Vertical Bar / Our thesis

Intelligence where the work happens.

Why the path to autonomous operations starts with the systems a business already runs.

A business does not operate inside a single application. An order may begin in a storefront, pass through an integration, become a record in an ERP, and trigger work in a warehouse. The people responsible for that journey see different parts of it. Improving the outcome requires understanding how those parts fit together.

The arrival of AI creates an opportunity to make this work easier. The “AI-native ERP” approach builds a new system around AI from the outset. A unified architecture can offer real advantages: consistent data, control over the application, and the ability to design automation into its core. For an organization choosing a new system, that can be a compelling proposition.

For an established business, there is a further question: must access to better intelligence depend on replacing the system of record? Our thesis is that a distinct opportunity lies in helping people understand and change the systems they already use. That calls for a workspace spanning the full course of operational work, from the first explanation of a problem to evidence that a change has improved it.

A system contains more than its data

An established ERP is also a record of how an organization has learned to operate. Approval rules reflect judgments about risk. Custom fields preserve distinctions that matter to a particular business. Workflows, integrations, and exceptions embody decisions made over years, sometimes with little documentation of why they were made.

Replacing that system therefore involves more than moving records. The organization must decide which behaviors to preserve, which to redesign, and how to validate the new arrangement. Data migration, integrations, testing, and user training remain separate responsibilities even when the destination has a better interface or a more capable model.1 A transition can be phased; its scope still extends beyond software installation.

This is not an argument for preserving every customization. Some complexity has outlived its purpose, and a system that no longer fits the business may deserve to be replaced. The relevant distinction is whether replacement is necessary to obtain the improvement being sought. An approval bottleneck or an unreliable order handoff may be addressable within the current environment.

Two approaches. Replacing the ERP: each new system arrives clean but inherits the same problem once it is in use. Improving in place: a shared layer forms above the existing ERP, learns how the business works, and fixes its records where they are.

That narrower decision matters. A team can investigate one process, test a proposed intervention, and assess its value before expanding the scope. The case for intelligence can be evaluated separately from the case for replacing an ERP. This preserves the option to modernize individual systems later, on their own merits.

Operating a system is different from owning it

When a team owns an application, it can inspect the source, change its internal behavior, and instrument the runtime. Tests and version control provide a foundation for reasoning about changes. Enterprise SaaS creates a different working environment: the vendor controls the underlying product, while the customer works through exposed interfaces, configuration, permissions, and extension points.

Calling these systems “black boxes” does not mean they are inaccessible. They may expose extensive APIs, logs, metadata, and scripting facilities. It means the customer’s view and authority are bounded. Available evidence and permissible actions differ by platform and role; even access to diagnostic logs can depend on permissions.2 An operating model for this environment must account for those boundaries.

Consider a Shopify order that passes through Celigo into NetSuite. A downstream failure might arise from a field mapping, a validation rule, a permission, or an assumption about what an order should contain. A developer can fix a script and still leave the business problem unresolved. The person who knows the intended behavior may be an operations user, while an administrator understands the configuration that enforces it.

The unit of work is a business change that crosses systems, people, and permissions.

A new ERP can improve the portion of the operation under its control. It does not, by itself, remove the need to coordinate with the other applications around it. We therefore see a lasting problem in the work between systems, including businesses that eventually adopt a new ERP.

Four connected platforms the business runs on but does not own. Seen from outside, attention only circles the boundary around them. Seen from above, one layer spans all four, sheds light over them, and understands what each one does.

The work needs a shared place to happen

Our vision for Vertical Bar is the place where administrators, developers, and business users carry a change from intent to outcome. Planning, domain expertise, implementation, deployment, and measurement should remain connected throughout the work.

The starting point is a business objective, expressed in terms the people doing the work can assess. “Reduce orders that need manual repair” is more useful than an instruction to edit a mapping in isolation. Domain experts contribute the exceptions and constraints. Technical owners connect them to records, workflows, scripts, and dependencies. The proposed change can then be reviewed against the reason it exists.

In this model, a shared record of the change brings together the objective, the evidence, the decisions, the implementation, and the observed result. Each platform retains its own permissions and execution mechanisms. The workspace provides continuity across them.

One insight layer above a conveyor of different platforms. Each platform stops beneath it in turn: its data rises into the layer, and the layer's light comes down to write the right change back into the platform.

This makes collaboration part of the reasoning process. An approver needs to see what will change and what has been checked. A business user needs to judge whether the result matches the intent. An agent needs both the technical context and the limits of its authority. Keeping these perspectives attached to the same work reduces the need to reconstruct them at every handoff.

