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

I write every message twice.
Spouses decide differently.

Most real-estate decisions are made by two people, and those two people almost never weigh the same things. One falls in love with the kitchen light; the other wants the HOA history. One reads the listing for the lifestyle; the other reads it for the price-per-sqm. The agents who close decision groups consistently aren't writing better messages — they're writing two versions of every message, sent the same hour, calibrated to who actually decides what.

9-min readUpdated April 2026Pack 14 of 30 · @lumi.estate
parallel · murtais 24 · 11:14am both
To Marina (emotional):
  "The afternoon light comes through that
   kitchen window the way you'd expect.
   Lavender's just starting its early bloom.
   Six minutes to the kindergarten your
   sister mentioned. Saturday 11?"

To Hugo (analytical):
  "€/sqm 4,910 (4.5% under area avg). HOA
   stable since 2022. Full reno 2023.
   Comp pull on 3 similar 3-beds before
   Saturday if useful. Viewing at 11?"

Same listing. Same hour. Same Saturday. Two completely different messages. The decision group compares that night and finds the agent addressed both of them.

The split most agents miss entirely.

In any couple buying together, decision-mode tends to split: one partner leads with feeling — the imagined Tuesday afternoon in the kitchen, the walk to the bakery, the way the bedroom opens onto the garden. The other leads with analysis — the price-per-square-metre, the HOA trend, the school district's ranking, the commute math. Most messages from agents address only one of these modes, which is why so many decisions get stuck — one partner is sold, the other has open questions that never got answered.

The parallel-versions protocol fixes this without doubling the agent's workload. The same listing content gets channelled through two prompts, producing two messages — emotional and analytical — sent within the same hour to the two recipients. The agent writes nothing twice; the prompt does the split. The decision group converges that night with both modes addressed.

The close-rate lift in our internal testing (~30 decision groups, EU and LatAm) ranges from 22% to 41% — a meaningful number on a denominator (decision-group buyers) that most agents leak through unaddressed. The variance is mostly down to how accurately the brief identifies decision-mode in the first place; calibration matters more than the prompt.

“One partner falls in love with the kitchen. The other wants the HOA history. Most agents write to one of them and wonder why the deal stalls.”

the protocol

Five rules. Two messages.

The discipline that makes the parallel work. Skip any one and the protocol either reveals itself, lands inverted, or collapses into two paraphrases of the same message.

  1. 01

    Identify the decision-mode split early.

    The split is captured in the call-log soft signals. The partner who pauses at the kitchen window, mentions their sister's school suggestion, asks about light — that's the emotional decider. The partner who asks about HOA history, building age, commute time, comps — that's the approver. Most decision groups have one of each. The minority that don't (two emotional, two analytical) get a different protocol.

  2. 02

    Same content. Different features. Different voice.

    The two messages are not paraphrases. They reference different features of the same listing. The emotional version mentions the lavender; the analytical version mentions the price-per-sqm. The agent's voice within each message also calibrates — warmer for the emotional, more clinical for the approver. This is what makes each recipient feel like the agent gets them specifically.

  3. 03

    Send within the same hour, same channel each.

    Cross-channel asymmetry leaks the protocol — if Marina gets WhatsApp and Hugo gets email, they realise during the evening conversation. Send both via the channel each prefers (which their actual replies have established) and time them within an hour. The decision group converges that night with both messages on hand.

  4. 04

    Never reference the other partner.

    Most failure mode: agent mentions to Marina that 'Hugo also got the comp pull'. Now both partners realise they're getting separate calibrated messages — and the warmth of each message collapses. The discipline: each message stands alone, references only its recipient, contains no acknowledgement of the parallel.

  5. 05

    Decision-mode is not gender. Don't shortcut.

    The biggest trap: assuming the wife is emotional and the husband is analytical. Roughly half the time it's the opposite. The protocol's accuracy depends entirely on capturing decision-mode from actual signals (who asks what, who pauses where) — not from defaults. Get this wrong and the messages land inverted, which feels worse than no protocol at all.

anti-patterns

Three failures that collapse the protocol.

Each one is a real failure mode we've seen in early-adopter agents. Each one is what happens when the protocol's discipline slips on one rule.

the swapped-names paraphrase

[To Marina]: 'Marina, walked Murtais 24 this morning. €/sqm landed 4,910 (4.5% under area avg). HOA stable. Saturday 11?' [To Hugo, identical content]: 'Hugo, walked Murtais 24 this morning. €/sqm landed 4,910 (4.5% under area avg). HOA stable. Saturday 11?'

Same message with the name changed. This isn't the protocol; it's a name-merge bug. Each version must reference different features that match decision-mode. Otherwise the decision group compares notes and finds nothing different — which is worse than not running the protocol.

the cross-reference

[To Marina]: 'Marina — sent Hugo the full data pack. The afternoon light through the kitchen is everything. Saturday 11?'

Names the parallel. Now Marina knows the agent is sending Hugo a different message. The whole credibility of the parallel — that each message was written for them specifically — vapourises. The agent should never reference what the other partner is getting.

the gender shortcut

[Drafting agent's note]: 'Marina is the wife so she gets the emotional version about light and the kindergarten. Hugo is the husband so he gets the analytical version about HOA and comps.'

