
October 29, 2026
A customer sees one retailer. Behind that experience sits an estate of commerce platforms, point-of-sale systems, order management applications, warehouse systems, customer service tools, financial systems, and partner services.
Each application may perform its own function well. The difficulty emerges when a business process crosses their boundaries.
A store cannot find an item promised for pickup. A supplier delay threatens a promotion that marketing has already launched. A returned product has arrived at a warehouse, but the customer is still waiting for a refund. Resolving these situations requires people to connect information, make decisions, and coordinate actions across multiple systems.
For enterprise architects, these handoffs deserve the same attention as APIs and data pipelines. Process Tempo Jupiter provides a way to build governed operational applications around them, bringing information, forms, analytics, and workflow into the same working environment. The same foundation can make AI assistants more effective by connecting their recommendations to reliable evidence, authorized actions, and measurable outcomes.
Large retailers have invested heavily in integration. The remaining gaps are not always missing connections. They can also be missing business context, unclear decision rights, or work that no application owns from beginning to end.
An order management system may know that fulfillment failed. A warehouse application may know that a shipment is late. A customer service platform may record a complaint. Yet the retailer may still lack a shared process for deciding what happens next.
Four architectural gaps help explain the problem:
These gaps vary by retailer. Architects should first assess capabilities already present in their core platforms. Microsoft’s Dynamics 365 Commerce architecture, for example, documents interconnected processes and integration points across commerce and back-office applications. The opportunity for an additional application arises where a retailer’s actual cross-system process remains incomplete. [1]
When the official architecture does not support the complete task, employees assemble a working process themselves.
A regional operations manager creates a pickup exception spreadsheet. Procurement maintains a supplier recovery tracker. Customer service builds a small application to reconcile refunds. An analyst writes a script that merges exports from several systems.
These solutions often begin with a legitimate need and a capable employee. The architectural problem develops when they become essential to operations without clear ownership, controlled access, supported integrations, or lifecycle management.
The result can be an unofficial system of record for decisions. The ERP records the purchase order, but the spreadsheet records the promise to expedite it. The customer service application records the complaint, but an email contains the refund authorization.
This weakens transparency in practical ways. Managers cannot reliably see how much work is waiting. Decisions lack a consistent record of supporting evidence. Different regions adopt different rules. When someone leaves, the organization can lose both the tool’s maintainer and the knowledge of how the process works.
It also slows rapid decision making. Before choosing an action, people must determine which information is current, whether someone else has already acted, and who has authority to approve the next step.
Shadow IT is therefore a useful architectural signal: it reveals business processes that the supported application portfolio has not fully accommodated.
Jupiter is Process Tempo’s low-code application development platform. It combines enterprise data access, analytics, and application design. Its companion, Voyager, supports graph exploration and shareable snapshots of connected data. [2]
For a retail architecture, a useful role for Jupiter is the operational application spanning an otherwise disconnected process. Employees can review the relevant information and work through a defined response in one place.
Jupiter’s documented workflow capabilities include forms of structured coordination such as steps, owners, approvals, and routing. Its spreadsheet replacement capabilities include guided entry, validation, role-based access, and change history. These provide building blocks for moving an informal process into a governed application. [3][4]
The source systems should retain their established responsibilities. An order management system continues to own order allocation. Finance continues to own accounting records. Jupiter can hold the exception case, supporting evidence, decision, and workflow status, while approved integrations carry out the required actions in those systems.
Connections must follow the interfaces available in the retailer’s environment. Direct access can suit supported data sources. SaaS services and partner platforms may require APIs, adapters, or the retailer’s existing integration platform. Reading information and executing a business transaction are separate design decisions.

Retail AI assistants are increasingly tailored to different users. A store associate needs help with product questions and fulfillment issues. A customer service representative needs order history and return policies. A merchandiser needs to understand supplier performance and promotion exposure. Microsoft’s current business application documentation describes role-specific assistance across customer service, commerce, finance, and supply chain, alongside agents that can execute tasks. [7]
For assistants that stop at conversation, however, the employee still owns the gap between a recommendation and a resolution. “Transfer stock from another store” is useful advice, but someone must validate availability, obtain approval, initiate the transfer, and confirm that the customer’s order can be fulfilled.
