Skip to main content
View Categories

AI Concierge: Operator Tools

Table of Contents

Everything you configure and monitor lives under HomeRunner -> AI Concierge in your WordPress dashboard. This guide walks through each tab.

Tab map #

The admin panel is organized into tabs. This guide walks through each one in the order you’re likely to use it.

TabWhat you do here
GeneralMaster enable/disable, debug mode, voice-input language
PerformanceCost dashboard, booking attribution, response times
LicenseLicense activation, deactivation, trial status
APIOpenRouter API key, default model, response temperature, daily spend cap
Search PromptTune which property attributes the AI prioritizes
ConciergeSetup checklist, brand voice, recovery messaging, lessons
ChatbotOn-page chat drawer, greeting, and search recovery
Live ConciergeLive human takeover: availability, Slack operator chat, session mirror
IntegrationsPush high-value Concierge events to Slack
Property KBProperty page experience toggle and bulk per-property knowledge editor
Self-AssessmentQuality loop where the AI grades its own responses
Photo AnalysisOne-time vision pass over property photos
Test SearchOperator sandbox for dry-running queries
Chat SessionsReview every chat conversation
Search HistoryAnalytics over guest search queries

General #

  • Master enable/disable: the single switch that turns the Concierge on or off site-wide.
  • Debug Mode: adds verbose logging to your PHP error log. Leave off for production.
  • Voice Input Language: language preference for the voice-input feature on the chat drawer. Default is “auto” (let the browser decide).

Save settings before leaving the tab.

Performance #

A live dashboard showing the business value AI Concierge is delivering. This is what you’d show your business partner at renewal time: cost trajectory, booking attribution, response times, and recovery-chip engagement. It has its own full write-up in Performance.

Concierge #

This tab is where you shape how the assistant talks and recovers.

  • Concierge setup checklist: at the top of the tab, a green/red checklist shows which core pieces are configured (license, brand voice, knowledge, photos). Use it to confirm you’re fully set up.
  • Brand Voice: a short description of how your team talks, applied to every reply. Give it examples and guidance. Specific input produces specific output. “Warm but professional, drop coffee-shop references” beats “friendly tone” every time.
  • Concierge Lessons Panel: over time the Self-Assessment loop surfaces patterns where the AI is off-target. This panel shows those lessons and lets you accept them into the prompt.

Chatbot #

Configure the on-page chat drawer guests interact with.

Drawer settings #

  • Position: left or right side of the screen
  • Greeting: the opening message guests see when they open the drawer
  • Tooltip: optional small nudge that appears next to the chat tab after N seconds on the page

Recovery & Outreach #

When a search returns zero results or too many results, the Concierge offers proactive recovery instead of a dead-end:

  • Progressive filter relaxation: it drops the least important filters first (features, then size, then guest count) and re-runs the search, so the guest still gets options.
  • Recovery chips: tappable suggestions the guest can act on in one click, such as relaxing a size or guest-count constraint.
  • Alternative-date chips: “Your dates aren’t available, but the same property is open the week before.”
  • Lead-capture prompt: for queries that genuinely exceed the catalog (“we couldn’t find that, want to tell us more so a human can follow up?”).

Live Concierge #

Live Concierge is where you set up handing a guest conversation to a real person. You choose the hours your team is available, connect the Slack app your operators chat from, and decide whether to mirror conversations into Slack and let an operator take over a chat the guest did not request. The “talk to a live person” option only appears to guests when live takeover is on, the current time is within your hours, and an operator has marked the team available. See Live Concierge Takeover for the full setup.

Integrations #

Push high-value Concierge events to the tools your team already uses. Slack notifications are available now: one-way alerts for moments worth a human’s attention, such as a captured lead. See Slack Notifications for what gets sent and how to connect it.

Property KB #

Property KB is where you turn the property-aware chat on and feed it per-listing knowledge. The Property Page Experience settings sit at the top of the tab, and a bulk knowledge browser sits below.

Property Page Experience #

A site-wide toggle for the property-aware chat behavior:

  • Enable on property pages: when ON, the chat drawer opens on individual listings with a property-specific greeting, three contextual chips, and the property’s knowledge base injected into the prompt. This master switch is off by default.
  • PMS Knowledge Base custom field: a picker of the custom fields found across your published properties. The Concierge reads this field per property and includes it in the prompt. Default: homerunner_concierge_kb. If your PMS doesn’t sync a knowledge field, leave it and author content manually below.

A notification dot and a short pulse animation draw guest attention to the chat tab on property pages without being intrusive.

Local knowledge browser #

A single table lists every published property, drawing on the PMS-synced custom field and a per-property local field you can edit by hand. Review or edit local knowledge across the whole catalog in one place: a few sentences on your top properties (house rules, local tips, owner notes) noticeably sharpens answers. See Property Chat.

Self-Assessment #

A quality loop where the AI grades its own responses over a rolling window of conversations and surfaces patterns where it’s off-target. Accepted lessons flow back into the Concierge prompt (see the Concierge Lessons Panel above). This is the foundation of how AI Concierge gets better at your specific catalog over time. Worth checking every 1 to 2 weeks once you’re settled in.

Photo Analysis #

A vision pass over your property photos that extracts both an overall summary and per-room observations the chatbot can cite when a guest asks about a specific space. Six canonical room types are captured for each property: living, kitchen/dining, bedrooms, bathrooms, outdoor, amenities. That room data feeds directly into the Concierge’s answers.

Click Edit on any property row to expand its analysis. You’ll see three sections:

  • Summary: a 2 to 3 sentence whole-property description the Concierge uses for top-of-conversation context.
  • Rooms (read-only): a table showing each room type the model observed, with visible features as pill chips and a qualitative note. This is exactly what the Concierge can cite in chat. A room with no visible feature pills won’t appear in the table.
  • Details (JSON): per-photo descriptions and tags. Operator-editable if you want to tweak captions.

The Rooms table only shows rooms the model actually saw in your gallery. A missing room means the gallery lacked a clear photo of it, not that the property lacks the space. The Concierge treats photo data as supportive observation, never as a complete inventory, and will never tell a guest a feature is absent based on photo absence. See Photo Analysis for the full detail, including regenerating a single property’s analysis.

Test Search #

An operator sandbox for dry-running queries the way a guest would, so you can sanity-check brand voice, recovery behavior, and matching before going live.

Chat Sessions #

Review every chat conversation guests have had with the Concierge. Useful for spotting where guests get stuck and where the assistant is performing well.

Search History #

Analytics over guest search queries: what people are asking for, where searches dead-end, and which recovery chips get clicked. Pairs with the Performance dashboard.

Related #