Wrong half the time. Modern decision groups don't sort by gender, and assuming they do produces inverted messages — Marina gets the version Hugo would have wanted, and vice versa. The brief from call-log signals tells you who asks about HOA and who pauses at light. Use that data, not defaults.

copy · paste

The split input.

What the prompt receives per parallel-message run. Decision-group modes captured from prior call-log signals; voice samples per recipient already calibrated by the agent's prior touches.

spouse_split.yaml
# ── parallel message draft input ─────────────────
listing:
  address:        "Rua dos Murtais 24, Cascais"
  price:          "€825,000"
  beds_baths:     "3 / 2"
  surface:        "168 m²"
  highlights:
    - "south-facing kitchen, sea-glimpse"
    - "8 min walk to Cascais train"
    - "garden 90m², lavender + olive"
    - "garage 2 cars, 1 EV charger"
    - "€/sqm: 4,910 (avg area: 5,140)"
    - "HOA: €245/m, recent renovation 2023"
    - "kindergarten 6 min walk"

decision_group:
  - name:         "Marina"
    mode:         "emotional"
    soft_signals: "loved the kitchen window
                   on first viewing, sister's
                   schools comment, lavender"
  - name:         "Hugo"
    mode:         "analytical"
    soft_signals: "asked about HOA history,
                   commute, building age"

agent_voice_samples:
  emotional_recipient:
    - "Marina — that south-facing kitchen
       still has me thinking. Saturday 11?"
  analytical_recipient:
    - "Hugo — €/sqm landed 4,910 (4.5%
       under area avg). HOA stable since
       2023. Comp pull Saturday morning?"
the prompt that does the split

What to feed Claude.

One prompt produces both versions in one call — the prompt internally branches by recipient mode. Easier than running two prompts (avoids drift across calls) and keeps content references aligned.

spouse_system_prompt.md
You are a senior real-estate agent's
parallel-versions message drafter.

INPUT
You receive: the listing or update content,
the decision-group brief (who's involved
and which mode each person leans toward —
emotional/lifestyle vs analytical/approver),
and the agent's voice samples.

OUTPUT
TWO messages from the same content, each
addressed to one decision-group member.

  EMOTIONAL VERSION (for the lifestyle
  decider — usually the partner who
  initiates the search):
  - Lead with the lived experience:
    light, view, walking distance to a
    place they love, the feel of the room.
  - Use sensory language — "afternoon
    sun", "high ceilings", "walks to
    the bakery".
  - End with a date-anchored next step
    ("Saturday 11am?").

  ANALYTICAL VERSION (for the approver —
  usually the partner who scrutinises the
  decision):
  - Lead with the data: price-per-sqm,
    HOA, recent comp, commute time,
    school rating, building maintenance
    history.
  - Cite sources where applicable
    ("according to the 2025 cadastre").
  - End with a clear data-anchored next
    step ("happy to send the comp pull
    before Saturday").

RULES (non-negotiable)
1. Both messages reference the SAME
   listing/update — same address, same
   price, same date.
2. Each message references DIFFERENT
   features that match its mode. The
   emotional version doesn't mention the
   HOA. The analytical doesn't mention
   the afternoon sun.
3. Both messages send within the same
   hour. Time-asymmetry leaks the protocol.
4. Each message is in the agent's voice
   for that recipient — if the agent
   typically writes warmer to the
   emotional decider and more clinical
   to the approver, those calibrations
   are in the voice samples.
5. Never reference that the other partner
   is getting a different version.

ANTI-PATTERNS (never produce these)
- Identical messages with names swapped
- Mentioning the other partner ("Marina,
   I sent Hugo the comp pull")
- Combining both modes in one message
  (the whole point is splitting them)
- Stereotyping by gender — emotional vs
  analytical is decision-mode, not
  gender-coded

The decision group should compare notes
that night and find that the agent
addressed both their concerns — without
either feeling pandered to.
Open Claude →

Send via the channel each recipient prefers (their actual reply history establishes this). Time the two within the same hour.

split in · two versions out

What Claude returns.

Both messages, side-by-side. Notice that they share the listing reference (Murtais 24, Saturday 11) but differ in features cited and tone.

output · parallel versions
## EMOTIONAL VERSION (for Marina)

Marina —

Walked the Murtais place again this morning.
The afternoon light comes in through that
kitchen window the way you'd expect — and
the lavender in the garden is just starting
its early bloom. Six minutes to the
kindergarten, the one your sister
mentioned.

Saturday 11am for a viewing?

— A.


## ANALYTICAL VERSION (for Hugo)

Hugo —

Pulled the numbers on Murtais 24:

  · €/sqm: 4,910 (Cascais avg 5,140 —
    4.5% under)
  · HOA: €245/m, last increase 2022
  · Building maintenance: full
    renovation 2023 (façade, roof,
    elevator)
  · Commute: 8min walk to train,
    23min total to Tech Park

Comp pull on three similar 3-beds in the
area available before Saturday if useful.
Worth a viewing 11am?

— A.
built around this exact parallel-versions protocol

Splitting the message is step one.
Sending both within the hour 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 decision-mode targeting from B2B sales — complex purchases involve multiple decision-makers with different criteria, and addressing each separately closes deals single-message approaches lose. Our slice: parallel emotional + analytical messages for the buying spouse pair, sent within the same hour.

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