---
title: "AI Concierge: Performance"
description: "> The Performance tab is your at-a-glance answer to \"is the AI Concierge actually doing anything for us?\" It shows how guests are engaging with it, what it is costing, and which questions and properties come up most. --- ##…"
url: "https://homerunner.io/docs/ai-concierge-performance/"
updated: "2026-10-17"
category: "Operator Tools & Tuning"
---

# AI Concierge: Performance

> The Performance tab is your at-a-glance answer to "is the AI Concierge actually doing anything for us?" It shows how guests are engaging with it, what it is costing, and which questions and properties come up most.

---

## What it does

Performance is a read-only dashboard that summarizes AI Concierge activity over a date range. It pulls from two places: the shared Conversion Kit events table (for engagement) and the per-message cost ledger the Concierge keeps for every AI call. Nothing here is guest-facing; it is for you.

## Where to find it

Go to **HomeRunner > AI Concierge** and open the **Performance** tab.

Use the **Range** selector in the top right to switch the whole report between **This month**, **Last month**, **Last 30 days**, and **Last 90 days**.

If Conversion Kit is not active on the site, the engagement counters will read zero (the assistant has nowhere to record events), and you will see a notice saying so. Cost figures still report regardless of Conversion Kit, because the Concierge logs those itself.

## What it reports

### Engagement counters

A grid of tiles, each a simple count for the selected range:

- **Drawer opens**: times the chat drawer was opened.
- **Unique sessions**: distinct guest sessions that interacted with the Concierge (a more meaningful "people helped" figure than raw event counts).
- **Messages sent**: guest messages.
- **Searches performed.**
- **Zero-result searches**: searches that returned nothing.
- **Lead captures**: times a guest handed over contact details.
- **Recovery chip clicks**: clicks on the suggested follow-up chips.
- **Property card clicks.**
- **Property link clicks**: times a guest clicked a property link the assistant wrote directly into one of its answers, separate from clicking a property card in the results grid.
- **View-all clicks.**
- **CTA clicks.**

### Cost

- **Total spend** in USD for the range.
- **Prompt tokens** and **Completion tokens** consumed.
- **AI messages** counted.
- A **per-model breakdown** table showing messages and cost for each AI model used, plus a note if any messages could not be priced.

### Top lists

- **Top guest requests**: the most common things guests asked, with counts.
- **Top failed searches**: the most common search or refinement requests that returned zero results, with counts. This is your demand-gap list: what guests want that your catalog (or its tagging) is not answering.
- **Top properties surfaced by AI**: the listings the Concierge put in front of guests most often, with how many times each was surfaced.

## Does the AI Concierge actually drive bookings?

Engagement counts tell you guests are using the assistant. They do not tell you whether those chats turn into money. A Concierge can field hundreds of questions and still leave you wondering whether any of it moved a guest from browsing to booking. Booking attribution closes that gap by tying confirmed bookings, and the value behind them, back to the guest's earlier conversation with the assistant.

Here is how it works in plain terms. Every time a guest clicks a property in the chat (a property card, a home's name in an answer, or a call-to-action button), the Concierge records the click against that visitor. If the same guest later completes a confirmed booking for that same home within **7 days** of the click, the booking is credited as **Booked within 7 days of a concierge click**. This is the headline figure, because it ties the booking to something the guest did after talking to the assistant.

The report also keeps softer lines so you can see how far the assistant's influence reaches. Looking back over the previous 30 days of that guest's activity, it checks whether the guest chatted with the Concierge at all (**Chatted within 30 days before booking**), whether the assistant showed the exact home that was booked, and whether the assistant had hand-picked that home. Each booking is counted at its strongest level only, so the numbers never double up.

A concrete example. A guest opens the chat on Monday, asks for a dog-friendly cabin near the lake, and the Concierge surfaces three homes. They click one and leave without booking. Four days later they come back and book that same cabin. That booking counts as **Booked within 7 days of a concierge click**, and its full booking value appears with it. If the same guest had waited ten days, the booking would still show under the softer lines, such as **Booked from AI-surfaced property**, because the assistant had shown that exact home, but it would not count as a click booking.

You will find this in the same place as the rest of the report, on the **Performance** tab. Look for four sections:

