Your ai chatbot resolution rate tells you whether your support bot actually solves problems or simply ends conversations. A bot can answer hundreds of questions a day and still leave most customers without a real solution. Measuring resolution properly shows what the bot does well, where it fails, and whether it is worth maintaining.
Resolution is not the same as deflection. Deflection counts every conversation that does not reach a human agent. Resolution counts only the conversations where the customer got what they needed. A customer who gives up or closes the chat window has been deflected, not helped. Reporting that as success keeps a weak bot in place while customers quietly leave.
This guide shows how to measure ai chatbot resolution rate fairly — what counts as resolved, how to record it, and how to improve a low rate.
What AI Chatbot Resolution Rate Really Means
The ai chatbot resolution rate is the share of chatbot conversations that end with the customer's problem solved, without a human agent stepping in. In plain terms, it answers one question: out of every 100 chats the bot handled, how many ended because the customer was helped?
A simple formula sits at the center of the measurement:
Resolution rate equals resolved chatbot conversations divided by total chatbot conversations, multiplied by 100.
If your bot handled 400 conversations in a month and 240 of them were resolved by the bot alone, the resolution rate is 60 percent. That number is only honest if your definition of resolved is clear and your data is complete. Two businesses can report very different rates for similar bots simply because they define resolution differently.
Before you measure anything, write down what resolved means for your business. A resolved chat is one where the customer received a correct and complete answer, or completed the task they came for, such as finding a delivery date or understanding a return policy. An unresolved chat is one where the customer escalated, repeated the question, or left without completing the task.
Resolution Rate, Containment Rate, and Deflection: Do Not Mix Them Up
Chatbot teams use several related terms, and mixing them up is the most common measurement mistake. The chatbot containment rate means the conversation stayed inside the chatbot channel and did not transfer to a human. That is a channel measure. Resolution is an outcome measure. A chat can be contained and unresolved at the same time, which happens whenever a customer gives up inside the bot.
For a small business, the practical rule is simple. Use resolution rate as your main quality measure. Use containment rate as a secondary measure of how much work stays with the bot. Use deflection only as a rough workload signal, never as proof that customers are satisfied. When you present numbers, label each metric clearly and explain what it includes.
Define Resolution Before You Count Anything
Define resolution before you count anything. If you start counting before you define, you will count the wrong things and argue about the results later. A good definition has three parts: the outcome, the evidence, and the edge cases.
Start with the outcome. List the five to ten most common reasons customers contact you, such as order status, delivery times, returns, pricing, and account access. For each reason, write one sentence that describes a solved case. For order status, a solved case might be that the customer received the correct delivery date and tracking link. These sentences become your test for resolution.
Next, decide what evidence you will accept. The strongest evidence is a direct customer signal, such as a thumbs up or a message like "that helped, thanks." The next level is behavioral evidence, such as the customer completing a task or ending the chat after a complete answer. The weakest evidence is silence. A customer who stops replying might be satisfied or frustrated, so treat silence as unknown, not as resolved.
Finally, list your edge cases in advance. Decide how to classify chats where the customer asked two questions and only one was answered, chats where the bot gave a partial answer the customer accepted, and chats that ended because of a technical error. Write these decisions down so two reviewers classify the same chat the same way.
To measure ai chatbot resolution rate honestly, define ‘resolved’ first, because a chat that ends is not the same as a problem solved.
For the test plan behind these numbers, use our AI customer support chatbot testing checklist.
Record Outcomes Fairly
Once the definition exists, you need a reliable way to record outcomes. Most chatbot tools offer conversation logs, tags, or analytics, though the exact feature names differ between tools, so check your own tool rather than assuming how it works. The goal is a simple record for each conversation: when it happened, what the customer asked about, and how it ended.
A practical setup uses a small set of outcome tags that every conversation receives. Keep the list short so reviewers actually use it. A useful starter set is: resolved by bot, escalated to human, unresolved and abandoned, out of scope, and technical failure. Every conversation should carry exactly one of these tags. If a chat could fit two tags, your definitions need sharpening, or the reviewer needs clearer guidance.