Experience should leave something reusable

A transaction log records events. A configuration snapshot records state. Neither necessarily explains why a team chose a particular change, which alternatives it rejected, or whether the outcome justified that decision. Much of this knowledge remains in conversations, tickets, and the memory of individual specialists.

A workspace that participates in the full lifecycle has an opportunity to retain that missing context. For a later problem, the useful question becomes more specific than “Have we seen this error?” It becomes: “Under comparable conditions, what did we change, who validated the reasoning, and what happened afterward?”

The potential advantage is cumulative. Each well-documented change could make the next investigation easier by preserving dependencies, decisions, and results within the customer’s authorized environment. A new model can draw on that context; it does not have to rediscover the business from a schema every time.

That advantage is conditional. A large volume of events is not the same as reliable operational knowledge. Records must stay current, contradictions must be resolved, and outcomes must be linked carefully to interventions. The value would come from the quality of those connections and their usefulness in subsequent work. It has to be demonstrated in practice.

A record of changes rises and one of its changes is copied, as dots, into a layer of knowledge beneath it. A second record, whose change went wrong, rises and passes on; nothing of it is kept.

The test is what improves in the business

Our approach rests on three foundations: process analytics, deployment with guardrails, and prediction with simulation. Together they form a discipline for change. Process analytics establishes how work actually unfolds. Prediction and simulation help compare possible interventions. Deployment with guardrails connects a chosen intervention to the checks, permissions, and recovery options needed to carry it out.

A deployment that succeeds technically may have little effect on the process it was meant to improve. Conversely, a modest change to a dependency or approval step may matter more than a substantial rewrite. The relevant measures are those of the operation itself: elapsed time, exceptions, rework, and the effort required from people.

These capabilities become more useful when they share context. A model needs an account of the process it is trying to predict. A deployment needs a reason for the change and a way to evaluate it. The observed result must return to the same body of knowledge, so that the next decision can build on what the organization has learned.

Process analytics, simulation and guarded deployment sit on one loop with the operation itself. A line runs through all four in turn: analytics ripples, the chosen forecast is drawn, the deployment overwrites old fields with new ones, and more of the operation lights up.

Prediction and simulation inform a decision; measurement tests it against what follows. An improvement after a change does not automatically establish causation. Transaction mix, staffing, and other interventions may also contribute. A useful workspace should preserve the assumptions behind a prediction, the scope of the action, and enough operational context to evaluate competing explanations.

Autonomy can be built through accountable work

This is a demanding product to build. An external workspace has to maintain an accurate view of changing systems without controlling their internals. Some actions will remain unavailable through an API. Some outcomes will be difficult to attribute. Some decisions will require human judgment. Supporting many platforms adds breadth, but useful work depends on depth within each one.

Our starting point is NetSuite, where financial and operational processes meet. CrossCheck helps teams understand their environments, investigate issues, and manage changes with testing and approvals. Shopify capabilities extend selected backup and synchronization workflows. The broader workspace across NetSuite, Shopify, Celigo, and other platforms is the direction we are building toward. MONIAC is our research effort into process prediction and simulation.

Work with customers and design partners provides a basis for learning from real operations. The effectiveness of a complete autonomous system must still be demonstrated through repeatable improvements in diagnosis, execution, and business outcomes.

We expect useful autonomy to develop through progressively better-understood work. The organization supplies goals, domain expertise, and authority. The system helps assemble evidence, propose changes, carry out permitted actions, and measure their effects. As that process becomes more dependable, the scope of delegated work can expand.

NetSuite stands at the centre of a company's infrastructure among other platforms. Beneath it the depth hangs down like a still waterfall of dots, and blue spreads through the part beneath NetSuite.

The opportunity is to make a business more capable of changing itself. For many organizations, the most practical place to begin is with the systems, people, and operational knowledge already in place.

Sources and scope

  1. Oracle NetSuite, The Complete NetSuite Implementation Guide. Opens in a new tab Describes migration, integration, testing, training, and organizational responsibilities in adopting a new ERP. Referenced for the work involved in implementation, not as evidence about a particular competing product.
  2. Celigo, Investigate & troubleshoot errors. Opens in a new tab Documents diagnostic surfaces and their permission requirements. The order example in this essay is illustrative.

This essay sets out our product thesis. The proposed cross-platform workspace and the path to broader autonomy describe our direction. Current product scope is described on the CrossCheck and MONIAC pages.