Separate saved-on-device work from accepted server transactions, and define retries, conflicts and stock promises before building a field app.
A salesperson visits a retailer, enters an order and sees “saved”. The connection is weak. Later, the salesperson tries again. At the warehouse, staff need to know whether there is one order, two orders or no accepted order at all.
An offline-capable field sales app needs clear business states as well as local storage. Saving something on a phone is different from the server accepting it, reserving stock or approving a price.
This article proposes a planning model for a hypothetical Nepal distributor. It is not a description of an existing WorkSphere mobile app or a guarantee that a particular sync design is appropriate for every business.
Decide what must work without a connection
List the work that should remain possible offline. It might include viewing a previously downloaded customer list, reading product information or preparing an order draft. Separately identify actions that require current server checks, such as confirming live availability or approving an exceptional discount.
Android’s offline-first architecture guidance explains the role of local and network data sources and making critical reads possible without network access. The practical business adaptation is to define what the local information may support and how its age is shown.
Display the last successful refresh and avoid labelling cached quantities as guaranteed availability. An app can help capture a request during poor connectivity without promising stock that another salesperson has already sold.
Suggested offline states; current stock and price promises need their approved authority.
Hypothetical field visit: capture a request without assuming server acceptance.
Name states that staff and customers understand
Use clear states such as draft, queued for sync, awaiting server result, accepted, rejected and needs review. Decide what the salesperson may tell the retailer in each state. “Saved on this device” is more informative than a generic success message when no server confirmation exists.
Retain the customer, item, quantity, units, agreed context and the local record reference. If a draft becomes a submitted request, preserve the relationship rather than losing its history in a new unexplained record.
| State |
Meaning for the salesperson |
| Draft |
Editable work saved locally |
| Queued |
Waiting for a submission attempt |
| Awaiting result |
Acceptance is not yet known |
| Accepted |
Server confirmation is recorded |
| Needs review |
A person must resolve a business or technical exception |
Conceptual artwork: reconnecting the device is only part of confirming an order.
Design retries around one intended request
In a hypothetical test, the phone sends order request M-204. The server accepts it, but the response never reaches the device. A retry should refer to the same intended request so the system can return or reconcile the earlier result instead of creating a second business order.
Conceptual mobile-sync panel separates the phone’s unknown outcome from the server’s accepted request in the hypothetical failure test. It is not a working app screenshot.
A stable request identifier and server-side duplicate handling are suggested design requirements here. They need careful implementation, storage rules and testing. A button labelled “sync” does not establish those behaviours.
Test the uncomfortable moments: a connection drops during submission, the app closes, the device restarts or the user presses retry several times. The user should see a recoverable status and a path to resolve uncertainty.
Hypothetical failure case; duplicate handling is an implementation requirement to test.
Keep stock and price decisions authoritative
Suppose two salespeople each request seven units while the server has ten available under the same allocation rule. Fourteen requested units cannot become fourteen guaranteed units merely because both phones saved their drafts offline.
The business must choose how the server handles the conflict: reject a request, accept a reviewed partial quantity or route it for an authorised decision. Record what was actually accepted. If the retailer must approve a change, include that step rather than silently substituting a smaller order.
The same principle applies to expired prices, credit restrictions and inactive products. Define which version the request used and which rules apply when it reaches the server. Do not let the phone invent approval by applying whichever rule is convenient.
Shared stock needs one acceptance policy; 14 requested does not mean 14 available.
Provide a visible exception queue
Give sales and back-office staff a shared way to identify requests that need attention. Include the intended request reference, last known result, error category, responsible person and next action. Keep technical diagnostics available to support without exposing confusing implementation detail to the retailer.
Restrict device data to what the user needs. Plan access removal, device loss, account changes and retention. A local copy can remain on a device unless the design deals with that possibility. Confirm the actual privacy and security requirements rather than relying on an offline label.
Hypothetical reconciliation: sales and dispatch staff resolve an uncertain submission together.
Pilot with realistic conditions
Use a small approved customer and product set, sample transactions and a defined reconciliation process. Compare accepted server orders with the intended requests, not only the number of green indicators on phones.
Include CRM customer references, Inventory availability, Accounting approvals and Task-based exceptions in the planned integration discussion. Each module needs to agree which event it accepts and what it must never post twice. The app should not independently create payroll, stock or financial effects simply because a draft was saved.