Lumi.
Lumi · private beta · EU · LatAm · MENA

Your second brain
for closing deals.

Speak after a showing. Forward an email. Pull up a client. Lumi captures the soft signals, fills the brief, and feeds Claude — automatically.

  • Voice → CRM, auto. No forms.
  • Works offline. Syncs when you're back.
  • Free for agents in EU · LatAm · MENA.

9-min read · Updated April 2026

9:41

Lumi · Wednesday

Good morning, Niki.

Two showings · three leads need a nudge.

Clara Ruiz
Tomorrow 11am showing at Passeig de Gràcia 84 with Clara Ruiz. She wants to bring her partner.
Got it — creating the showing.
Suggested event · 92%

Showing · Passeig de Gràcia 84

Thu · 11:00–11:45Gràcia
What’s the HOA for Apt 4?
€210 per month, covers elevator, concierge, and rooftop.DOC 12
Ask Lumi or speak…
Calendar
Todos
Lumi
Clients
Settings
agent toolkit · field guide

Voice note in.
Three drafts out.

The follow-up message after a showing is the single highest-leverage communication an agent sends — and the one most often delayed because sitting down to write it from scratch takes 8-12 minutes. The 60-second loop fixes that: speak the memo, get three drafts back in different tones, pick one, send. The whole thing fits between the front door of the property and the first red light on the drive home.

9-min readUpdated April 2026Pack 09 of 30 · @lumi.estate
three_drafts_v1.txt — example output
--- friendly ---
Sofia — quick one. The kitchen window at Rua
da Prata is going to be the standard now,
fairly. I've got two ground-floor + balcony
options pulling together for Thursday — both
in Lapa, both within 10 min walk of the same
café we ended up at. — A.

--- professional ---
Hi Sofia,

Thanks for taking the time at Rua da Prata
yesterday. Noted Carlos's HOA question — I'll
include the figure (280€/mo) in tomorrow's
note. The stairs concern is fair; I have two
ground-floor + balcony options to send
Thursday by 14:00 as discussed.

Best,
Andre

--- urgent ---
Sofia — Saturday 11am viewing slot just
opened on a ground-floor + balcony in Lapa.
Same Estoril-train walking distance.
Want me to lock it before someone else does?

Same client. Same week. Three tones. Pick the one that fits the context — send in 30 seconds.

Why three tones beat one message.

The instinct when AI started writing follow-up messages was to ask for one perfect output and send that. The result was an industry-wide drift toward a single safe-middle tone — warm but not casual, brief but not urgent, ends with a soft “let me know”. Recipients learned to recognise the shape of an AI follow-up within 6 months and reply rates dropped accordingly.

Asking for three drafts in different tones forces the model to commit to actual tonal choices — the friendly draft has to be friendly enough that the professional one feels different. The agent reads all three, picks the one that fits the moment, and the message lands as intentional rather than templated. The same brief, the same memo, three messages — and the agent picks based on context the model doesn't have access to (mood of the showing, recent text-thread tone, whether the spouse just texted in).

The deeper benefit: the three-draft format reveals what the AI is actually doing. When the friendly draft and the professional draft differ only in punctuation, the model is being lazy and you can re-prompt. When they differ in word choice, sentence length, and call to action, you know the model has the brief well enough to vary output meaningfully.

“One AI draft regresses to the safe middle. Three forces the model to commit — and gives you the lever.”

the loop

Four steps. Sixty seconds.

The whole loop fits between the property's front door and the first red light on the drive back. Practiced agents don't think about it — they just speak the memo and the rest happens.

  1. 01

    Voice memo, 60-90 seconds, in the agent's voice.

    Speak naturally as if telling a colleague what just happened. Don't try to structure — the AI will. Cover: what happened, what they cared about, what you committed to next.

  2. 02

    Whisper transcribes; brief is auto-attached.

    The brief for this client (7-field YAML) is pulled from the CRM and prepended to the prompt automatically. The model sees both the new memo and the existing context.

  3. 03

    Three drafts return in 6-8 seconds.

    Friendly, professional, urgent. Same content, three tones. The agent reads, picks one, taps to send — or taps to edit. Total elapsed time from memo end to message ready: under 30 seconds.

  4. 04

    Send. Brief auto-updates with last_touch.

    Whatever was sent gets logged as last_touch in the CRM with timestamp, channel, and outcome. The next reactivation message — if needed — will know exactly where things were left.

three tones · three jobs

When each draft wins.

The three tones aren't synonyms — they map to three different communication contexts. Default-friendly is fine; never-urgent is a leak.

Friendly

whenDefault for WhatsApp/SMS to clients you have an existing rapport with. Most post-showing follow-ups, most check-ins after a casual touch.

why it worksFriendly tone signals continuity of relationship. The recipient reads it as a message from someone they already know, not a formal communication — which means it gets opened and replied to faster.

Professional

whenDefault for email, especially when there's a structured next-step or a financial detail (HOA, bank pre-approval, valuation). Also default for new-client first contact.

why it worksProfessional tone signals competence and creates a paper trail. The recipient files it differently from a casual message — they're more likely to forward it to a spouse, a lender, or an attorney for review.

Urgent

whenWhen a window is closing — viewing slot, offer deadline, lender response window. Use sparingly; the urgent tone burns out if used more than ~once per client per month.

why it worksUrgent tone forces prioritisation. The recipient understands the message is time-sensitive and replies within hours instead of days. Overuse and they stop trusting it; underuse and you lose deals to slower communication.

worked example

Voice memo in. Three drafts out.

