Google AppSheet

What AppSheet is

AppSheet is Google’s platform for building applications and automations without conventional programming. It combines structured data, configurable interfaces and process rules. It integrates with Google Workspace and external sources, extending beyond turning a Google spreadsheet into a form.

Its declarative approach lets creators specify data, conditions and intended behaviour. Process design is still required. An application may be easy to build and difficult to maintain if it mixes master data, requests and outcomes without defining what each record represents.

Evaluate a proposal by identifying the interaction to improve: consulting current information, capturing an observation or requesting materials. Then decide how it is presented, who may act and what evidence establishes completion.

Reference documentation: AppSheet product overview.

Founding, investment and Google acquisition

Praveen Seshadri and Brian Sabino founded AppSheet in 2014. NEA records its first, seed-stage investment in 2015. The project addressed a common need: using existing business data to create mobile applications without commissioning every interface from a programming team.

In May 2018 AppSheet announced Spec, a feature for describing applications in natural language. Google announced the acquisition on 14 January 2020 and its incorporation into Google Cloud. Current assisted creation belongs in this development, without treating Spec as equivalent to today’s Gemini features.

The history explains the product’s close relationship with spreadsheets and tables and its expansion into automation and enterprise administration. Before following an older tutorial, check the editor, licence and specific capability: names and screenshots may have changed.

Reference documentation: AppSheet announces Spec, May 2018 · NEA investment record: AppSheet · Google acquires AppSheet, January 2020.

Data sources and native databases

AppSheet supports Google Sheets, Excel, SQL databases and services such as Salesforce, alongside its own databases. An app can combine tables from different sources. Documentation distinguishes available connections from Integration Connectors in preview; a table added from Google Forms is read-only.

Native AppSheet databases organise data within the service without requiring an external spreadsheet. Storage decisions should consider volume, relationships, ownership and other applications needing the same information.

For a materials catalogue, separate resource descriptions from requests that reference them. Correcting a title should not erase the record of what was requested. A recognised source for each data item reduces discrepancies between supporting apps and the operational record.

Reference documentation: Manage data sources · What is an AppSheet database?.

Record keys and relationships

A key uniquely identifies a row and must remain stable. AppSheet discourages row numbers as identifiers because moving or deleting records can change them. Duplicate keys can cause updates to affect the wrong row.

This matters when an informal spreadsheet becomes a shared application. A site’s name may be corrected, but its identity should not depend on spelling. Requests should reference that identity while displaying a readable label.

Before importing data, check duplicates, empty values and broken relationships. A useful exercise is renaming a test material and verifying that previous requests still reference the same resource. A polished interface cannot compensate for ambiguous identification.

Reference documentation: What is a key?.

Views, forms and mobile capture

Available views include tables, details, forms, calendars, maps, charts and dashboards combining several views. Forms can capture text, photographs, location and signatures. NFC capture requires a compatible mobile device; its scan button does not appear in a browser.

Select the view around the user’s question. Maps locate sites, tables support comparison and forms guide requests. A screen attempting to display everything may hinder a short field task.

In an OSH campaign, a photograph should have an explained purpose and recognisable context. An isolated image does not establish when an issue was resolved or who validated the outcome. Similarly, a drawn signature does not by itself establish identity or understanding of a document.

Reference documentation: View types · Form view type · Use NFC scanning.

Expressions, calculations and virtual columns

Expressions retrieve values and references across tables. A virtual column calculates a result through a formula without corresponding to a physical source column. Many virtual columns or complex expressions can impose a performance cost.

Distinguish captured facts from calculated interpretations. A request date is a fact; a label saying it has been pending for several days depends on a rule and the viewing time. Preserving that distinction explains why a view can change without editing the request.

Test rules with ordinary, empty and boundary cases. Agree units and decimal precision for quantities. Decide how dates behave across days and time zones. Document those decisions through understandable examples rather than hiding them in a lengthy formula.

Reference documentation: Column value expressions · Use virtual columns.

Bots, processes and execution monitoring

A bot connects an event to a process and its tasks. It can update records, issue notifications or call a service through a webhook. Monitoring exposes executions, errors and activity; some functions do not run fully on a free account even when they can be configured.

Automation design requires deciding what constitutes a relevant change. If a request is corrected three times, should it produce three notifications or remain one conversation? Messages should include an identifier and link to the authorised record to avoid contradictory copies.

A test should follow the complete route: event received, condition evaluated, action executed and destination result verified. Sending a notification does not establish that someone read it. Tasks needing a response require ownership, a deadline and a way to record resolution.

Reference documentation: Bots: The Essentials · Monitor your apps.

Gemini: app creation and AI in processes

Gemini for App Creation proposes tables, columns and an application from a description and is included in paid AppSheet licences. Gemini in AppSheet Solutions adds extraction or classification tasks to automations with Enterprise Plus and credit consumption. These are distinct AI uses.

The initial description should identify the task, required data, participants and intended outcome. Suggested tables can provide a draft, but do not establish suitable structure or correct permissions.

When extracting information, preserve the source document and verify critical fields before use. In OSH, automatic classification may support later review; it should not become a risk assessment or work authorisation without validation. Recording corrections helps identify recurring process errors.