Customer feedback is the best way to confirm resolution, but keep it light. A single question at the end of a chat, such as "Did this answer your question?" with yes and no options, gives you a direct signal without annoying the customer. Ask it only once, after the bot has given its main answer, and make it optional. If only a small share of customers answer, do not treat their answers as representative of everyone.
Human review is the second layer. Each week, read a sample of 20 to 30 conversations across all outcome tags, including ones marked resolved. Look for answers that sound confident but are wrong, answers that are correct but incomplete, and cases where the customer said thanks out of politeness rather than satisfaction. Keep short notes on what you find.
Privacy matters during review. Conversation logs can contain names, email addresses, and order numbers, so limit access to the people who genuinely need it and follow your own retention rules. The bot should never ask for passwords or payment card details in chat. For more on this, see the guide to AI chatbot customer data privacy.
The way you measure ai chatbot resolution rate matters as much as the number, so record outcomes the same way every week.
Most reporting features live in the chatbot plugin, so compare them in the official WordPress plugin directory.
A Worked Example: An Example KPI Calculation Sheet
An example KPI calculation sheet is the unique value of this article, so here is a simple worked example you can copy into a spreadsheet. Imagine a small online shop whose bot handled 500 conversations in one month: 280 resolved by bot, 90 escalated to a human, 70 unresolved and abandoned, 40 out of scope, and 20 technical failures.
The resolution rate is 280 divided by 500, multiplied by 100, which is 56 percent. The escalation rate is 18 percent, abandonment is 14 percent, out of scope is 8 percent, and technical failures are 4 percent. These five numbers add up to 100 percent, which is a useful check that no conversation was missed or counted twice.
Now add a second layer. Of the 280 resolved chats, 120 customers answered the end of chat question, and 102 of them said yes. That gives a confirmed resolution signal of 85 percent among respondents. A good report states both facts: the overall rate is 56 percent, and 85 percent of customers who gave feedback confirmed the answer helped.
Add a third layer by topic. Suppose order status chats resolved at 78 percent, delivery questions at 64 percent, returns at 41 percent, and pricing at 52 percent. The overall rate of 56 percent hides the real story, which is that returns need urgent work. Build your sheet with one row per topic and columns for total chats, resolved chats, escalated chats, abandoned chats, and resolution rate.
This worked example shows how to measure ai chatbot resolution rate with a simple sheet, so you can repeat the same calculation monthly.
Which Support KPIs Matter Alongside Resolution
Resolution rate is your headline measure, but it needs supporting chatbot support metrics to make sense. A bot can have a decent resolution rate and still be slow, rude, or confusing. Track a small set of companion measures and review them together.
First, track escalation rate, the share of chats that move to a human. A rising escalation rate often means the bot is failing on a new topic. A falling escalation rate is good only if resolution is steady or rising. If both fall, customers may be abandoning chats instead. Second, track first response time and total conversation time. A slow first response usually points to a technical problem. Very short chats can mean quick answers or customers leaving fast, so read a sample of the longest and shortest chats before drawing conclusions.
Third, track customer satisfaction for bot handled chats separately from human handled chats, because mixing the two hides the bot's real performance. A short thumbs up or thumbs down question is enough. Fourth, track repeat contact rate. If a customer chats with the bot and then emails or calls about the same issue within a few days, the first contact probably did not resolve the problem. Finally, track the share of chats that are out of scope. Those requests need a clear route to a person, not a forced bot answer.
The ai chatbot resolution rate is only one number, so track it next to handoff rate and customer satisfaction to see the full picture.
Tools like Tidio report resolution and handoff numbers automatically — see Tidio’s official site for how reporting works.
Review Trends and False Positives
A single month of data tells you where you stand. Several months tell you whether you are improving. Review trends monthly, using the same definition and the same tags each time. A sudden drop in resolution often follows a website change, a price update, or a knowledge base edit. A slow decline often means content is going stale.
