Choose the job the assistant should do.
A service website may want visitors to understand eligibility before making an enquiry. A software marketing site may want published product questions answered accurately. An agency may need reliable client separation and a manageable follow-up process.
Write a goal that you can review directly, such as whether questions about a published service are answered with the right conditions. Avoid starting with a target percentage before you know what the current visitor mix looks like.
Separate activity, outcomes, and quality.
Sairo’s analytics count conversations by their start time and use UTC day boundaries. The available reporting windows are 7, 30, and 90 days, with a comparison to the preceding period. Site scope follows the member’s permissions.
A conversation can have a recorded AI outcome and a separate team review status. A lead record indicates that contact details were captured. Neither automatically establishes a correct answer, a qualified prospect, or a completed sale.
| Question | Evidence to use |
|---|---|
| Are people using the assistant? | Conversation volume over a consistent site scope and period. |
| Are answers supported? | A reviewed sample checked against the actual source material. |
| Are visitors sharing contact details? | Conversations with a lead record, plus the context of the request. |
| Does someone handle escalations? | Assignments, follow-up records, and the actual completed response. |
| Is the website missing information? | Unanswered questions grouped by the content change they need. |
Build a representative review sample.
Include answered, unanswered, escalated, and lead-producing conversations. Also inspect pending conversations instead of quietly excluding them. A sample drawn only from successful-looking answers will miss the main failure modes.
For each reviewed conversation, record whether the answer addressed the full question, matched current sources, communicated uncertainty where needed, and offered an appropriate next step. Use the same criteria across reviewers.
Example sampling note
Review a fixed number of recent conversations from each important outcome, record why any record was excluded, and keep the same method for the next review. The sample size should fit your actual volume and the decisions you intend to make.
State the denominator when using a rate.
If you calculate a lead-capture rate, specify whether the denominator is all conversations, a selected campaign’s conversations, or only those judged relevant to the business. These are different measurements.
For example, 12 conversations with a lead record out of 80 conversations is 15%. That arithmetic describes capture in that set. It says nothing by itself about qualification, revenue, or whether the assistant caused the visitor to contact you.
Lead-capture rate = conversations with a lead record ÷ conversations × 100
12 ÷ 80 × 100 = 15%Control the context around a comparison.
Record changes to traffic sources, campaigns, website content, assistant settings, and team coverage. Compare periods of the same length and note whether the assistant was installed for the whole period.
A higher unanswered count after a new service launch may reveal a useful content gap. A lower conversation count may reflect lower traffic rather than poorer engagement. Read the conversations and the surrounding business context before concluding.
Finish each review with an owned action.
Choose a small number of improvements: fix a source, rewrite an approved answer, clarify a greeting, or assign responsibility for follow-up. Record an owner, a due date, and how the change will be checked.
Keep sales results and satisfaction measurements in the systems where they are actually recorded. Sairo’s operational analytics can inform those discussions without being presented as automatic attribution or a verified return on investment.
Review what needs attention. Use visitor activity to choose the next improvement.
Before you start: Conversations have started arriving in your workspace. The workspace, visitors and figures shown are fictional examples.
Read the transcript
- See the pattern. Improve the answer.
Make a short review part of your week. Use Sairo's analytics to choose one useful improvement.
- Set the scope
Open Analytics. Choose the website and reporting period, then apply the filters. The report uses UTC calendar days.
- Read the signals
Compare conversations, leads and bookings, and capture rate. Look at unanswered outcomes for signs that visitors need information your assistant could not provide.
- Keep the context
Use the activity chart and outcome breakdown to understand the pattern. These outcomes describe conversations; they are different from the list of knowledge gaps still needing attention.
- Export if useful
Select Export daily report when you need a file for a team review. The export follows the report's selected filters.
- Find open gaps
Open Conversation inbox, choose Unanswered, and apply the filters. This view helps you find questions that still need a verified answer.
- Choose one fix
Open a conversation and read the actual question. Here, a visitor asked about skylight repairs. Confirm the answer with your team, then save it in Knowledge library.
- Review. Learn. Help the next visitor.
Choose one gap, save a confirmed answer, and check it. That simple weekly habit keeps your assistant useful as your business changes.