Sara Chen

President · Hearth Hotels

Sara runs Hearth Hotels, a seven-property boutique chain in the American West. She’s preparing a franchise pitch for the next eighteen months and looking for the brand-differentiator that will carry it.

The setup

Hearth is a seven-property boutique chain Sara’s family started in the late nineties. Each property runs forty rooms, sits in a different town across the American West, and trades on a single positioning: attentiveness. The brand promise is that staying at Hearth feels like being taken care of by people who know your name.

That promise works during the day. It fails at 2am. The smaller properties can’t staff the night desk; the larger ones run one person who is also doing turn-down service and the morning prep. Guest requests pile up between 10pm and 7am, pool hours, extra towels, the occasional locked-out room, and the morning-shift GM walks in to a list of overdue tickets and a satisfaction score that dropped three points overnight.

Sara has been losing GMs to burnout. The chain has lost two in the past eighteen months, both citing the on-call overnight load as the reason. The cost of replacing a GM at a 40-room boutique property runs $80-120K once you count the search and the ramp.

She came to Zynkex preparing the franchise pitch deck for next year and looking for the brand-differentiator that would carry it. Her opening: “I want every guest to feel like they have a concierge, but I can’t hire one for every shift at every property.”

The session

The session opened with Sara framing the project as "an AI for guest service." Zynkex pushed back on the framing inside the first dozen rounds. AI-for-guest-service is the category competitors are already losing in, every general-purpose hotel chatbot has the same problem, which is that guests can tell within two messages they’re talking to a machine, and the brand promise dies on contact.

The reframe came from a different angle. What is the request category where guests don’t expect an immediate human voice? After-hours. A guest texting at 2am isn’t expecting the front-desk manager personally; they’re expecting "the front desk", an abstract entity that handles things. That abstraction is where an agent can live without breaking the trust.

From there, the scope tightened. Not "guest service", that’s daytime, that’s humans, that doesn’t change. "After-hours concierge." Active only between 10pm and 7am. Reservation-scoped read access. Property-aware fact lookup. Slack escalation to the on-call GM for anything outside the boundary. The narrow scope itself becomes the trust signal: the agent earns credibility by what it refuses to do.

The strategic moment was identifying the GM-retention problem as the real ROI driver. Sara had been framing this as a satisfaction-score project, but the math doesn’t work that way, $4-point NPS improvement is real but slow to capitalize. The fast win is keeping the GMs from quitting. At $80-120K per replacement and two departures in eighteen months, the after-hours load is a $200K-per-year burnout-driven attrition cost.

By the end of the session, the product narrative was clear. The agent isn’t about guests at all. The agent is about the GMs, and the boutique-luxury brand position the chain can hold only if the GMs stay long enough to learn what attentiveness means at their property.

The Build Brief

For the engineering vendor building the SMS agent and the Slack escalation pipeline. Names the interaction surface, the tool inventory, the conversation patterns, the voice rules, the escalation SLA. The .docx the build runs on.

Build Brief for the Hearth Hotels after-hours concierge agent. Paste into your AI coding tool; the tool will determine the file structure for the build environment. The implementation produces an SMS-based agent reachable from every property's guest-facing materials, a property-aware tool inventory, and an escalation pipeline that hands off to the on-call GM via Slack. Read end-to-end before generating code, the after-hours scope in Section 1 governs every interaction decision downstream.

Hearth Hotels, After-hours concierge agent

What this is. An SMS agent that handles guest requests at all seven Hearth properties between 10pm and 7am local time, when front-desk staffing is thin and GMs are off-shift. The agent operates as the after-hours concierge, not as a general-purpose chatbot, not as a replacement for human service during the day.

The after-hours-scope constraint. Every decision in this brief is gated on the fact that this agent exists only for the hours when human service is materially constrained. During daytime hours, guest requests flow to front-desk staff as they always have. The agent does not compete with humans; it covers the gap humans cannot.

Hearth is not adding AI to guest service. Hearth is adding the after-hours concierge the chain has never been able to staff at every property.

