Best Practices & Troubleshooting

Best practices

Practice	Why
Name metrics descriptively	The analysis reasons about the metric by its name. `Conversion Rate` guides it; `metric_2` does not.
Keep endpoint parameters minimal	Send only what is needed to fetch the value. Fewer parameters mean fewer failure modes and faster runs.
Test the cURL before saving	The generated Dummy cURL is the exact request Callkaro will send. Confirming it works now avoids an empty report tomorrow.
Be specific in the RCA prompt	Scope, exclusions and priority are what separate a useful report from a generic one.
Exclude known artefacts explicitly	Telling it to ignore sub-5-second disconnects stops the analysis spending its attention on dialer noise.
Set targets you have actually achieved	An unrealistic target produces daily reports about a gap that has no fix — and the analysis will eventually tell you so.
Return a diagnostic string from Python sources	A short message alongside the number gives the analysis context it cannot get from the number alone.
Work the board, not just the reports	Moving cards through the columns is what lets the next report tell you whether the fix worked.
Read before you assign	Every fix is reviewable for a reason. Check the evidence and the prompt excerpt before sending anything to AI FDE.

Troubleshooting

No report was generated for a day

Either every metric met its target, the shortfall was under 1 percentage point, the RCA prompt did not cover the metric that missed, or Auto RCA is off. Open the day-wise timeline — it states which of these applied.

A metric shows as "could not be measured"

No value came back from its data source for that day. Check the timeline entry for the endpoint or Python source, and verify the Source path on the metric matches the field the source actually returns.

The metric value is always empty

Almost always a referencing mistake. API sources need the extra .data segment; Python sources do not. Compare your Source field against the response shape.

An endpoint was "partly delivered"

The batch was split and some parts failed. The timeline shows how many calls were processed and how many were not, along with the status codes returned.

"Save the agent first" appears

Data sources are configured against a saved agent. Save the agent, then return to the KPI Manager tab.

The report says data was insufficient

There were too few audited calls that day to conclude anything reliably. This is stated deliberately rather than guessed around. Widen the scope in your RCA prompt, or wait for a higher-volume day.

The same cause keeps appearing

It will be flagged as recurring with its first-observed date. Check the Pending Tasks board — the fix may never have been assigned, or may have been rolled out to only part of your traffic.

I changed the config but the report looks the same

Changes apply to the next run, not to reports already generated. Use Run Manual RCA if you need to see the effect immediately.

I want analysis on one metric only

Say so in the RCA prompt. Other metrics stay measured and charted, but will not trigger an analysis.

Defaults and limits

Setting	Default
Automatic run time	5:00 AM IST daily, covering the previous calendar day
Shortfall required to trigger an analysis	1.0 percentage point below target
Auto RCA	On for every agent
Default RCA model	GPT-5.4
RCA prompt maximum length	10,000 characters
Built-in metric	`connected_to_conversion_percentage`, calculated by the platform
Dashboard default date range	The last 7 days
Recurrence history window	Recent reports from the past 14 days
Automatic handoff of fixes to AI FDE	Off — fixes are assigned by a person
Report email	Sent to the agent owner whenever a report is generated

Access

Action	Permission required
View the KPI Manager Dashboard	Analytics — View
View an agent’s KPI Manager configuration	Agents — View
Save configuration or run a manual RCA	Agents — Create
Delete an agent’s KPI Manager configuration	Agents — Delete

In-product help. The Guide button in the KPI Manager header opens a short quick-start reference inside the product, covering the same setup steps in condensed form.

Getting started

Pricing

Voice AI Agents

Chat Agents

AI FDE

AI Auditor

KPI Manager

Phone Numbers

Calls

Webhooks

Website Widget

Whatsapp (Facebook)

Integrations

Best Practices & Troubleshooting

Best practices

Troubleshooting

Defaults and limits

Access

​Best practices

​Troubleshooting

​Defaults and limits

​Access

Best practices

Troubleshooting

Defaults and limits

Access