When the Vendor Says the Agent Is Only 5%: What Infor’s Hospitality Strategy Reveals About the Agentic AI Gap
The enterprise software market is full of announcements about AI agents. But Infor’s recent commentary on hospitality takes a notably different approach: it argues that the agent itself is only a small part of the overall AI equation.
In two articles published in July, Infor highlighted three issues that are often overlooked in the rush towards agentic AI.
AI capabilities are becoming increasingly commoditised. Many organisations are struggling to turn AI investments into measurable operational value. And, according to Infor, the agent itself represents only around 5% of the work required to make an AI initiative successful.
The remaining 95% sits in areas such as data, context, governance, implementation and organisational adoption.
That argument deserves attention because it shifts the conversation away from what an AI agent can demonstrate and towards the infrastructure required to make it useful in a real business environment.
Is Enterprise AI Becoming an Arms Race?
Alan Young, Infor’s Vice President of Product Management for Hospitality, describes the current market as an AI adoption arms race.
His commentary points to several examples of organisations experiencing problems after introducing AI, including a widely reported case involving unexpectedly high AI token costs. He also references concerns from business leaders about AI expenditure exceeding productivity improvements, alongside a case involving a Pizza Hut franchisee where an AI-enabled delivery system reportedly contributed to a major operational failure.
The important point is not simply that these projects encountered problems.
Infor’s analysis suggests that the technology itself was not necessarily the fundamental issue. In the Pizza Hut example, the system was reportedly poorly suited to the operational realities of the particular business, which relied heavily on DoorDash. The implementation also lacked sufficient adaptation, training and ongoing support.
This highlights a broader problem with enterprise AI.
A system can work exactly as designed and still fail to deliver value if it does not fit the business process around it.
Context, implementation, change management and accountability can therefore matter just as much as the underlying AI model.
The Commoditisation of AI Features
One of Infor’s more significant observations is that the hospitality technology market is becoming increasingly difficult to differentiate through AI features alone.
Capabilities such as natural-language interaction and integrations with property management systems, once viewed as differentiators, are increasingly available from multiple providers at broadly comparable price points.
For buyers, this changes how AI vendors should be evaluated.
Instead of focusing primarily on demonstrations or claims about model sophistication, organisations should examine the supporting capabilities that determine whether an AI deployment can operate successfully over the long term.
These include:
- Industry experience
- Data and security architecture
- Governance and compliance
- Implementation methodology
- Change-management capabilities
- Ongoing support
- Post-go-live accountability
In other words, the AI feature may be becoming a commodity, while the ability to implement, govern and operate it effectively becomes the real differentiator.
The 95/5 Problem
Infor’s second article develops the argument further through its Architecture of Intent framework.
The framework was discussed by David Poprawka, Innovation Strategist at Infor Hospitality, during a session at HITEC 2026.
The central idea is that businesses can have AI, automation and conversational interfaces without necessarily having the foundations required to make those technologies useful.
Infor’s framework is built around three core elements:
- Clearly defined intent — understanding what the business wants the system to accomplish.
- A unified data layer — referred to by Infor as the Portico.
- Contextual understanding — giving the system knowledge of the property’s specific rules, processes and workflows.
Only after these foundations are in place does the AI agent become useful.
This leads to Infor’s 95/5 argument: the agent, chatbot or automated interface may account for only around 5% of the overall implementation challenge, while the remaining 95% involves the data, processes, context and governance underneath it.
The concept extends well beyond hospitality.
Across enterprise software, AI demonstrations tend to focus on the visible layer — the chatbot answering a question, the agent completing a task or the automation triggering an action.
The less visible work involves connecting data, defining permissions, establishing governance, understanding business processes and ensuring employees can trust the results.
That foundation is arguably where much of the real complexity lies.
Where the Vendor Argument Needs Scrutiny
Infor’s analysis provides a useful warning about the limitations of AI demos. However, buyers should also recognise that the company’s recommendations ultimately support its own market position.
Young argues that an impressive demonstration from a new entrant cannot replace the experience, compliance capabilities and institutional knowledge built by established enterprise technology providers.
That may be a reasonable consideration, but it should not automatically make an incumbent the preferred choice.
Infor’s own framework should be applied to Infor with exactly the same level of scrutiny as any other vendor.
If data architecture, governance and post-go-live accountability are the real differentiators, buyers should ask Infor to provide evidence of those capabilities rather than accepting them as part of the product narrative.
The same principle should apply to every vendor competing for the business.
The Bigger Competitive Question: Vertical or Horizontal AI?
There is another issue beneath Infor’s argument.
Its Architecture of Intent approach can also be viewed as a competitive response to technology companies building horizontal AI and orchestration platforms above traditional systems of record.
Companies such as Palantir, ServiceNow, Microsoft and Salesforce are developing technologies designed to coordinate work across multiple enterprise applications.
This creates a strategic question for ERP customers:
Should AI intelligence live inside the industry-specific ERP environment, or should an independent orchestration layer sit above the ERP and other business applications?
Infor’s position is that AI should be deeply grounded in the industry’s operational context.
Hospitality businesses, for example, have highly specific processes, terminology, rules and operational requirements. An AI system that understands those factors may have advantages over a generic layer attempting to interpret them from outside the industry stack.
Horizontal technology providers can make the opposite argument: that contextual intelligence can be developed through data, integrations and configuration, without requiring the orchestration layer to belong to the underlying ERP vendor.
Neither approach has definitively won the argument.
The outcome will depend on how effectively each model performs in real-world deployments, particularly as AI moves from providing recommendations to taking autonomous actions.
What This Means for ERP Buyers
Evaluate the 95%, not just the 5%. AI demonstrations can be impressive, but buyers should spend as much time examining the underlying data architecture, governance, implementation approach and support model as they do testing the agent itself.
Model the failure scenarios before deployment. Unexpected consumption costs, poor workflow fit and inadequate user adoption can quickly undermine an AI investment. Organisations should identify these risks during the pilot stage and deliberately test how the system behaves under difficult operational conditions.
Look beyond the AI feature list. If natural-language interfaces, assistants and basic agents are becoming widely available, vendors will need to differentiate through implementation quality, industry knowledge, security, governance and measurable business outcomes.
Treat the vertical-versus-horizontal decision as an architectural choice. Choosing whether AI should sit inside an industry ERP or operate through an independent orchestration layer can influence integration costs, data ownership and vendor dependency for years. Buyers should consider open APIs, portable data and contractual flexibility before committing.
Demand evidence after go-live. The strongest test of an AI platform is not the quality of its demonstration but what happens once employees use it in production. Vendors should be able to provide measurable outcomes, reference customers and evidence of ongoing governance.
The Agent Is Only the Visible Part
Infor’s 5% argument cuts through much of the current excitement around agentic AI.
The agent is the part customers see. The difficult work happens underneath it.
Data needs to be connected. Processes need to be understood. Context needs to be defined. Governance needs to be established. Employees need to adopt the technology, and someone needs to remain accountable when the system makes a poor decision.
For ERP leaders, this changes the question they should be asking.
Instead of asking “What can your AI agent do?”, the more revealing question may be:
“What have you built around the agent to make sure it works safely and effectively in our business?”
That is where the difference between an impressive AI demonstration and a sustainable enterprise deployment is likely to emerge.