When uncertain whether a feature belongs in the agent, ask: would this matter at 2am when no human can respond? If yes, build it. If it makes sense only because daytime staff are busy, redirect it to a staffing decision instead.

Section 1. Interaction surface

Channel. SMS only. The agent's number is printed on the in-room key card and posted in the property elevators with the line: "Need something after-hours? Text the front desk, we'll get back to you fast."

Purpose: guests don't download apps for a single hotel stay. SMS works on every phone, every carrier, with zero onboarding. The "front desk" framing keeps the agent invisible, guests think they're texting the desk, which they are; the agent answers when staff can't.

Active hours. 10pm to 7am, property-local time. Outside those hours, the SMS line forwards to the property's front-desk system as before.

First-response target. Under 30 seconds from inbound SMS to outbound reply. The agent acknowledges every message immediately, even when escalating to a human.

Section 2. Tool inventory

The agent has read access to the property management system (Mews) for the requesting guest's reservation. It can:

Look up reservation details, room number, length of stay, special requests on file, prior interactions.
Confirm in-property facts, pool hours, breakfast service start time, parking validation rules, wifi credentials (property-specific).
Send room-supply requests to the night manager via Slack, extra towels, pillows, toothpaste, charger cables.
Escalate to the on-call GM via Slack with a one-tap "needs human" handoff.

The agent does NOT have access to:

Booking modifications. Changing dates, room types, or rates requires a human.
Charge approvals. Any request that adds a line item to the folio is escalated.
Information about other guests. Privacy boundary, hard.

Section 3. Conversation patterns

The agent answers in the Hearth brand voice. Warm but not chatty. Plain language. Never asks the guest to repeat themselves.

Standard request (in scope, no escalation needed). Guest texts asking about pool hours. Agent replies in under 30 seconds with the answer + a single follow-up sentence: "Pool's open 6am to 10pm. Can I get you a robe to your room?"

Purpose: the follow-up question is the load-bearing brand differentiator. Hearth's positioning is hospitality-as-attentiveness; the agent's job is to anticipate the next request, not to mechanically answer the one in front of it.

Out-of-scope request (escalation). Guest texts at 2am asking to extend their stay by two nights. Agent replies: "Let me get our overnight manager to look at availability. I'll text back within 10 minutes." Then pings the on-call GM via Slack with the guest's reservation context.

Purpose: never say "I can't help with that." Always frame the escalation as a routing decision, not a capability gap. The guest experience is "someone is handling it," not "the chatbot didn't understand me."

Resist the urge to make the agent capable of more, booking modifications, charge approvals, anything that touches the folio. The boundary between "AI can answer" and "human must answer" is the trust signal. Every line crossed erodes the boutique-luxury position Hearth charges for.

Section 4. Escalation pipeline

The on-call GM receives Slack pings in a dedicated #night-desk channel. Each ping shows:

Guest name + room number.
The original SMS thread.
The reason the agent escalated.
A single button: Take over (replies into the SMS thread as the GM, with the agent stepping back).

GM SLA on the escalation: respond within 10 minutes. If no response within 10, the ping re-routes to the regional manager.

Section 5. Voice rules

Plain, attentive, never effusive. Right. Got it. Let me check., never Absolutely! or Of course!
Address the guest by name when known.
Anticipate the next request. The pool-hours follow-up ("Can I get you a robe?") is the brand differentiator.
Never apologize for being an agent. The guest is texting the front desk; that's true at 2am whether a human or the agent is on the other end.

Section 6. Operational details to provide

Mews API credentials for read access to the seven properties.
The Slack workspace + the #night-desk channel created in advance.
Property-specific facts (pool hours, breakfast times, parking rules, wifi credentials) compiled into a per-property knowledge base.
The list of GMs and the on-call rotation schedule.
A small (~20-message) sample of the desired voice, past texts from staff that read as "Hearth", for the eval pipeline to calibrate against.

The Strategy Plan

For Sara to take into the franchise pitch and the next ownership meeting. Names the GM-retention thesis, the brand position, the year-one targets, the explicit list of what the agent will not do. The .docx that opens the pitch deck.

