Enterprise AI’s real product is not the agent – it is the 95% you do not see
Enterprise AI adoption costs are the total burden of making AI agents work inside complex organizations, including data unification, workflow redesign, system integration, and change management, which together dwarf the effort of building the model or chatbot itself and often decide whether projects deliver measurable value or fail outright. Infor has now said the quiet part out loud: the agent, chatbot, automation, or augmented reality overlay is only about 5% of the work. The other 95% lives in intent design, data plumbing, governance, training, and operational fit. That admission cuts through years of AI vendor claims reality distortion, where demos imply that enterprise workflow automation arrives once you click “enable AI.” If you are budgeting for licenses while ignoring integration, migration, and process change, you are budgeting for disappointment.
Infor’s 5% confession: when vendors admit where the pain really is
In hospitality tech, one vendor has broken ranks. In two July posts, an executive there called the current landscape an “AI adoption arms race” and detailed how a company managed to run up roughly USD 500 million (approx. RM2.3 billion) in AI token spend in a single month. Another example: an AI-driven delivery platform at a Pizza Hut franchisee helped trigger an operational collapse that wiped out more than USD 100 million (approx. RM460 million) in business and enterprise value when it clashed with a DoorDash-dependent operation and lacked adaptation, training, and support. Those are enterprise AI adoption costs in their rawest form: not model inference prices, but badly modeled failure modes. Then came the sharper point. Recapping a hospitality event, the same vendor said the agent itself is only 5% of the work, with data unification, contextual grounding, and governance making up the other 95%. That is a direct challenge to almost every agentic AI implementation pitch on the market.
Property ERP: where AI becomes credible – and shows the limits
If Infor explains why agentic AI implementation is hard, property ERP shows where it starts to make sense. At a recent property summit, a major housing and real estate ERP provider put artificial intelligence and “future readiness” at the center of its agenda, arguing that early AI investments respond to skilled labor shortages and help customers through AI integrated directly into its property management system. The company is embedding AI into specific workflows: invoice intake, mail processing, operating cost settlement, and day-to-day property management operations. Its AAVA assistant already automates incoming mail and invoice processing, and it announced an AI-optimized operating cost settlement process that its CEO called a milestone for simplifying processes and unlocking additional value. This is enterprise workflow automation at its most credible: high-volume, repetitive tasks inside an ERP, grounded in domain templates rather than generic copilots. But even here, the 95% looms large.
The migration and data iceberg: why “practical” AI still costs a fortune to adopt
Look under the surface of that property strategy and you see the same iceberg Infor described. The housing ERP in question is built on SAP S/4HANA with an industry template that customers can run on-premise, in the vendor’s data center, or via SAP’s cloud ERP. That sounds tidy until you meet the migration clock: mainstream support for SAP ERP 6.x ends in 2027, so laggards must complete their move to S/4HANA to secure maintenance until 2040. According to research cited there, 55% of organizations have deployed SAP S/4HANA or its cloud version, but only 34% have completed the transition. In the same study, 43% of organizations say SAP’s AI announcements are the top external factor shaping ERP strategy, ahead of the 2027 support deadline at 39%. That means AI buzz is pulling buyers into transformations they have not finished planning, while data quality and cloud architecture work – the true 95% – remain underfunded.
How enterprise buyers should budget for AI: price the 95%, not the 5%
The market signal is clear: AI features have commoditized, adoption failures are common and expensive, and vendors that still sell “our model is better” are selling yesterday’s advantage. In hospitality, one executive argues buyers should evaluate AI claims through factors like industry experience, security and governance architecture, implementation methodology, and the accountability structure that survives long after contract signature. In property ERP, the most convincing AI comes from narrow, measurable workflows such as invoice processing and operating cost settlement, not catch-all copilots. To align AI vendor claims reality with outcomes, buyers must treat the agent as 5% and budget for the rest: data architecture, integration into live workflows, training, support, and post-go-live optimization. Price the failure modes before you fund the software: the enormous token bill and the Pizza Hut collapse happened because no one modeled what occurs when AI hits messy operations without adaptation, training, and support. If your RFP centers on features and license costs, you are underwriting someone else’s demo, not your own success.