Reference documentation: Use Gemini in AppSheet.

Identity, filters and application governance

AppSheet distinguishes authentication, app access, data access and auditing. Security filters restrict downloaded records; a slice selects among records after download. Hiding a view does not replace information-access policy.

Organisation policies can restrict external sources, AppSheet databases, APIs or AI features. Configuration should reflect context: a public materials catalogue differs from an internal application containing personal information.

Test at least three roles: administrator, authorised user and user without access. Check both visibility and permitted changes. Offboarding should address devices, accounts and shared applications. The person maintaining content need not hold every technical permission.

Reference documentation: Security: The Essentials · Scale using security filters · Predefined governance policy templates.

Performance, versions and deployment

App capacity depends on the data source as well as AppSheet infrastructure. For a major upgrade, documentation proposes working on a copy, testing it and running deployment checks before replacing the version in use.

Measure representative volumes: time before consultation is possible, saving delays and sync duration. A ten-row demonstration cannot establish how a catalogue with years of requests will behave. Include phones and connections resembling actual field use.

Document responsibility for tables, rules and access. Before withdrawing a version, establish who still uses it and how required records will be preserved. Continuity also depends on another person understanding the design and resolving ordinary incidents.

Reference documentation: App performance: Core concepts · Upgrade an app.

Licensing, users and maintenance costs

The catalogue includes Starter, Core, Enterprise Plus and Publisher Pro for public access. Some Workspace editions include Core. User Pass uses a licence pool: an eligible user consumes one for each app accessed and retains access to that app for the month.

Calculate scope using actual users, including external participants, and verify the application’s required tier. A free prototype does not establish that all its automations work in production or that distribution has no cost.

Operating costs also include design, support, permission review, data retention and adaptation to change. Compare that effort with maintaining a simple form or using an existing capability. Initial speed is valuable when it leads to a sustainable solution.

Reference documentation: How to choose a subscription · Allow app access with AppSheet User Pass.

Documented adoption and interpreting cases

Google’s Globe Telecom case describes an internal creation programme growing from 59 apps during a hackathon to 350, with uses including station maintenance and auditing. This is the supplier’s account of that deployment, not an estimate transferable to every organisation.

To assess a case, ask which task was replaced, which data already existed, how much support the team received and how outcomes were measured. Time reduction alone does not describe quality, stability or user acceptance.

For your own pilot, choose a frequent, bounded interaction. Measure corrections, incomplete requests and time to resolution alongside capture time. Compare the previous process using consistent definitions and preserve examples that explain the results.

Reference documentation: Globe Telecom customer case.

Applications in occupational safety and health

In occupational safety work, it can support straightforward activities such as consulting approved resources or collecting requests for materials by site. Its usefulness depends on a clearly bounded task and a well-defined information source, avoiding multiple applications that create competing versions of the same preventive record or obscure which version is authoritative.

One scenario proposed by Sabentis is consulting approved materials at a site with intermittent connectivity and recording requests for later review. Another is collecting language and format preferences for a campaign. It can also display current resources by site. In these cases, distinguish a request’s local status from the status confirmed by the receiving team, avoiding expectations of immediate handling.

Worked example and checks

For example, an application could display the current catalog of campaign materials and accept requests for copies in a particular language. Requests would be reviewed before fulfillment and linked to the relevant material and site.

1. Prepare a catalog with identifiers, titles, languages, versions and current status; link requests to those identifiers.
2. Configure who can consult or modify the catalog, verifying access with test users.
3. Open the app online and check that data and, where needed, files are downloaded.
4. Disconnect the device, create a request and verify that it is clearly awaiting synchronization.
5. Reconnect and check the source record, its identifier and absence of duplicates.

Then modify a material at the source and check when the device updates. Also test two users acting on the same record. The acceptable result is knowing which version was consulted and whether the request arrived; a form with a working save button is insufficient. Use fictitious data and keep trial requests separate from real orders. Record what happens if the user closes the application before synchronization completes.

Limitations and conditions of use

No-code development does not eliminate configuration risks. Test permissions, synchronization and data rules, especially for mobile use; a quickly created application should not be assumed reliable enough to authorize hazardous work or record clinical decisions.

Offline use requires an initial online launch and file-cache configuration. Local saving and completed synchronisation are different outcomes. Prepare and test the device before field use.

Reference documentation: Offline and Sync: The Essentials.

Choosing and complementing the tool

Choose AppSheet when a structured-data lookup or capture interface is needed and there are owners able to maintain its rules. Google Sheets can serve as a source for a bounded process with editing controls; Power Apps is another catalog option where the Microsoft ecosystem predominates. Compare identity, sources and synchronization, rather than creation speed alone.

Before expanding use, define support, recovery and withdrawal of outdated versions. An auxiliary app needs a clear owner and recognized source; separate copies maintained by each site can hide content differences behind a convenient form. Define which status users can rely on and how to report a request that remains pending.

Frequently asked questions

Does saving offline mean a request has been delivered?

No. A change may remain only on the device until synchronization completes.

Does a slice protect confidential information?

Not by itself. It filters the presentation of already downloaded data; security controls require a different design.

Does viewing a PDF mean it will be available offline?

No. Configure compatible caching and test availability after actually disconnecting the device.