A real-feeling 60-second memo on the left. AI's three drafts on the right. No editing applied — first-pass output.

voice memo · transcribed

Just finished with Sofia and Carlos at Rua da Prata. Sofia loved the kitchen window — said "this is where I'd make coffee." The stairs killed it though, twins concern. Carlos asked about HOA. I promised them two ground-floor options with balconies by Thursday. Need to send the follow-up tonight.

three drafts · ai-generated
--- friendly ---
Sofia — quick one. The kitchen window at Rua
da Prata is going to be the standard now,
fairly. I've got two ground-floor + balcony
options pulling together for Thursday — both
in Lapa, both within 10 min walk of the same
café we ended up at. — A.

--- professional ---
Hi Sofia,

Thanks for taking the time at Rua da Prata
yesterday. Noted Carlos's HOA question — I'll
include the figure (280€/mo) in tomorrow's
note. The stairs concern is fair; I have two
ground-floor + balcony options to send
Thursday by 14:00 as discussed.

Best,
Andre

--- urgent ---
Sofia — Saturday 11am viewing slot just
opened on a ground-floor + balcony in Lapa.
Same Estoril-train walking distance.
Want me to lock it before someone else does?
the prompt that writes them

What to feed Claude.

The system prompt that turns voice memo + brief into three tonal variants. Tested against Claude Haiku — generates all three in 6-8 seconds at $0.0008/triplet.

three_drafts_system_prompt.md
You are a real-estate agent's follow-up drafter.

INPUT
A voice-memo transcript (10-90 seconds) recorded
by the agent immediately after a showing or
client conversation. Plus the client's brief
(7-field structured data).

OUTPUT
Three message drafts — exactly three, no more,
no fewer — returned in this format:

  --- friendly ---
  [draft text]

  --- professional ---
  [draft text]

  --- urgent ---
  [draft text]

Each draft is for the SAME upcoming send to the
SAME client. The agent picks one. They differ
in tone, not content.

TONE DEFINITIONS

  friendly
    - 2-3 sentences
    - first-name, no honorific
    - one specific reference to the showing
    - casual sign-off ("— A.", "— Andre")
    - default channel: WhatsApp / SMS

  professional
    - 3-4 sentences
    - full name on first reference, first
      name on close
    - one specific reference + one structured
      next-step with date
    - measured sign-off ("Best, Andre")
    - default channel: email

  urgent
    - 2 sentences
    - names a specific deadline or window
    - one yes/no question with a 24h response
      cue
    - same channel as last contact (continuity)

RULES (non-negotiable for ALL three tones)
1. NEVER fabricate a fact not in the brief or
   the voice memo. If the agent didn't say it,
   the message can't reference it.
2. Every draft references at least ONE soft
   signal verbatim from the brief. Not
   paraphrased.
3. Every draft ends with a concrete next step
   that has a date or time, not "let me know".
4. Match agent.language. Localise greetings,
   sign-offs, calendar formats.
5. NO emoji in any tone. NO exclamation
   marks except mid-sentence ("call!" only
   if quoting somebody).
6. Honor channel conventions: SMS no subject
   line; email always has a subject.

ANTI-PATTERNS (never produce these)
- "Hope you're doing well"
- "I wanted to follow up"
- "Just touching base"
- "Looking forward to hearing from you"
- Any agency-promotional sign-off

Voice: warm, brief, specific. Tone differs
between the three drafts; voice is consistent.
four ways the loop breaks

The failure modes.

Each one is the result of treating the loop as a writing tool instead of a decision tool. The fix is to remember why three drafts beat one.

failure mode

Trying to write the message instead of speaking the memo.

The whole protocol depends on voice in. If you're typing the memo, you're already structuring — and the structure leaks soft signals. Speak the memo, even if you feel awkward. The transcription is what powers the tone variations.

failure mode

Always picking the friendly draft.

If you're using friendly for every send, you're missing the urgency lever when you actually need it. The protocol works because the three tones are reserved for three contexts. Default-friendly is fine; never-urgent is a leak.

failure mode

Editing all three drafts before deciding.

The point of three drafts is fast selection, not editorial review. Read all three, pick one, edit only that one. Spending 5 minutes editing each draft defeats the 30-second loop.

failure mode

Sending without the brief auto-update.

If your tooling doesn't write last_touch back to the CRM after send, you've broken the next step in the chain. The 7-day silent buyer reactivation depends on knowing exactly when last_touch happened.

built around the 60-second three-drafts loop

Speaking the memo is step one.
Picking the right tone is step two.

Lumi is the app that runs this workflow for you. You speak after a showing — Lumi captures the soft signals. You forward an email — Lumi updates the constraints. You open the app at 8am — the brief is already there, ready to feed Claude.

  • Voice → structured CRM, automatically
  • No forms. No data entry. No copy-paste.
  • Free for agents in EU · LatAm · MENA
9:41

Lumi · Wednesday

Good morning, Niki.

Two showings · three leads need a nudge.

Clara Ruiz
Tomorrow 11am showing at Passeig de Gràcia 84 with Clara Ruiz. She wants to bring her partner.
Got it — creating the showing.
Suggested event · 92%

Showing · Passeig de Gràcia 84

Thu · 11:00–11:45Gràcia
What’s the HOA for Apt 4?
€210 per month, covers elevator, concierge, and rooftop.DOC 12
Ask Lumi or speak…
Calendar
Todos
Lumi
Clients
Settings

A real-estate adaptation of the voice-first multi-variant follow-up pattern from operator communities — returning multiple drafts forces the model out of the safe middle. Our slice: 3 tones (friendly / professional / urgent), drafted before the agent walks back to the car.

More guides like this on @lumi.estate. Follow if any of this was useful — it's how we know to keep writing.