Copy the complete prompt below into your coding agent inside your existing project. It adapts to your stack and business model. A membership app, marketplace, internal tool and SaaS should not all get the same metrics.
Inspired by the Agent Ops Club admin implementation: signup sources, member activity, free-to-paid activity, learning progress, downloads and labeled revenue estimates. These are examples of useful questions, not a promise that your project already collects the required data.
Act as a senior product engineer, analytics engineer and admin UX designer. Build a production-quality admin insights and operations dashboard for THIS application. Work in the existing repository and deliver functioning software, not only a plan or a mockup.
- Understand the product before building.
Inspect the project instructions, framework, routes, authentication, authorization, database schema, event instrumentation, jobs, integrations, billing model and design system. Identify the administrator's main decisions, core entities, user journey and definition of successful usage. Reuse the existing architecture and components. Do not replace the stack or introduce a new analytics vendor unnecessarily. Check existing admin screens before adding duplicate functionality. Ask only for missing decisions that materially block implementation; otherwise state reasonable assumptions and proceed.
- Design the information around decisions.
Create a concise question-to-metric map: business question, metric definition, data source, time window, limitation and action the administrator can take. Select the 4 to 6 most useful headline KPIs for this product. Examples include activation, meaningful active users, completed jobs, fulfillment, retention, revenue, support backlog or failed runs. Omit irrelevant concepts: a noncommercial internal tool does not need MRR. Every number should help answer a question or trigger a useful action.
- Build the overview and drill-downs.
Provide a clear navigation structure with Overview, relevant entity management, Activity, Health and Settings. Adapt the labels to the app. Start with a concise status summary, key changes and a prioritized needs-attention list. Add date presets and custom ranges, comparison with an equivalent prior period, timezone clarity, meaningful segment filters and a visible data freshness timestamp. Preserve filters in the URL where appropriate.
Use readable trend lines for change over time, ranked bars for comparisons, and funnels only when the data supports a defined ordered journey. Include units, tooltips, denominators, absolute counts beside percentages and accessible table alternatives. Link cards and charts to filtered records or detail drawers so users can investigate the underlying evidence. Keep small screens usable without turning the dashboard into a wall of miniature charts.
- Make the analytics trustworthy.
Compute real metrics on the server from authorized data. Define what counts as active, activated, retained and converted. Distinguish unique people from events and accounts from organizations. Handle duplicate events, deleted users, bots/test accounts, nulls, currency, refunds and subscription states as applicable. Distinguish revenue, cash collected, profit, MRR and estimates. Never call an estimate actual revenue.
For funnels, specify cohort entry, step ordering, conversion window and denominator. Independent all-time event counts are activity coverage, not a verified conversion funnel. For retention, use eligible cohorts and do not compare incomplete cohorts as though mature. For zero prior-period values, show the absolute change and an honest label rather than infinite growth. Expose incomplete tracking, row limits, sampling, stale data and unavailable integrations. Missing data is not zero. If instrumentation is missing, add a small documented event contract and the required instrumentation where feasible, with stable IDs and deduplication. State when useful history starts. Do not fabricate historical events. Keep synthetic fixtures confined to tests or a clearly labeled demo mode that cannot be mistaken for production.
- Add insights that earn their space.
Generate evidence-backed observations such as a change in activation, an unusual increase in failures, concentrated feature usage, a backlog or a drop between onboarding steps. Each insight must show the observed change, timeframe, supporting counts, why it may matter and a specific investigation or next action. Use minimum sample sizes and sensible thresholds. State uncertainty and separate observation from hypothesis; do not invent causes or guaranteed outcomes. Prefer deterministic rules initially. An optional AI summary may summarize authorized aggregates, but must not invent statistics or send personal data to a new provider without an approved integration.
- Add convenient, working controls.
Provide searchable, sortable, paginated entity tables with useful filters, saved views when justified, detail drawers and safe CSV exports. Include the relevant app controls: account or membership status, content visibility, moderation, feature flags, quotas, notification settings, integration health or job retry. Implement only controls supported by this application. Do not display decorative buttons or switches that do nothing.
Validate settings on the server and show saving, success and failure states. Explain scope and consequences before impactful changes. Require confirmation for destructive or bulk actions, and provide undo where feasible. Make retries idempotent and protect against concurrent edits. Record who changed what and when in an audit log, without logging secrets. Mask stored credentials and never return secret values to the client.
- Enforce permissions at every boundary.
Use the existing authentication and least-privilege role system. Check authorization on every admin page, API, server action, query and export. Enforce tenant boundaries in the data layer. A hidden navigation link is not access control. Use appropriate CSRF protections, input validation and rate limits for the framework. Protect sensitive fields, prevent CSV formula injection, and log access to sensitive exports where appropriate. Do not expose service-role keys or privileged queries in browser code.
- Make it polished and fast.
Match the app's visual language with intentional typography, spacing and a restrained semantic palette. Use status labels as well as color, sufficient contrast, keyboard navigation, visible focus and accessible chart descriptions. Include skeletons, genuine empty states with useful next steps, partial-error recovery and stale-data states. Use bounded queries, server-side filtering and pagination, appropriate indexes and aggregation, and documented caching freshness. Avoid N+1 queries and fetching the entire database into the browser. Keep current data visible during background refresh when safe.
- Verify the behavior, then hand it over.
Test representative metric calculations with known fixtures, time boundaries, duplicates, empty datasets and zero comparison periods. Verify anonymous, ordinary-user, admin and cross-tenant access for reads and writes. Test one full drill-down, a settings update, a failed mutation, pagination and a safe export. Run the project's relevant checks and inspect the interface at desktop and mobile sizes. Fix errors rather than hiding them.
Deliver the implementation, necessary migrations, instrumentation, meaningful tests and a concise handoff. Explain metric definitions, environment-variable NAMES only, setup steps, data limitations, verified behavior and remaining gaps. Do not claim deployment or production readiness unless verified. If an external credential blocks a module, finish the unblocked work and show a clear unavailable state with the exact setup required. Never substitute fake production data.
Begin by inspecting this project, show a short tailored implementation plan, then build and verify it. Stop for approval before destructive migrations or production deployment unless I have already authorized them.
The useful test: can you spot a problem, inspect the evidence and take an appropriate action from the same admin area?
This prompt gives you one useful upgrade. Inside Agent Ops Club, I teach you how to build and deploy AI agents, put them to work, and turn software into something you can offer clients.
| What you want to do | What you get inside the Club |
|---|---|
| Put AI to work in your business | AI Employee roles for recurring work, with training to get them running and keep them reliable. |
| Build and deploy with a clear process | The Agent Ops Masterclass and live training to help you move from an idea to a working system. |
| Start your next project with more already built | A premium library of systems, apps, skill kits and SaaS boilerplates you can adapt. |
| Create an offer for your clients | Software with resale licensing, plus sales tools to help shape and present your offer. |
Bring a real project:
- Automate a recurring job in your own business.
- Adapt a useful app for a client's workflow.
- Build a software or agent service you can demonstrate and sell.
You can start with the free walkthrough and introductory Masterclass modules, then explore the full training and software library.
See the AI Employees, browse the software, and find your next build.
<sub>This prompt is free to use in your projects. Training, downloads and resale rights vary by membership and product license.</sub>