Artificial Intelligence doesn’t just resolve tickets, it changes how much work reaches your team and how much that work costs.

In Crisp, AI ROI is not a vague promise. You can measure it concretely through your Analytics dashboards and a simple AI performance calculator that turns deflection and time saved into real financial impact.

This guide walks you through the essentials of measuring AI performance in Crisp. You'll learn which metrics truly matter for evaluating your chatbot’s impact, how to import a ready-made AI Performance dashboard into Crisp Analytics, how to interpret each metric correctly, how to plug those numbers into the AI ROI Calculator to estimate time and cost savings, and finally, how to improve your AI setup over time to continuously increase your ROI.

Understand what AI ROI means in Crisp

Before looking at numbers, it’s important to align on what “ROI” means for AI in customer support.

With Crisp, AI ROI is mainly driven by:

• Deflection rate: how many conversations are fully handled by AI
• AI-helped conversations: how often AI contributes to a resolution
• Resolution time (AI vs. human): how fast issues are resolved
• Automation rating / CSAT: how customers rate AI answers
• Misunderstanding rate: how often AI fails to understand the request

A strong ROI means:

• fewer repetitive conversations handled by humans,
• faster resolution for customers,
• stable or improved satisfaction,
• and visible time & cost savings for your team.

The goal of the rest of this article is to connect these metrics to your Crisp data and make ROI visible.

Import the AI Analytics dashboard into Crisp

The quickest way to start is to import a pre-configured dashboard into Analytics. It groups all key AI metrics in one place.

How do Crisp Analytics work?

How to import the dashboard

1. Go to Analytics → Create new dashboard.
2. Click Import dashboard.
3. Upload the .json file you downloaded.
4. The new dashboard appears in your list (for example: “AI Performance Calculator”).

Download the AI Performance Analytics dashboard (.json)

Once imported, your dashboard will look similar to this:

Each widget on this dashboard will be used later in the ROI Calculator.

Read each metric in the AI Performance dashboard

Conversations

Definition
Total number of conversations handled by your support team for the selected period.

Use it as a baseline to understand how much volume your AI and your agents are dealing with.

Automation first-hand

Definition
Number of conversations where the first reply to the user comes from AI (Chatbot or Overlay).

High “automation first-hand” means AI is acting as the front line, which is required for strong deflection.

Deflected conversations

Definition
Conversations fully resolved by AI without human intervention.

This is the main driver of AI ROI: more deflected conversations → fewer tickets for humans → more time saved.

Avg. resolution time – Human

Definition
Average time it takes for human agents to resolve a conversation.

This is used in the ROI calculator to estimate how much time an AI-resolved or AI-helped conversation would have cost in a typical human-only workflow.

Avg. resolution time – AI

Definition
Average time it takes for AI to resolve a conversation.

AI resolution time is usually significantly lower than human resolution time. This gap directly contributes to faster overall support.

Automation rating

Definition
Average rating of AI answers, when users rate the conversation.

This metric helps validate that deflection and speed do not degrade customer experience. Good ROI requires both high deflection and good ratings.

Fallback messages / month

Definition
Number of messages where AI could not answer and fell back (“I’m not sure”, “I don’t know”, etc.).

High fallback indicates:

• missing content in your AI Data Hub,
• or intents not covered in your AI Chatbot scenario.

Reducing fallback is one of the most effective ways to improve both ROI and customer satisfaction.

Total AI messages / month

Definition
Total number of messages generated by your AI (Overlay answers, chatbot replies, MagicReply suggestions counted as automated responses, etc.) during the selected period.

Use it to understand how actively your AI is participating in conversations and how much workload it is absorbing compared to human agents. A high number indicates strong automation coverage and directly contributes to measuring ROI and efficiency gains.

Use the AI ROI Calculator

Once your Analytics dashboard is in place, you can export the key metrics into the AI ROI Calculator to estimate time saved, FTE equivalent, and monthly cost savings.

You can follow the full walkthrough in this video:

And download the Crisp AI Performance Calculator (Google Sheet)

Your calculator will look similar to this:

Step 1: Copy your data from Crisp Analytics

In your AI Performance dashboard, note the following values for the chosen period:

• Monthly conversations
• Automation first-hand
• Deflected conversations
• Avg. resolution time – Human
• Avg. resolution time – AI
• Overall rating (or CSAT)
• Automation rating
• Fallback messages / month
• Total AI messages / month

These correspond to the INPUT section (blue cells) in the spreadsheet.

Step 2: Paste the values into the INPUT section

In the calculator:

1. Fill each blue field with the value from your Analytics dashboard.
2. Adjust Agent cost/hour, Support software cost, and Working hours / FTE to match your organisation.