A chatbot that summarizes a problem can save research time. Its operational value grows when it can participate in the process required to resolve that problem.
The same architectural gaps that frustrate employees can undermine an AI assistant’s answers. An inventory extract may show stock that has since been reserved. A supplier scorecard may omit a recovery commitment recorded in email. A return may appear unresolved because its inspection result sits in a separate application.
An assistant drawing on those sources could recommend an action that looks reasonable but is no longer appropriate. Giving separate departments their own chatbots can reproduce the fragmentation when each assistant sees only part of the process.
Connecting systems provides access to more evidence, but trust also requires consistent definitions, current records, and visible data quality. A connected source can still contain an incorrect stock count. The application needs a way to flag the discrepancy, assign a correction, and record verification.
Jupiter’s workflow capabilities provide a foundation for combining human decisions with AI assistance. Process Tempo describes embedded agents that classify requests, summarize information, draft responses, and populate data within structured processes. [3]
In the supplier disruption scenario, a configured AI agent could summarize the delay, identify affected promotions from approved sources, and prepare a recovery request. Jupiter could route the request to procurement and the appropriate approvers. Authorized integrations could then execute the approved changes and return their status to the case.
The workflow defines what the agent may do, which evidence it needs, when approval is required, and what confirms completion. Missing information should create a review task. An unsuccessful transaction should remain visible until someone resolves it.
This gives each user an assistant connected to a shared operational process, with access and authority appropriate to that person’s role.
A structured workflow can capture more useful feedback than a simple thumbs-up on an AI response. The retailer can record the recommendation, the evidence available at the time, the decision taken, the reason for an override, and the eventual result.
Suppose procurement repeatedly rejects an alternative supplier because its products lack a required certification. Capturing that reason exposes a missing qualification in the data or decision rules. Once reviewed and corrected, future recommendations can account for it. Recording whether an approved recovery actually protected the promotion provides a stronger measure of usefulness than approval alone.
That feedback can inform curated retrieval content, business rules, prompts, and evaluation cases. Changes should be reviewed and tested against outcomes; approval by a human does not establish that a decision was correct. This learning loop must be designed into the solution and does not imply that Jupiter automatically retrains an AI model.
Trust improves when users can inspect the basis of a recommendation and verify its result. A Jupiter application could be designed to show:
Enterprise architects can then assess AI using correction rates, override reasons, failed actions, and successful resolutions. The aim is to make reliance on AI proportionate to the evidence and its demonstrated performance.
The following scenarios illustrate how such applications could work in retail. They are proposed solution patterns, rather than accounts of named customer deployments.
A customer orders a coffee machine for collection that afternoon. The commerce platform accepted the order, but the store associate cannot find the item. The stock record could be wrong, or the available unit could be damaged.
In a disconnected process, the associate contacts another store, customer service opens a separate case, and a manager authorizes an alternative through email. Several people may contact the customer without seeing each other’s actions.
A Jupiter application could bring the order, current inventory information, picking exception, and customer service history into a shared case. The associate records what happened using a structured form. The application presents the relevant recovery options and routes any required approval.
One option might be a transfer from a nearby store. Another might be a substitute product or shipment to the customer. The order management system remains responsible for validating availability and confirming the chosen fulfillment action through its supported interface.
The case stays open until the action is confirmed and the required customer communication is recorded. A failed update becomes visible work for an owner, rather than an invisible integration error.
The intended outcome is a faster, coordinated response before the customer makes an unnecessary trip. Useful measures include time to acknowledge an exception, time to confirm recovery, and the proportion resolved before the collection deadline.
A supplier reports that a shipment will arrive late. Procurement knows about the delay, but merchandising is preparing a promotion and stores are allocating display space.
The missed handoff is the connection between supply disruption and commercial commitment. Knowing that a purchase order is late does not establish which promotion, stores, or customer promises are exposed.
A Jupiter application could combine purchase orders, expected receipts, inventory, promotion schedules, and supplier information. Process Tempo describes multi-source dashboards, supplier relationship mapping, and forms and workflows for responding to supply chain exceptions. [5]
With an appropriate graph model, Voyager could help users explore the relationships between a supplier, affected products, distribution centers, stores, and promotions. A saved graph snapshot could preserve the dependency view used during a review; current operational data would still need to be checked before execution.