Hearth Hotels, Strategy Plan

The after-hours service gap is the boutique-luxury industry's worst-kept secret. Every seven-property chain in the segment is losing guest satisfaction at 2am because no human is at the desk. Hearth solves this with an after-hours concierge agent and turns the solution into the brand-differentiator that supports the franchise pitch.

Opportunity

Seven properties. Average property size 40 rooms. Night-desk coverage runs lean, one person at most properties, none at the two smallest. Inbound after-hours requests average 8 per property per night; ~30% of them go unanswered until morning.

The recovery: ~$340K in retained ADR per year from improved guest satisfaction scores. The hidden upside: a brand-differentiator that opens the franchise pitch deck Sara is preparing for the next eighteen months.

The audience is the after-hours guest, typically a business traveler arriving on a late flight, a family with a sick child, an early-departure guest who needs a 4am wake-up handled differently. They are not the audience the chain markets to; they are the audience the chain measures retention on.

Boutique hotels charge boutique prices because of attentiveness. Attentiveness fails at 2am when nobody's at the desk. Every hour of unanswered request is an hour the brand promise is broken.

Our Position

Hearth is not adding AI to guest service. Hearth is adding the after-hours concierge the chain has never been able to staff at every property.

The conversational-agent category is full of general-purpose chatbots that promise to handle everything and deliver friction at the boundaries. The after-hours-only concierge category is empty. By scoping narrowly, only the hours human service is materially constrained, only the request types where a guest doesn't expect a human voice, Hearth ships a tool whose boundaries are themselves a trust signal.

If the agent ever feels like "a chatbot" to a guest, the strategy has failed. The guest believes they are texting the front desk. They are. The agent is the desk after-hours.

The Strategy

After-hours scope only. The agent answers between 10pm and 7am local time. During the day, SMS forwards to staff. The narrow scope is the differentiator, every line crossed dilutes the trust.

Property-aware tool inventory. The agent can look up reservation details, confirm in-property facts, send supply requests to the night manager, and escalate to the on-call GM. It cannot modify bookings, approve charges, or access other guests. The hard boundaries protect the boutique-luxury position.

Voice as brand. Hearth's brand is hospitality-as-attentiveness. The agent's job is to anticipate the next request, not to mechanically answer the one in front of it. "Can I get you a robe to your room?" after the pool-hours question is the brand-differentiator beat.

Escalation as feature, not failure. The agent never says "I can't help." It frames every escalation as a routing decision. Guests experience "someone is handling it."

Year-one targets.

80% of after-hours requests resolved by the agent without escalation.
Median first-response time under 30 seconds.
Guest-satisfaction NPS improvement of +4 points across the chain.
Zero GM burnout-related departures (the chain lost two GMs in the last 18 months to exactly this work).

The decisions we made.

SMS only. No app, no portal, no kiosk.
After-hours only. The agent does not exist during the day.
Reservation-scoped only. The agent cannot see other guests' data.
Escalation framed as routing. The boundary protects the boutique position.
Brand voice: warm but not chatty. Right. Got it., never Absolutely!

Risks

The single largest risk is feature creep. After launch, every GM will request expanding the agent's capabilities, booking modifications, charge approvals, daytime coverage. Each expansion erodes the after-hours-concierge scope that makes the agent work. Mitigation: a written scope-change policy that requires regional-manager sign-off and a documented rationale tied to the after-hours-scope constraint.

Second risk: brand voice drift. If the eval pipeline doesn't catch the agent shifting toward generic-chatbot patterns, the brand-differentiator dissolves. Mitigation: weekly voice-drift report flagging any response that scored below the threshold against the calibration sample.

Open, whether to extend the agent to the chain's two future properties opening next year, or wait through six months of operational data on the current seven before scaling. Deferred until the first quarter post-launch tells us whether the per-property setup cost is the bottleneck or the eval-pipeline maintenance is.

What’s the gap your team can’t cover?

Already have an account? Sign in