An Enterprise Execution OS connects understanding a business process, carrying out its approved steps and checking the result. Duvo uses the term for a platform that captures how a process runs across people and existing systems, supports execution within agreed controls and gives the process owner evidence for deciding what to improve.
Consider a carrier invoice that differs from the agreed rate. Finding the difference is useful, but finance still needs the applicable contract, shipment evidence, a person authorized to decide, a carrier response and an accurate payment record. The process includes that whole sequence. Some steps need a person, some use ordinary software and some can use AI to interpret evidence.
Key Takeaways
- An Enterprise Execution OS connects process capture, approved execution and result verification across the applications and people involved in a business process.
- Guided interviews and screen walkthroughs capture explanations, exceptions and handoffs; the relevant operators and process owner still need to review their coverage and accuracy.
- A first purchase can produce a standalone discovery output, reviewed audit findings or an SAP migration handoff. Recurring automation is a separate choice with its own scope and acceptance criteria.
Why processes break between people and systems
An ERP can record that an invoice arrived and was paid. It may not explain why an AP analyst accepted a quantity difference, where the delivery evidence was found or who agreed an exception with the supplier. Those explanations often live in email, spreadsheets and the experience of the people doing the process.
This matters in a retailer reconciling goods received, a manufacturer checking transport charges or a transformation team deciding which SAP requirement still serves a business purpose. The next person needs both the records and the reason for the decision. When that context is missing, they repeat the investigation or apply a rule to the wrong situation.
Duvo starts process capture with AI interviews, screen walkthroughs and available documents. The participant chooses when to start and stop. An operator can show the carrier portal and explain what happens when a reweigh certificate is missing; a follow-up question can establish who reviews the dispute and what evidence they require.
That account needs checking with the person receiving the handoff. An operator may describe how a process usually runs while finance applies a different approval rule. The reviewed record should preserve the discrepancy until the responsible people resolve it. One recording cannot establish every market, role or exception.
System-event analysis can corroborate frequency and timing where records exist. Screen capture adds the visible actions, and guided questions add explanations. These methods can complement each other. See the related guide to process mining and process discovery for the questions each method helps answer.
What understanding, running and improving a process mean
Understanding produces a reviewed description of the current process: its trigger, steps, decisions, owners and completion criteria. A proposed improvement remains a proposal until the process owner approves it. Removing an unnecessary spreadsheet join is different from removing the finance review that protects a payment decision.
Running means carrying the approved process through the relevant people and systems. In Duvo's operating model, the process defines required steps, permitted decisions and responsibilities. AI can interpret an invoice or help select an allowed route within a step. That role does not give it authority to skip a required approval.
The freight audit example makes the distinction concrete. The captured XPO invoice shows $3,550.00 billed against an expected $1,265.02, leaving a potential $2,284.98 discrepancy. The finding concerns reweigh evidence and a contract discount. The record presents an approval or dispute decision; it does not establish that the carrier accepted a claim or that cash was recovered.
A production evaluation should demonstrate the relevant controls before relying on them. Show what happens when approval is absent, a required document is missing or a write fails. Confirm the human handoff and the permission boundary in the target system. A convincing interface or a rule written in a prompt is insufficient evidence that an action cannot bypass a control.
Improving starts with a checked result. A submitted dispute, a corrected unpaid invoice and recovery of a paid overcharge are different outcomes. Preserve their evidence separately so the process owner can see what succeeded and what still needs attention. A recurring exception may justify a proposed rule change, but that change needs approval before it governs later runs.
This process perspective also appears in BCG's public analysis of scaling AI, which argues for examining the complete process rather than automating isolated tasks. It is wider industry context, not evidence of Duvo's delivery performance.
What the customer controls as models change
The customer remains responsible for the process outcome. That includes approving operating rules, assigning decision authority, providing appropriate access and deciding when a proposed change is acceptable. The customer also validates source records and supplies the people who can explain exceptions. Software cannot infer a missing commercial agreement into existence.
Usable process knowledge helps preserve continuity when tools or models change. The reviewed map records how the process works; the rules and supporting evidence explain why. BCG's public discussion of AI vendor dependence likewise recommends keeping business knowledge and rules under enterprise control while allowing model choices to evolve.
