AI Block Conversion — This section shows how effectively AI blocks are working within your business processes. You can see how customers progress through each block: how many of them reach their goals, where they get lost, and at what stages manager intervention is required.

Metrics help evaluate the actual effectiveness of AI logic, identify bottlenecks, and understand which scenarios require refinement. Analytics are built for each block separately and take into account both automated AI actions and customer behavior during dialogue.

Key statistics columns

• BP is the business process to which this block belongs.
• Entered - shows how many times the client entered this block during the selected period.
• Completed - the number of times the block was successfully completed.

For an AI Agent, the agent independently determined that the goal was achieved or a postback was received.

For the AI Response, the business process has moved forward - the AI condition within the block has been triggered (or there were no conditions, and the transition to the next step has occurred).

• Abandoned - The number of times the interaction was interrupted or the process stopped.

For AI Agent - the deal is archived or the client has not responded for more than 25 hours (one hour after the automatic reminder).

For AI Response - the process has not progressed further within 48 hours (the client has not responded, or responded in a way that the AI Condition has not been triggered).

• Takeover - the number of times the manager took over the conversation before the agent completed the task. This does not count as a successful completion.
• In progress - the number of sessions that have not yet been completed.

Formula:
In progress = incoming − (completed + abandoned + takeover)

• Conversion is the proportion of successful sessions.

Formula:
Conversion = completed ÷ entered

In this case, only closed sessions are taken into account (excluding “in progress”, “abandoned” and “intercepted”), which reflects the real effectiveness among those who achieved the result.

The “ Evaluate Conversions ” button allows you to compare current versions of prompts, identify regressions, and find weak points in scripts.

Grouping and filtering statistics

Grouping and filtering allow for flexible analysis of AI block statistics: identifying relevant data segments, comparing the effectiveness of different business processes, blocks, and their versions, and quickly identifying problematic stages and growth points.

Statistics grouping

Available grouping options:

• By BP - displays a list of business processes.
• By Block - shows a list of AI blocks (AI-Agent and AI-Response) within the selected business processes.
• By Version – grouping by block version (prompt hash – short identifier). When an instruction is changed in a business process, a new version is created, and statistics for different versions are not merged.
• By Date - grouping by the calendar day on which the events within the block occurred.

Filtering statistics

Filters allow you to select data according to the required criteria:

• Period - specifies the time interval for analysis.
• Business process - filtering by selected business processes.
• Block - filtering by specific AI blocks (AI-Agent and AI-Response).
• Version - filtering by block prompt versions.

After changing the filters, click the Refresh button to recalculate the data in the table.

Block analysis and prompt update

To get an analysis for a specific block:

1. Select the period and business process .
2. Set grouping by BP and by block .
3. Click the Update button.
4. In the table, click "Open dialogs in block" .

In the window that opens, you will see examples of dialogues:

• successful,
• abandoned,
• intercepted,
• as well as dialogues that are still in the works.

This allows you to quickly understand how the block behaves in real-world scenarios.

When you click the "Analyze block" button, the system:

• will identify the main problems of the current proposal,
• will offer a new prompt option,
• compare the current and proposed versions,
• will describe the key changes, the expected effect and possible risks.
• shows the top phrases sent before a client leaves (the agent’s last messages in abandoned sessions).

After this, you can decide whether to upgrade.

If you confirm the update, the new prompt will be automatically applied to the business process. When grouping by version, the new version of the block will appear in the conversion section, allowing you to compare "before" and "after" metrics and evaluate the effectiveness of the changes.

Block metrics (Median and P95 for dialogue length in messages and time to completion for each outcome):

• Median Messages — the 50th percentile of the number of messages per session (agent + client). Half of the sessions are shorter, and half are longer. Calculated using completed sessions for the selected outcome. Sessions longer than 1000 messages are capped at 1000 (safeguard limit).

• Median Time — the 50th percentile of the time from entering the block to its completion. Measured in wall-clock time, not machine processing time.

• For “Abandoned” sessions, the value is always greater than or equal to the waiting threshold (24–48 hours) — the system intentionally waits before marking a session as unsuccessful.

• P95 Messages — the 95th percentile: 95% of sessions are shorter than this value, while 5% are longer. Highlights “tail-end” cases — the longest conversations.

• P95 Time — the 95th percentile of time to completion. These are the “long-running” sessions. Useful to compare with the median: if the P95 value is several times higher, it indicates a long tail worth investigating.

Conclusion

The "AI Block Conversion" section is a tool for systemic improvement of business processes using AI. It allows you to not only track metrics, but also understand customer behavior, identify weaknesses in your processes, and make informed decisions on optimization.

Using groupings, filters, and dialogue analysis, you can pinpoint which blocks are working effectively and which require refinement. Built-in AI analysis and prompt updates help you quickly implement improvements and validate their impact.

Regularly working with this section allows you to gradually increase conversion, reduce the number of abandoned conversations, and make customer interactions more precise and effective.