controls include risk management, data quality, logging, documentation, human oversight, cybersecurity, robustness and accuracy. The dates are regulatory milestones, not evidence of compliance or effectiveness. [2]

The study therefore supports a staged, human-in-the-loop experiment. It does not demonstrate autonomous end-to-end procurement, bespoke quote accuracy, supplier pre-vetting, order conversion, logistics, certification or payment performance.

Sources and next reading

Explore OfferSignal · Focused Business & AI Workshop · All research

References

[1]: https://www.bcg.com/publications/2026/scaling-agentic-ai-in-tech-procurement "Scaling Agentic AI Across Tech Procurement Functions". [2]: https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai "Regulatory framework on artificial intelligence". [3]: https://www.oecd.org/en/publications/progress-in-implementing-the-european-union-coordinated-plan-on-artificial-intelligence-volume-2_3ac96d41-en/full-report/ai-in-manufacturing_5df4a60d.html "AI in manufacturing: Progress in Implementing the EU Coordinated Plan on Artificial Intelligence". [4]: https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/tech-forward/state-of-ai-trust-in-2026-shifting-to-the-agentic-era "State of AI trust in 2026: Shifting to the agentic era". [5]: https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises "Use of artificial intelligence in enterprises".