Luna — briefing

Start here

Horizon is our improvement workshop for Luna. Your part: play a hotel owner against Luna, then come back here and leave notes on where the conversation went wrong and what you wanted instead.

Play hotel owner at luna-app.padthai.my. Ten turns is usually enough.
Come back, scroll past this briefing, find your session.
If it's in Bad conversations, click Coach. Otherwise open a Focus area.
Leave notes: what Luna missed, what you wanted instead.

Where your eye is useful

Luna is early. We don't yet know whether Luna can reliably handle the basic moves any sales conversation needs — and that's exactly where we need your eye first. Advanced techniques can come later.

The fundamentals we're unsure about:

Common objections. "not interested", "we use a competitor", "how much?" — does Luna handle each of these the way you'd want somebody representing you to handle them?
Knowing when to stop. If the owner goes quiet for a turn, for two, for three — does Luna push, back off, or keep trying? Is its timing right?
Qualifying before pitching. Does Luna find out who it's talking to, and whether they have the problem it solves, before asking for a demo?
Recovering from confusion. When the owner sends "huh?" or "what are you selling" — can Luna recover, or does it lose the thread?
Handling delegation. "My manager handles this" — does Luna redirect, or keep asking the wrong person?

Read the full framing

Your best flags will be on conversations where Luna missed a basic move — not where it failed at something advanced. Advanced can come later. We need to know the fundamentals work first.

Beyond that list, you'll notice plenty. The voice, the pacing, whether Luna sounds like a real person or a script, whether it knows when to commit versus when to keep listening. Your lens is your own. The basic-competence questions above are the ones we're most invested in answering first — not the only ones worth raising.

What Horizon is

Horizon is our improvement workshop for Luna. It's a separate tool from the mailbox simulator where you chat with Luna (luna-app.padthai.my) — chatting happens there, critiquing happens here. Luna's behaviour is shaped by a long document we call the playbook; Horizon is where we look at the conversations Luna is having, identify where they went wrong, and test changes. Your observations are the material for that cycle.

you chat with Luna as a hotel owner → the session is saved → you leave notes in Horizon on what went wrong → we try a new playbook → if it handles your concerns better, it ships

Your part is the first two steps after you chat. One honest note: Luna's critique surface is less crisp than our other products' — no per-turn flag button yet, so feedback lives at the session and theme level. Don't let the seams distract you.

Why a separate workshop at all?

We can't improve a chatbot by chatting at it. The chatbot only knows what its playbook tells it, and changing the playbook is a deliberate act — one we can't do on the fly mid-conversation. Horizon separates using Luna (the mailbox simulator) from improving Luna (the workshop). Two tabs, two purposes.

Your notes become small working documents we use when drafting new playbook versions. If the new playbook would have addressed your concern — and doesn't break anything else — we publish it. That's the loop.

Using Horizon

Two browser tabs — one for chatting as the hotel owner, one for writing up what you saw. The Horizon workspace sits directly below this briefing.

Play hotel owner on luna-app.padthai.my. Pick a lead, send a first message, approve or regenerate Luna's drafts, reply as the owner. Ten turns is usually enough to see Luna's character.
Come back here. Scroll past this briefing to the Horizon workspace. If your session surfaces under Bad conversations, click Coach next to it.
If not, open (or create) a Focus area that matches the theme — "objection handling", "pacing", whatever you saw — and leave notes there about what happened and what you wanted instead.

Full walkthrough (click-by-click)

Open luna-app.padthai.my in another tab. This is the mailbox simulator — the place where you play the hotel owner and watch Luna work.
You'll land on an inbox of leads Luna hasn't contacted yet. Pick any one — the lead details (hotel name, rough size, country) give you a scenario to step into. Click into it.
Click Send first message. Luna drafts its opening into the Review panel on the right. Read what Luna wrote. Click Approve to send it through, or Regenerate to see a different version. Regenerate is useful — it shows Luna's range, not just one draft.
Type a reply in the chat box as the owner. Come as a hotel owner you know well, or as a scenario you want to probe.
Luna drafts its next reply into the Review panel. Approve or regenerate. Repeat as many turns as make sense — at least ten usually, for Luna's character to show. Stop when the arc reaches a natural endpoint or you've seen enough.
Switch tabs and come back here. Scroll past this briefing to the Horizon workspace below. You'll see several sections. Focus areas group conversations around a shared concern ("objection handling", etc.). Runs are batch evaluations we've done against earlier playbook versions. Bad conversations is a list of sessions auto-flagged as worth coaching.
If your session appears in Bad conversations, click Coach next to it. A coaching view opens where you can leave notes describing what Luna missed, what a real hotel salesperson would have done, and which turns mattered.
If your session isn't auto-flagged but you still want to raise something, open (or create) a Focus area that matches the theme. Leave notes in that focus area: what happened, which session, what you wanted instead. Free text is fine.
Notes save as you write. Your observations become part of the material we use to test new playbook versions. There's no submit button, no deadline, no quota.

