Conversation Efficiency Analysis — Fragmentation Index Report

Your engagement reports track two related but different signals for every conversation:

Turns (how many times the two sides actually went back and forth) and Messages

(the raw number of individual chat lines each side sent). On their own, these are just

numbers in a grid. The Conversation Efficiency Analysis report — and Velaro's AI

reporting layer — go further, reasoning about what the relationship between the two

numbers actually means for your team.

What "Fragmentation Index" Means

The Fragmentation Index is Messages ÷ Turns, computed per agent, per bot, and per

site each week:

• A high index means that party is splitting single replies into many separate

messages — sometimes intentional (a bot script that sends a multi-part answer by

design), sometimes a habit worth coaching (an agent typing several short messages

instead of one complete reply).

• A low index (closer to 1) means each side of the conversation tends to say what

it needs to say in a single message per turn.

Who Can See It

The Conversation Efficiency Analysis report is available on Professional and

Enterprise plans. Accounts without the report enabled won't see it in the reports

navigation.

How to Interpret Outlier Flags

Each row in the report — whether it's an agent, your bot, or the site-wide average —

can be flagged as an outlier. A row is marked as an outlier when its Fragmentation

Index sits meaningfully outside the normal range for its peer group that week (for

example, an agent whose index is well above the rest of the team). An outlier flag is

a prompt to look closer, not an automatic verdict:

• Outlier + high index often means that person or bot flow is splitting answers

more than the rest of the team — worth a quick listen-in or script review.

• Outlier + low index usually just means that party is unusually concise, which is

typically fine.

• No outlier flag means the row is within the normal range for that week — no

action needed.

Use the week-over-week change and trend views alongside the outlier flag —

a single high week is less concerning than an index that keeps climbing.

Comparing Your Bot to Your Agents

It's common for your bot's Fragmentation Index to look very different from your

agents' — a difference by itself isn't a problem to fix:

• Bot consistently higher than agents is usually intentional. A bot built around

a multi-step flow — a guided intake, a booking sequence, a checkout walkthrough —

will naturally send more messages per turn than a person typing free-form replies.

Treat this as expected behavior, not something to correct.

• Bot fragmentation that's inconsistent across similar conversations is the

pattern worth a closer look. If the same type of conversation (say, every product

question) produces very different Fragmentation Index values from one conversation

to the next, or swings week to week, that can point to a workflow issue — a branch

that isn't firing the way it was designed to, or a step that's unintentionally

repeating. Check the trend view for that conversation type before assuming

something's wrong, but don't ignore a pattern that keeps recurring.

Fragmentation and Conversion Outcomes

Beyond survey ratings, this report can also check fragmentation against what your

customers actually did, not just what they said afterward. If your store or CRM is

connected, the report automatically pulls in whether the conversation led to a

completed purchase or a won deal, and checks whether fragmented conversations

correlate with fewer of those outcomes.

For example: if conversations with a high Fragmentation Index are noticeably less

likely to end in a completed order or a closed deal than conversations with a low

index, that's a stronger signal to dig into than survey ratings alone. If there's no

meaningful difference, that's useful too — it means the extra messages aren't costing

you outcomes.

This layer only appears when your site has a connected store or CRM providing

outcome data. Sites without one still get the full Fragmentation Index and outlier

reporting — just without the conversion correlation.

Automatic Recommendations When a Pattern Is Costing You Sales

When one of your bots shows a fragmentation outlier AND conversations matching that

pattern are converting noticeably worse than the rest of your account, the report will

surface a specific, plain-language recommendation — not just a flag. For example: "this

bot's replies are splitting into more messages than usual, and those conversations are

converting less often — consider consolidating multi-message replies in this flow."

You stay fully in control:

• Nothing changes automatically. A recommendation only ever appears for your review —

it never modifies your bot on its own.

• You choose to apply it. Applying a recommendation adds a targeted instruction to

that bot's configuration, asking it to reply in fewer, more complete messages for that

pattern. Everything else about the bot stays exactly as you set it up.

• You can undo it anytime, with one click. Rolling back removes only the instruction

that recommendation added — nothing else about your bot's setup is touched.

• If the pattern comes back, you'll hear about it again. Rolling back a

recommendation doesn't silence it forever — if the same fragmentation-and-conversion

pattern shows up again in a later reporting period, you'll get another chance to review

and apply it.

This appears in the Recommendations tab of the Conversation Efficiency Analysis

report, alongside the Overview, Agents, Trends, and Correlation views.

Asking Copilot About It

You can also ask Copilot natural-language questions about the same data, for example:

• "Which of our agents send the most fragmented replies?"
• "Is our bot more chatty than our agents on this site?"
• "Has message fragmentation gone up this month?"
• "Do shorter, more fragmented replies correlate with lower star ratings here?"
• "Do fragmented conversations convert to sales less often than concise ones?"

Copilot will lead with a direct answer, show the supporting numbers, and give one clear

recommendation only when the data actually supports one — it will say the data is

inconclusive rather than guess. It also checks whether fragmentation correlates with

survey ratings and, when a store or CRM is connected, with completed purchases or won

deals — before drawing any conclusion, rather than assuming fragmentation is

automatically bad.

Requirements

Requires a Professional or Enterprise plan. Works on any site with engagement report

data — including sites still on the legacy (v10) engagement platform, since the

underlying Turns/Messages data is the same regardless of which platform recorded it.

See also: [Agent Performance & Conversation Quality Scoring](agent-performance-guide.md)

for AI-scored quality/sentiment on individual conversations — Conversation Efficiency

Analysis looks at message patterns across many conversations, not a single transcript.