The real breakthrough will not be a better chat box. It will be an adaptive layer that turns intent into coordinated action and disappears when no interaction is needed.
The interface has not caught up with the intelligence
I have been thinking about whether enterprise software has kept pace with the intelligence now available to it. My sense is that we have upgraded what the technology can understand, but not yet fully reconsidered how people should experience it. That question has become more concrete for us this year, as we have worked with SAP to bring procurement orchestration directly into systems enterprises already run, rather than asking them to replace what is already there.
AI can interpret language, understand images, analyze documents, reason across information and take action. Yet most enterprise work still begins in a familiar way. Someone opens a laptop, finds the right application, navigates a menu, completes a form and waits for the request to move between teams.
Even many AI experiences begin with a chat box. That is a meaningful improvement on learning a complex application, but it is still a box on a screen. It asks the user to stop what they are doing, translate their need into words and manage another digital interaction.
The larger opportunity is not to replace every form with conversation. It is to redesign the boundary between human intent and enterprise action.
There should not be one interface for work
I do not think the interface of the future will be a single interface. It will change according to the person, the task, the environment and the consequence of getting the decision wrong.
It will listen when speaking is the most natural way to express a need. It will see when an image carries more useful information than a written description. It will present relationships visually when patterns matter, and speak when the user is moving and cannot look at a screen. When routine work is already understood and authorized, it may not ask the user to interact at all.
Research comparing speech with mobile typing found that speech recognition was almost three times faster for short English messages. That does not mean everyone should talk to software all day. Voice is a poor choice in a confidential meeting, a noisy warehouse or when exact figures must be reviewed. The point is that typed text is now a design choice, not a universal requirement.
A 2026 study of 34 professional naval maintenance workers completing 3,128 subtasks found that augmented-reality guidance reduced completion times compared with desktop instructions. The benefits were strongest for less experienced workers and when people switched between task types. The broader lesson is that information becomes more useful when it appears in the environment where the work is happening.
This matters because work is already fragmented. Microsoft reports that employees are interrupted, on average, every two minutes during core working hours, while 48 percent say their work feels chaotic and fragmented. Another application or dashboard can add coordination work even when each individual feature is well designed.
The same principle applies to every interaction. A useful interface should respond to context rather than force context into one standard channel.
What an adaptive interface looks like in procurement
Consider an engineer standing beside a failed pump on a production line. The operational need is clear: restore production by Friday. The administrative route is not. The engineer may not know the component number, the category, the correct buying channel, the relevant contract or which teams need to be involved.
Today, the engineer may have to leave the work, find a laptop, identify the right system and complete a request designed around organizational ownership. Each step asks them to translate a business outcome into procurement administration.
An adaptive interface would let the engineer point a camera at the pump and explain the problem. The system could combine the image, voice, location and operational deadline with the asset record, service history, available inventory, supplier agreements and production impact. It could check whether a spare is already on site, whether an approved supplier can meet the deadline and whether an expedited purchase falls within delegated authority.
The engineer might see only one question: Should they approve the premium for next-day delivery or accept a two-day delay? Procurement might be involved because a commercial exception is required. Finance might be involved because the cost exceeds authority. If neither is necessary, the routine work could continue without creating another inbox for either team.
Now consider a marketing leader engaging an agency for a new campaign. The interface may begin with a statement of work dropped into an existing collaboration channel. It could identify the intended outcome, personal-data exposure, intellectual-property terms, budget and deadline. Instead of presenting a generic supplier-onboarding form, it might ask two missing questions, route a privacy review, reuse a valid supplier record and show the leader a short decision summary.
For a CPO reviewing supplier resilience, the same underlying intelligence may appear as a ninety-second audio briefing between meetings, then become an interactive network of suppliers, sites, materials and incidents when deeper investigation is needed. One intelligence layer, three very different interfaces.
From forms to intent
Forms are useful because they create structure. They help organizations collect consistent information, apply policy and create an audit trail. The problem is not the form itself. It is making the form the default starting point for every kind of demand.
A form transfers organizational complexity to the requester. It assumes that the person knows which function owns the need, which system to use, which information matters and what process should follow. We have become so accustomed to this translation work that we often treat it as unavoidable.
AI gives us the opportunity to reverse that relationship. Demand could arrive as a conversation, image, email, document, system event or pattern in operational data. The interface could establish the intended outcome, gather missing context and introduce structure only where verification, comparison, approval or auditability require it.