The spreadsheet automatically computes the OUTPUT section:

• Deflection rate
• % AI-helped conversations
• Blended time to resolution (TTR)
• CSAT impact
• Misunderstanding rate

And the ROI block:

• Time saved per month (hours)
• Equivalent FTE saved
• Monthly cost saved

Step 3: Interpret the results

Some general guidelines:

• Deflection rate: Healthy setups often reach 40–70% deflection once mature.
• % AI-helped conversations: After a few weeks, aim for 20–40%+ of conversations where AI assists.
• Misunderstanding rate: Try to stay below 10% after the first month, then below 5% as your Data Hub improves.
• Time saved & FTE saved: These values show how many equivalent full-time agents your AI is saving at current automation levels.
• Monthly cost saved: This is an estimate based on your own agent cost and software cost; it helps communicate AI ROI internally (to leadership, finance, etc.).

Improve your AI ROI and troubleshoot common issues

Measuring ROI is only the first step. This section gives practical actions to increase your AI performance and short answers to common questions.

5 concrete actions to improve ROI

1. Strengthen your AI Data Hub

Your AI can only answer what it has actually learned. Keep your knowledge base fresh, turn recurring agent replies into short articles or snippets, and import any useful documents such as pricing sheets, policies, or internal workflows. Past resolved conversations are also an excellent source of training data.

The cleaner and more complete your Data Hub becomes, the lower your fallback rate and misunderstanding rate will be.

How can I add data sources to Crisp AI Hub and why?

2. Review fallback messages regularly

Once per week, check how often your AI falls back and use this signal to improve its knowledge.

1. Go to Analytics → Fallback messages (or your dedicated chart) to monitor how many fallbacks occurred during the selected period.
2. Open a few recent conversations with fallbacks directly from your Inbox (via search or tags) to see the exact questions the AI failed to answer.
3. For each recurring theme you identify in these conversations, add or update a help article, enrich your Data Hub, or map the intent inside your AI Chatbot scenario.

Each fixed gap reduces fallback frequency and improves both deflection and user satisfaction.

How can I format Knowledge Base articles?

How can I preview my Knowledge Base articles before sharing them with my visitors?

3. Optimise your Snippets

Your Answer Snippets are the core knowledge your AI relies on. If they’re clear, complete, and up to date, your automation rate increases and fallback messages drop.

In AI Automations → Data Sources → Answer Snippets, make sure your content covers what users ask most often. Aim for short, factual, easy-to-understand answers.

To keep your snippets effective:

• Cover your top recurring questions (billing, account access, errors, basic how-to).
• Make each snippet clear and self-contained, so the AI doesn’t need extra context.
• Update outdated information as soon as your product or policies change.
• Add missing snippets when you notice repeated questions in the inbox.

Good snippets let the AI confidently handle simple, frequent, well-structured requests, freeing your team for more complex conversations.

4. Monitor misunderstanding rate and ratings

From your calculator and dashboard:

If misunderstanding rate is high:

• Improve training data, article titles, and wording.
• Add examples of real customer phrasing to your Data Hub.

If automation rating is low:

• Review AI answers for the lowest-rated conversations.
• Adjust tone or content (too short, too vague, too technical, etc.).

Good ROI requires both efficiency and a positive experience.

5. Re-calculate ROI on a regular basis

We recommend:

• Updating the calculator monthly using the same period in Analytics.
• Tracking how: deflection rate evolves, fallback volume changes, and time and cost saved increase.

This helps demonstrate progress and justify further investments in AI content and scenario design.

Ticket deflection: How AI Chatbots reduce support backlogs

FAQ: Common questions about AI ROI in Crisp

“What’s a good deflection rate for an AI Chatbot?”

It depends on your industry and volume, but most mature setups fall in the 40–70% range for deflection on eligible intents.

“My deflection rate is low. Where should I start?”

Check, in order:

1. Is the AI Overlay visible and easy to use on your website?
2. Are frequent questions covered by your knowledge base and Data Hub?
3. Are your top intents mapped correctly in the AI Chatbot scenario?
4. Are many conversations still falling back or being escalated too early?

How do I setup Crisp Overlay on my chatbox widget?

Understanding and Mastering the Crisp AI Chatbot

“Does a higher deflection rate always mean better ROI?”

Not if ratings collapse. Always combine deflection with:

• Automation rating / CSAT,
• Misunderstanding rate,
• And where necessary, a manual review of conversations.

The best ROI comes from high deflection + good satisfaction.

“Can AI replace my support team?”

No. In Crisp, AI is designed to:

• absorb repetitive volume,
• speed up responses,
• give agents more context,

So that humans can focus on complex, emotional, or high-stakes cases. The calculator helps you quantify this shift, it doesn’t aim to remove humans from the loop.

Check out these helpful resources:

Nail 80% AI-resolved conversations through human-grade support with AI

Strategies to reduce time to resolution with examples from companies around the world

From FAQs to AI: How Chatbots Deliver AI Self-Service Support at Scale

Updated on: 26/11/2025