Choose the work to replace
A coordinator opens a dashboard, filters late orders, exports the list and assigns internal follow-ups. The migration candidate is the whole job: querying data, explaining an exception and preparing the next task. List frequency, users, sources, filters and the decision made after each report.
Recurring inventory queries, procurement issues and at-risk deliveries are useful evaluation scopes. Free-form visual exploration, official reporting and highly interactive analyses may still require BI tools. Decide by use case; removing the whole platform before measuring equivalence increases risk and can raise costs.
Compare paths that preserve the data investment
Evaluate three paths: retain the dashboard and add an agent for queries and actions; replace a reporting routine with a conversational workflow; or migrate the data and metrics layer too. Effort and risk grow when existing definitions and controls must be rebuilt.
Microsoft documents Fabric data agents answering questions over sources including Power BI semantic models. Natural-language querying is therefore not exclusive to Zild. Compare alternatives available in your environment. Evaluate Zild through its connection of data, people, tools and process actions, with integrations confirmed in the pilot.
Keep a contract for every metric
Define name, formula, unit, period, timezone, filters, granularity, refresh time and owner. “Late” might mean an overdue promised date, missing shipment or missing delivery confirmation. Without one definition, dashboard and agent can produce different yet plausible answers.
Synthetic example: there are 120 open orders, twenty overdue; five of those twenty are suspended under a commercial rule and excluded from the operational metric. The actionable queue contains fifteen orders, and the answer should explain the exclusions. Calculation belongs in a validated query or rule; the model presents results and context without inventing missing rows.
Preserve permissions in the new query path
Microsoft documentation explains that RLS restricts rows in semantic models and its enforcement depends on access roles; workspace Admin, Member and Contributor roles do not have the same restriction as Viewer. Testing only with an administrative account is therefore insufficient.
If an agent queries the database directly, report rules do not transfer automatically. Define effective identity, authorized filters and query limits in the backend. Test users from different sites and prohibited-data requests. Prompts do not replace technical authorization. Also confirm APIs and permissions actually available through the integration path; this guide does not assume a native Zild Power BI connector.
Benchmark from question through action
Build questions with approved answers, language variations, missing filters, stale data and empty results. Run dashboard and agent against the same data version and user identity. Reserve questions not used in configuration. Compare numbers, filters, sources, explanations and proposed actions.
Measure correct-answer rate, critical failures, time to decision, human review and cost per correctly completed task. For a late-order queue, also check whether the right owner received the right task and whether duplicates occurred. The first pilot can propose actions only; enable writes after the process owner approves rules and system confirmation.
Calculate savings using contracts that can actually end
For the current scenario, add licenses, capacity, support and effort for the evaluated routine. For agents, include the Zild subscription, APIs, integration, operations, review and retained tools. Add migration and coexistence. A license used for other reports cannot count as full savings if it remains necessary.
Synthetic BRL example: current cost of R$ 3,000 per month and future cost of R$ 2,200 including retained components. With R$ 9,600 implementation, recurring savings are R$ 800 and simple payback is twelve months. Over twelve months both scenarios total R$ 36,000; there are no accumulated cash savings yet. Freed hours represent separate capacity unless actual expenses fall. Use the calculator below to test your own assumptions.
Migrate by routine with a recovery criterion
Select one engineering, logistics, inventory or finance report and document the decisions it supports. In Zild, the design can combine Chat queries, Admin access and connectors, and Flow action tracking. Validate the source, identity and permitted operations before promising the experience.
Keep the previous path during evaluation. Define when to stop the new workflow: a critical numerical discrepancy, unauthorized access or missing action confirmation. Retire a license or routine only after demonstrating coverage, adoption and total cost. The business benefit is reducing effort between noticing an exception and completing controlled work.
Technical references
- Microsoft — Row-level security with Power BI Checked on
- Microsoft — Create a Fabric data agent Checked on
Build an assessment using your numbers
Gather this information to scope the work with Zild. During the conversation, we assess integrations, owners and how to measure outcomes.
- A financial process and its owners
- Current costs, volume and review time
- The correct outcome and approval limits
Describe the process in the form; samples and documents can be prepared later without confidential data.
The form opens with this guide’s title. You can edit everything before sending.