- **Concierge results** sits at the top as a strip of tiles, each with a one-sentence definition under it: **Conversations** (chat sessions started in the period), **Questions answered** (assistant replies with content, not counting searches that found nothing), **Zero-result questions** (searches the Concierge could not match to any home), **Property links clicked** (clicks on a property card, title link or call-to-action inside the chat), **Booked within 7 days of a concierge click** (with the gross booking value per currency), and **Chatted within 30 days before booking**.
- **AI-surfaced gross booking value** shows the total value of bookings the assistant helped drive, broken out by currency. There are five figures per currency: **Booked within 7 days of a click** (the headline number), **Chatted within 30 days** (any booking that followed a Concierge chat), **AI-surfaced** (the assistant showed the exact home that was booked), **AI-handpicked** (the assistant specifically curated the home that was booked), and **AI link-clicked (older rule)** (the guest clicked a link to that home inside a reply, with no 7-day limit).
- **Booking attribution** shows the counts behind those values: **Booked within 7 days of a concierge click**, **Chatted within 30 days before booking**, **Booked from AI-surfaced property**, **Booked from AI-handpicked property**, **Clicked an AI link within 30 days (older rule)**, and **Total attributed bookings**. Below the tiles is a per-booking table so you can see each attributed booking, when it was booked (UTC), its value, which level it was credited at, and how it was matched to a guest.
- **Leads & bookings by source** breaks chat leads and AI-attributed bookings down by the first-touch channel that brought the guest to your site in the first place, a UTM-tagged campaign, a Google Ads click, a Meta Ads click, or the external site that referred them. Each row shows the channel, how many chat leads it produced, how many bookings it is credited with, how many of those booked within 7 days of a click, and the booking value, so you can see which marketing channel is actually converting through the Concierge. A visitor with no recorded campaign, ad click, or referrer shows up as **Unattributed** rather than being guessed at, so the totals stay honest even when the signal is missing. This needs Conversion Kit's visit tracking; without it you will see a note here instead of the table.

About that gross booking value figure. It is what the guest paid: rent, cleaning, booking fees, and taxes, net of discounts, with refundable deposits left out. It is a booking total, not your take-home revenue, since it still carries pass-through taxes and fees.

Honest limits, so the number means what you think it means:

- **Confirmed bookings only.** Attribution records a booking when it is confirmed; it is not counting carts, inquiries, or quotes. This follows the same confirmed-reservations-only rule used everywhere in HomeRunner analytics.
- **Two look-back windows.** A concierge click earns the headline credit only when the booking lands within 7 days of it. The softer lines look back 30 days, so a guest who chatted two months before booking will not be tied to that earlier conversation.
- **The guest has to be recognizable.** The assistant connects a booking to a chat by matching the guest's identity or browser session. If a guest chats on their phone and later books on a different device without identifying themselves, the link can be missed, so treat these figures as a confident floor rather than a perfect count.
- **Conversion Kit has to be active.** Conversion Kit supplies the visitor identity that connects a click to a later booking. If it is not active on the site, expect no attributed bookings.
- **Original value at time of booking.** The value shown is the booking total when it was made. Later cancellations and modifications are not yet deducted, so a cancelled stay still counts toward the total for now.
- **Bookings without a clean value or currency** are counted toward the booking totals but left out of the value sums, and the section calls out how many were excluded so the money figures are not quietly overstated.

For the wider question of how guests reach the assistant and where leads come from, the engagement counters and top lists above are your starting point.

## How to read it

- **Drawer opens and unique sessions** tell you reach: how many guests are actually engaging.
- **Zero-result searches and Top failed searches** are the most actionable signals. A recurring failed request often means a gap in your inventory, or a listing that exists but is not tagged the way guests phrase it.
- **Top guest requests** show you, in guests' own words, what they care about, useful for listing copy and marketing.
- **Top properties surfaced** highlight which homes the Concierge leans on, so you can sanity-check that your best inventory is getting airtime.
- **AI-surfaced gross booking value and Booking attribution** tell you whether the engagement is paying off. Watch **Booked within 7 days of a concierge click** as your most defensible read on revenue the assistant helped drive, and compare it with Property links clicked to see how often guests who follow the assistant's recommendations go on to book.
- **Leads & bookings by source** tells you which marketing channel is worth funding. If one campaign consistently produces leads but no bookings while another converts at a high rate, that is a budget conversation, not just a Concierge one.
- **Cost** keeps spend visible. The per-model breakdown shows where the spend goes. Set total spend against the gross booking value of click bookings to gauge return.

For everything below the surface (model selection, spend limits, data handling), see [Privacy, Data, and Cost Controls](/docs/ai-concierge-privacy-data-and-cost-controls/).

## Health

Next to Performance is a **Health** tab: a read-only diagnostics panel, mainly there for troubleshooting rather than day-to-day use. It shows internal pipeline counters for the last 14 days. All zeros is the healthy reading, meaning the assistant is settling every request cleanly.

You will not normally need this tab. If chat ever seems to be misbehaving (slow answers, an odd fallback reply) and you contact support, they may ask you to check the Health tab and read off what it shows; that saves a round of back-and-forth diagnosing the issue blind.

## Tips

- **Empty report?** If you see "No AI activity yet," open the chat drawer on a guest-facing page to start recording, then come back.
- **No attributed bookings yet?** The headline line only counts a booking confirmed within 7 days of a guest clicking a property in the chat (the softer lines look back 30 days), and Conversion Kit has to be active. Early on, expect this section to stay quiet.
- **Work the failed-searches list monthly.** It is the cheapest product feedback you will get.
- **Compare ranges** (this month vs. last month) to see whether engagement, attributed bookings, and cost are trending the way you expect.
- **Want a heads-up the moment a lead comes in, not just at report time?** [Slack Notifications](/docs/aise-slack-notifications/) posts each new chat lead straight to your Slack, badged with the same acquisition source shown in the Leads & bookings by source table here.