This is where orchestration matters. The employee should not have to navigate Procurement, IT, Legal, Risk and Finance as separate organizational queues. Orchestration should manage those dependencies behind the experience, coordinating people, processes, systems, data, decisions and AI agents around the outcome.
Orchestration also cannot assume a blank slate. Most enterprises already run SAP or other ERP and procurement systems shaped by years of investment. An adaptive interface that arrives as another system of record would simply become one more queue for the engineer or marketing leader to navigate. The value comes from coordinating against the enterprise that already exists, connecting to SAP and the rest of the stack rather than replacing it, so the people, policies and systems underneath can stay as they are while the experience on top changes. That is the premise behind our recent work with SAP: orchestration that works with the systems already in place, not a rebuild that asks the organization to start over.
A conversational front end is not a contextual system
A natural interface is only as useful as the architecture behind it. An agent may speak fluently and interpret an image, but that does not mean it understands the organization in which it is operating.
A prompt is closer to a job description than a brain. A capable enterprise agent also needs current context, trusted records, relationships, unstructured evidence, policies, permissions, tools, memory and feedback. It needs to know not only that a contract exists, but which supplier, asset, business requirement, risk, decision and employee are connected to it.
Systems of record remain essential for trusted facts and transactions. Unstructured information preserves the language and nuance contained in contracts, emails, meeting notes, images and reports. Graphs can make the relationships between those sources part of the information model. Retrieval brings the relevant context into the moment, while policies and permissions govern what an agent can know and do.
Research into GraphRAG shows why relationships matter when questions require sensemaking across an entire body of information, rather than the retrieval of one relevant passage. Research into generative agents makes a related point: memory, reflection and planning materially affect coherent behavior. The language model provides general capability. The surrounding architecture makes that capability relevant to a particular organization.
AI should remove administration, not accountability
A more natural interface should not become an excuse for invisible decision-making. The less the user sees, the more deliberately the organization must design authority, evidence, controls and escalation.
For routine, low-risk and reversible work, an agent may be allowed to act and record what it has done. When confidence is low, financial consequences are material, policies conflict, or commercial judgment is required, the interface should slow down and make the decision visible.
A supplier bank-account change is a good example. The system may collect the request, compare it with existing records, identify anomalies and prepare the verification steps. It should not quietly complete a high-risk change merely because the interaction feels seamless. The right interface is the one that brings an accountable person into the process at the right moment, with the evidence needed to exercise judgment.
This is an operating-model question as much as a technology question. Roles, decision rights, risk tolerances, incentives and governance determine when an agent can proceed, when it must recommend and when it must stop.
Five questions for leaders
Where are people translating intent into administration? Look for moments when employees must choose a system, category, process or owner before they can express the outcome they need.
Which context should change the experience? Role, location, urgency, value, risk, device and working environment may all affect the best interaction.
What should remain visible? Decisions involving judgment, ethics, material risk or commercial consequence need explanation and accountable ownership.
What intelligence sits behind the interface? Trusted records, documents, relationships, policies, permissions and feedback are what make an agent organizationally useful.
What are we measuring? Completion inside an application is a weak measure of experience. Time to outcome, interactions removed, adoption, control effectiveness and decision quality are more revealing.
Where Intake & Orchestration fits
The 'Idea-to-Pay' concept is that people should not need to understand organizational structures and systems before they can ask the organization for help. They should be able to express what they are trying to achieve, while the platform coordinates the people, policies and systems required to support them.
I also think it is important to distinguish a design direction from a claim about what has already been delivered. Like much of the enterprise software market, today’s experience remains more screen-led and text-led than the future described here. Voice, visual intelligence, audio, graphs and contextual memory create opportunities to go further.
For me, that is the next design challenge for orchestration: not simply giving every user a better place to manage work, but reducing how much of the machinery of work they need to manage at all.
The best interface may be no interface
Sometimes the right interface will be a conversation. At other times, it will be a visual explanation, an audio briefing, an interactive graph or a traditional form. There will also be situations where the system has understood the need, completed the authorized routine work and involved a person only when their judgment adds value.
We should stop measuring enterprise software only by how many activities it allows people to perform inside it. We should also ask how much unnecessary interaction it can remove.
The interface of the future is not a screen that people learn to navigate. It is an adaptive boundary that understands intent, applies organizational context, coordinates action and disappears when human interaction adds no value.
Want to learn how ORO can transform procurement at your organization? Book a demo with one of our experts.