Make that ownership specific in a purchase. Identify the discovery outputs and process maps the customer receives, available export formats, the supporting evidence included and the agreed exit process. Duvo's published Process Intelligence capability includes customer-owned reviewed outputs and BPMN 2.0 XML export. BPMN is an Object Management Group standard with machine-readable schemas for process models.
A process-map export does not establish that every decision record or runtime configuration is portable, or that another platform can execute it unchanged. Inspect the promised export with the receiving team. If a model changes, evaluate the supported process again, including its permitted actions and human review requirements.
Processing arrangements are another decision. A compatible customer endpoint or dedicated model capacity may fit an existing requirement. Agree the scope, location and operating responsibilities for the process; an inference endpoint alone does not place the entire platform on the customer's premises. NIST's AI Risk Management Framework offers a voluntary basis for considering trustworthiness through design, use and evaluation.
How a contained first engagement leads to a useful result
Choose the first scope around a decision someone needs to make. A finance team may need to know whether its invoice records support a useful audit. A process owner may need a training guide. A SAP programme may need to decide which exceptions belong in the target design. Each can have an accepted deliverable without a company-wide automation programme.
For process discovery and improvement, agree the process boundary, participants and chosen output. Operators contribute walkthroughs and interviews; the team reconciles handoffs and reviews the result. Live ERP access is not required for that starting point. A reviewed catalogue, improvement plan or training output can be used by the customer's own team.
For freight or payables audit, start with the agreed records and checks. The deliverable should identify the source evidence, discrepancy, validation status and next decision. Missing records should appear as incomplete checks. A sample does not automatically include supplier follow-up, ongoing file delivery or production posting; each needs a supported scope.
For SAP migration, process discovery captures what happens around SAP. An optional read-only scan adds findings from the agreed datasets inside the system. The business owner and SI can use the combined evidence to decide what to standardize, extend, retire or keep outside the core. Their acceptance of the handoff is the first result.
Before starting, agree customer effort, prerequisites, delivery expectations, price and acceptance criteria. A date depends on usable inputs, the necessary access and reviewer availability. A buyer who already has sufficient proof can purchase production directly. Requiring another evaluation only helps when it resolves an actual buying question.
Expansion should follow evidence from the accepted process. Invoice reconciliation may expose a recurring master-data problem; a reviewed SAP process may identify an administrative step suited to automation. Each addition has its own owner and controls. Duvo's customer results show specific deployments and their evidence, while the platform explanation describes how the capabilities connect. Technical readers can continue to the developer resources.
Sources
- Scaling AI Requires New Processes, Not Just New Tools, BCG
- Do You Own Your Enterprise Cortex?, BCG
- Business Process Model and Notation 2.0.2, Object Management Group
- AI Risk Management Framework, NIST
Frequently Asked Questions
Q: What is an Enterprise Execution OS?
It is Duvo's category for a platform connecting process understanding, execution across existing systems and verification of the result. The process includes human decisions and ordinary software as well as model-assisted steps.
Q: Does an Enterprise Execution OS replace an ERP?
Duvo starts with the customer's existing systems. The agreed process defines where records are read or changed, with permissions and approval requirements scoped for each operation.
Q: Can we buy process discovery without buying automation?
Yes. A reviewed process catalogue, training output, improvement plan or transformation handoff can be the complete purchase. Further implementation is a separate decision and commercial scope.
Q: Do guided process interviews require live ERP access?
No. Discovery can start with participant-controlled walkthroughs, interviews and documents. A live system scan or production action has separate access requirements.
Q: Who decides whether an AI-assisted process can change?
The authorized process owner must approve changes to the governing process and controls. Results can support a recommendation. Define the approval requirement in the agreed operating process and verify it before relying on automated execution.
Q: What should we check before purchasing the first engagement?
Confirm the inputs, customer effort, controls, deliverable and acceptance decision, together with the pricing basis and delivery dependencies. Inspect relevant source evidence and test the supported controls. Choose the smallest scope that answers the decision your team needs to make.