The merchandising team could initiate a recovery request from that context. Procurement evaluates an alternative source. Logistics assesses an expedited shipment. Marketing reviews whether the promotion needs to change. Finance approves additional cost where required.
Each decision becomes part of the same case. Approved changes pass to the relevant systems through controlled integrations, with confirmations recorded against the recovery plan.
The retailer can then measure detection-to-decision time, overdue approvals, and incremental recovery cost. Linking these measures to promotion availability helps establish whether the process improves commercial outcomes.
A customer returns an online purchase to a store. The product was supplied by a marketplace partner. The store accepts it, a logistics provider moves it, and finance needs evidence before the refund or partner settlement can proceed.
A spreadsheet may become the only place connecting the return authorization, carrier reference, partner correspondence, and unresolved payment.
A Jupiter application could organize these records around a single return case. Store staff record acceptance. Operations reviews movement and inspection evidence. Customer service sees the current resolution status. Finance receives the evidence and approval needed for its next action.
Partner information should enter through the retailer’s established exchange mechanisms. Where relevant, GS1’s EPCIS standard offers a common approach to sharing visibility events within and across enterprises. It can help describe what happened to a product as it moved between parties; using that information in Jupiter would require an appropriate integration. [6]
The workflow should distinguish between “refund approved” and “refund confirmed.” That distinction prevents an internal task completion from being mistaken for a completed customer outcome.
Measures could include refund cycle time, cases awaiting partner evidence, manual touches per return, and unresolved partner settlement value.
These applications need clear architectural boundaries to remain manageable as adoption grows.
Preserve authoritative ownership. Define which system owns each business object and which application may change it. Store the case and decision evidence without creating an uncontrolled duplicate of the order, inventory, or financial ledger.
Define shared identifiers and meanings. Agree how orders, products, locations, suppliers, and partners are matched across systems. Establish definitions for statuses such as available, received, approved, and complete.
Match freshness to the decision. A customer recovery decision may need current availability; a supplier review may tolerate less frequent updates. Display timestamps and distinguish missing data from a confirmed business state. Real-time customer experiences depend on source freshness, integration latency, and response time across the whole process.
Make cross-system actions recoverable. Design integrations to handle retries, duplicate requests, partial failures, and unavailable partners. Require confirmation from the authoritative system before marking a transaction complete.
Apply access controls to both information and action. A store associate, regional manager, procurement specialist, and external partner need different permissions. Approval authority should be explicit, and source-system permissions should remain enforced.
Give each application a lifecycle. Assign an owner, support arrangements, change controls, and retirement criteria. Reuse shared connections and patterns, and retire superseded spreadsheets and scripts after the replacement is proven.
These are implementation requirements for the overall solution. They should be assessed across Jupiter, the integration services, and the connected applications.
Start with one process where the gap is visible and the outcome can be measured. A strong candidate has multiple participating systems, frequent manual handoffs, a clear business owner, and a consequential exception.
Baseline the time employees spend finding information, obtaining approval, and confirming execution. After implementation, measure those same activities alongside the customer or commercial result.
For pickup recovery, that result may be fewer unresolved orders at collection time. For supplier disruption, it may be earlier decisions on threatened promotions. For returns, it may be a shorter interval between receipt and confirmed refund.
A connected retail enterprise makes the path from an event to a completed response visible. Jupiter gives enterprise architects a way to build applications around that path, while preserving the responsibilities of the systems already running the business. It also creates a stronger foundation for AI: shared business context, defined authority, and a record of decisions and outcomes that can guide improvement.
The starting point is a process your teams currently hold together with spreadsheets, email, and manual coordination. Map its gaps, define its outcome, and use Jupiter to give that work a governed home.
1. Microsoft Learn — Dynamics 365 Commerce architecture overview
2. Process Tempo — Jupiter platform
3. Process Tempo — Workflow modernization
4. Process Tempo — Spreadsheet replacement
5. Process Tempo — Supply chain analytics
7. Microsoft Learn — Agents, Copilot, and AI capabilities in Dynamics 365 apps