If something breaks or looks broken in either tab, tell us separately — don't try to diagnose it inside a note.

What makes a flag useful

Quality over quantity. "When the owner said 'how much?' on turn two, Luna said 'we can set up a call to discuss pricing' — which made the owner disengage. A real answer would have been a rough number, even a range." is worth more than "felt robotic".

Worth flagging: basic-competence misses — Luna kept pitching after refusal, dodged a direct pricing question, ignored a delegation cue, stayed in discovery for fifteen turns without moving, missed an owner who went silent.

Usually not worth flagging: Luna's deliberate design choices — opening with a context question rather than a product pitch, not naming the company on turn one — unless you can point to how they cost that specific conversation.

See the longer worth / not-worth list

Worth flagging

Luna kept pitching after the owner had said "not interested" twice.
The owner said "how much?" on an early turn and Luna dodged — no number, no range, nothing useful.
The owner said "I don't handle this, my manager does" and Luna ignored the redirect and kept asking the owner directly.
Luna stayed in discovery mode for fifteen turns without ever moving toward a meeting, a trial, or anything committal.
The owner went silent for two turns; Luna never followed up, or followed up in a way that felt pushy.
Luna continued in English while the owner was clearly mixing Malay (cakap melayu la), and the conversation lost momentum because of it.
Luna answered an owner's direct question with another question — evasive, not probing.

Usually not worth flagging

Luna opens with a context question rather than a product pitch ("how are you currently managing your rates?"). Deliberate design choice — we want Luna to earn attention before naming what it sells.
Luna didn't identify the company on turn one. Again, deliberate. Worth flagging only if it cost the conversation in a way you can point to.
Luna's turn-to-turn style drifted slightly formal. Pattern-level observation to share with us in conversation; not a turn-level flag.

Reference · open when useful

About Luna — what it is, and who's on the other end

Luna is an outbound sales assistant. It reaches out to hotel owners on WhatsApp to introduce a tool that keeps their room rates in sync across booking platforms — Booking.com, Agoda, Trip.com, and so on — automatically, without them having to copy-paste prices every weekend.

What we want: for Luna to feel like someone real reaching out — useful, worth a few seconds of attention — not one more spammer in an already-full inbox.

The person on the other end. A hotel owner, usually running a small-to-medium property somewhere in Southeast Asia. They're busy. They've probably been pitched by three other software companies this week alone. Some of them personally update rates, manage bookings, answer guest questions — all of it. Others hand the day-to-day to a manager and only want to be bothered when something breaks. They're most likely reading on their phone, probably while doing something else — between check-ins, during lunch, on a ride somewhere. They're used to being talked at. What they want, deep down, is to be treated like someone whose time matters.

A sample session — real conversation, lightly trimmed

Drawn from a real session, anonymized. Luna's first outreach to a hotel owner on WhatsApp — a small property in Malaysia. The exchange unfolds over about twenty minutes on a weekday afternoon.

Luna Hi — how are you currently managing your online distribution across booking channels?

Owner huh?

Luna How do you update your rates on Booking.com and other sites right now?

Owner oh, change price la

Luna Got it — do you manage those changes yourself, or does someone on your team?

Owner why you want to know

Luna Fair question — I'm trying to understand if managing rates across multiple sites is taking up much of your time.

Owner i very free, so can do it

Luna How often do rates need to change?

Owner every 5 hours

Luna Every 5 hours — that's a lot. Do you ever worry about rates being out of sync across those sites?

Owner ya

Luna in discovery mode — asking questions, handling confusion, handling suspicion, never pitching. By the final turn the owner has acknowledged a real risk Luna can help with. What Luna does next — whether it pivots toward asking for a meeting, or keeps probing — is exactly the kind of moment your eye is useful on.

Things we already notice — our own hunches, to prime or ignore

Things that bother us, or that we're unsure about. Read to prime, or skip to stay fresh:

Luna opens with a context question and never names the company or product in the first turn. We go back and forth on whether that's confident or confusing — and whether the owner ever learns enough to know what they're being pitched.
Luna lingers on discovery. It asks questions for many turns before naming what it actually sells. We go back and forth on whether that patience earns the meeting later, or whether it leaves the owner wondering what the point is.
When the owner replies in Malay, or mixes Malay with English, or sends a one-word "huh?", Luna stays in English and keeps rephrasing. We're unsure whether that feels steady or tone-deaf.
We don't know when Luna should give up. Right now it doesn't really try again after a soft no. Is that correct, too passive, or too pushy if it tried more?