False positives deserve a dedicated check every month. A false positive is a chat marked resolved that was not actually resolved. Common causes include a customer saying thanks to end the conversation politely and a bot giving an answer that sounds complete but misses the real question. To find them, take a random sample of chats tagged resolved and re-read them, asking for each one whether the customer got what they came for. If you cannot tell from the log, mark it as unclear rather than resolved.
Segment your trends so a problem in one area does not hide behind a good overall average. Break resolution down by topic, by device, and by time of day. A bot that resolves 70 percent of desktop chats but only 40 percent of mobile chats may have a display problem, not a knowledge problem. These splits turn a vague worry into a specific fix. For the after hours angle, the guide to AI chatbot after hours support covers message design and escalation expectations.
How to Improve a Low Resolution Rate
When the rate is lower than you hoped, resist the urge to rewrite everything at once. Pick the topic with the most chats and the lowest resolution, and fix that first. Read 20 conversations on that topic. You will usually find one of four problems: the answer is missing from the knowledge base, the answer exists but the bot cannot find it, the answer is correct but unclear, or the question needs a human and should escalate faster.
If the answer is missing, add a short, grounded source document that states the facts plainly. If the bot cannot find an existing answer, check how real customers phrase the question and add those phrases to your source material. If the answer is unclear, shorten it, put the key fact first, and add one clear next step. If the question needs a human, write a better escalation message that explains what happens next. Test the change with ten fresh conversations before rolling it out widely.
Keep a simple change log. Record the date, the topic, what changed, and the resolution rate for that topic in the month before and after. Without a log, you will forget which change helped and repeat changes that did not work.
To improve a low ai chatbot resolution rate, fix the top unanswered questions first, because a few missing answers usually cause most failures.
For the root cause, read our guide to reducing AI chatbot hallucinations in support.
Limitations and Honest Reporting
No measurement system is perfect, and resolution rate has real limits. It depends on honest tagging, and tagging depends on people or tools that can make mistakes. It misses customers who never open the chat at all, and it cannot capture the customers who were so frustrated that they left the website entirely. It also says nothing about whether the answer was kind, clear, or easy to follow. Use it as your main quality signal, not as the only signal.
Be careful about comparing your rate with other businesses. Published figures differ because definitions differ, industries differ, and customer questions differ. A shop that sells one simple product will naturally resolve more chats than a business with complex custom orders. Compare your current rate with your own past rate, using the same definition, and aim for steady improvement.
Privacy and consent belong in every report. Resolution tracking uses real customer conversations, so keep the data inside the tools meant to hold it. Remove or mask personal details before sharing examples outside your team, and delete logs on the schedule your privacy policy promises. If your bot collects an email address or order number during a chat, explain why it is needed and store it only where it belongs.
Frequently Asked Questions
What is a resolved chat?
A resolved chat is a conversation where the customer received a correct and complete answer or finished the task they came for, without a human agent stepping in. The definition should be written down for your business, with examples for your most common topics. A chat that ends because the customer gave up, timed out, or repeated the same question is not resolved, even if it never reached a person. Direct customer confirmation, such as a yes answer to "Did this answer your question?", is the strongest evidence, but silence alone should be treated as unknown.
Which support KPIs matter?
The most useful companion measures are escalation rate, customer satisfaction for bot handled chats, repeat contact rate, first response time, and the share of chats that are out of scope. Escalation rate shows how much work still reaches humans. Satisfaction shows whether customers liked the experience, which is different from whether the problem was solved. Repeat contact rate catches false positives, because a customer who contacts you again about the same issue was probably not helped the first time.
Conclusion
Measuring ai chatbot resolution rate is a habit, not a one off calculation. Write a clear definition, tag every conversation with a single honest outcome, confirm resolution with customer feedback and human review, and track a small set of supporting chatbot support metrics alongside the headline number. Watch for false positives, segment the results by topic, and fix the weakest topic first.
Your next step is simple. This week, write your one paragraph definition of a resolved chat, set up five outcome tags in your chatbot tool, and read 20 conversations from the last seven days. You will learn more from that small exercise than from any dashboard. If your knowledge base needs work first, the guide to creating a knowledge base for an AI support bot is a good place to start.