采购中失败的代理人工智能计划背后的三大陷阱

CLEARING THE HURDLESTO AGENTIC AI NEW TECH, FAMILIAR CHALLENGES The future may be agentic, but there are many traditional, familiar,issues to be faced before organisations can reap its benefits. In fact,just 8% of Procurement Leaders’ Ovation community – a group of CPOsat leading-edge procurement organisations – consider their function’sdigital capabilities to be highly mature. These issues revolve around change, transformation anddigitalisation. If AI is based on data, then is your business’s data fitfor purpose? Can its systems and APIs cope with another layer oftech? Will your people help drive AI’s successful take-up – or willthey block progress? SOURCEOvation Pulse 2025, Procurement Leaders 80%The percentage of AI project failures in which poor data quality is acontributory issue DETANGLING DATA DEFECTS Procurement data is often scattered across enterprise resource planning andprocure-to-pay systems while also being in unstructured formats. Its impacton even ‘starting pistol’ AI projects such as spend analysis can be flawed – sobusinesses miss out on straightforward, easy-win savings. “Poor outputs mean teams having to spend time validating information, thosewhich could have led to significant savings or missed opportunities,” saysBrian Bergerson, regional VP at GEP. SOURCE Essential Procurement Tech webinar, Procurement Leaders/GEP webinar, 2025 “Poor data quality contributesto project failures because largelanguage models ingest thatand try to compute from it” “If you don’t get yourdata right, you’re justautomating stupidity” Brian Bergersen, regional VP – software sales, GEP CPO, media company The percentage of companies thatsays legacy systems hinder theirAI adoption efforts Source:ForresterCloud Adoption Trends ENSURING THE TECH STACKS UP The ability for source-to-contract and P2P tools to process information quicklyis critical in aligning with the computations that agentic AI needs to take toprovide timely insights. While more than 86% of organisations have a dedicated digital P2P solutionin place, it’s no guarantee that these systems are ‘AI-friendly’. Without goodconnectivity and integration, then time spent on workarounds will erode the ROIthat the CFO has been pledged. SOURCE Data and Digital Cohort Pulse Survey, Procurement Leaders, 2024 “Your innovation can be hampered or throttledby the weakest link in the chain within yoursystems. A piece that’s behind the others,can’t connect etc. Unlocking that legacyroadblock and how to move beyondthat is critical to AI agent projects” “We want to transitiontowards becoming ahands-free procurementorganisation where all ourpeople can really free up theirtime for business partneringand building relationships withour strategic suppliers” Brian Bergersen, regional VP – software sales, GEP CPO, consumer goods company The percentage of CPOs confident in theirteam’s ability to effectively leverage AI toautomate processes in the next 12 months Source:GEPOutlook Report 2025 BUILDING DATA CAPABILITIES A lack of confidence in teams’ ability to handle an agentic AI-drivenenvironment is illustrated in Procurement Leaders’From chaos to clarity:delivering data to drive AI-powered procurementreport, which found limitedtechnical skills or experience was the biggest barrier to AI adoption. If teamscannot interpret data then, as with poor data outputs, the ROI of an AIimplementation will likely be lower than predicted. SOURCE From chaos to clarity: delivering data to drive AI-powered procurement, Procurement Leaders, 2025 “Teams often struggle to interpret AIoutputs and underlying data science.Immediate investment in targetedupskilling is essential – not to replacetalent, but to enable teams to worksmarter and deliver greater value.Clear change managementcommunications must emphasisethat AI elevates human expertiserather than displace it” “The future procurement workforceis not just tech-literate. It’s valuefocused, analytically savvy andcommercially sharp. Upskilling yourteam isn’t optional, it’s essential.” Rich Rivera, regional VP of software sales, GEP CPO, consumer goods company LOOK BEFORE YOU LEAP: HOW TO CLEAR THREE COMMON AI HURDLES 1. FRAGMENTED DATA 2. LEGACY SYSTEMS 3. INTERNAL SKILLS Steps to build AI readiness:Targeted training. Strategies to modernise:Assess compatibility. Steps to address:Assess data quality. llStandardise formats.lIntegrate sources.lImplement governance. llAdopt cloud solutions.lUse modular AI tools.lPlan phased upgrades. llHire experts.lFoster learning.lCommunicate benefits. Quick wins: Quick wins: Quick wins: lIntegrate AI with supplier portals forinstant automation.lUse middleware to bridge legacy systemsand AI tools. lUse AI-driven data-cleansing tools to fix errorssuch as duplicate records.lStart with high-impact datasets, for instance,spend data, supplier records. lUse AI copilots with intuitive interfaces toenable easy adoption.lStart with user-friendly tools requiringminimal exp