Financial Services & Advisory
See the cost behind every research report.
A single research request can reach multiple enterprise MCP servers and trigger more model calls. Connect the recorded steps to one report or audit execution to understand what it costs to serve the customer.
Where costs compound
One request. Several paid steps.
External tools may have their own invocation charges. Their results can also generate additional model input, follow-up queries, and retries. Looking only at the initial model call misses the recorded cost of the wider workflow.
Link
Keep the report under one execution ID
Assign an opaque execution ID when the research or audit job begins. Use that same ID for each model call and MCP step, with unique step IDs and parent-step links to preserve the reported sequence.
Propagate IDs through asynchronous jobs and tool handlers. Ganivra links the events you send; it cannot discover uninstrumented downstream calls.
Capture
Include direct enterprise tool costs
Send each MCP server and tool identifier, latency, outcome, and actual invocation cost when known. Record downstream model calls separately with their real token usage.
Report the tool's own charge, not a total that already includes the separately reported model calls. Keep unknown charges unpriced rather than silently treating them as free.
Inspect
Follow expensive tool paths
Filter Cost breakdowns by MCP server or tool to inspect linked execution costs. Open an execution to review individual steps, direct charges, and the model calls connected to them.
A server or tool filter can include the wider execution. Inspect parent links to distinguish the tool's direct cost from other work in the same report.
Attribute
Connect report cost to customer economics
Add a pseudonymous customer ID and a workflow such as investment research or audit evidence review. Connect customer pricing or execution-linked revenue to compare tracked cost with what the report earns.
Human analyst time, data subscriptions, and general infrastructure are not automatically captured. Treat the result as tracked AI and tool cost-to-serve until your required costs are included.
Illustrative execution
The lookup costs $0.10. The tracked workflow costs $1.20.
| Step | Parent | Recorded cost |
|---|---|---|
| Plan research · model | Root | $0.10 |
| Market-data lookup · MCP | Plan research | $0.10 |
| Analyze findings · model | Market-data lookup | $0.45 |
| Evidence lookup · MCP | Analyze findings | $0.15 |
| Validate and draft · model | Evidence lookup | $0.40 |
These invented amounts demonstrate attribution, not vendor prices. At $2.00 of linked report revenue, the execution leaves $0.80 after tracked AI and tool costs, or 40% contribution margin before other costs.
What to send
Link the workflow with IDs, not its financial content.
execution_idUse a stable, opaque job identifier across the whole workflow. Add customer, application, and workflow labels for attribution.
step_id + parent_step_idGive each actual call a unique step and event ID. Point to the step that triggered it; reuse the event ID only for telemetry delivery retries.
tokens + MCP costSend model usage for pricing and known tool invocation charges. Never send the research query, account numbers, tool arguments, or report content.
Use the breakdown to choose your next change
Investigate repeated lookups, redundant validation, or expensive model steps. Test a change in your application and compare equivalent reports alongside your own evidence-quality and completeness checks.
Start with one research workflow
Trace the report from first query to final model call.
Verify the captured steps, fill pricing gaps, and connect revenue to understand the margin behind delivery.