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 7-field brief, and feeds Claude — automatically.

  • Voice → 7 fields, auto. No forms.
  • Works offline. Syncs when you're back.
  • Free for agents in EU · LatAm · MENA.
Join the waitlist →Read the field guide

11-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

Your CRM is not a database.
It's an unfinished prompt.

Most real-estate agents treat their CRM like storage. Names, phone numbers, a checkbox or two. The agents who close more in 2026 treat it differently — as the context layer their AI draws from every time it writes a message, books a showing, or preps a meeting. This is the 7-field brief they're using.

11-min readUpdated April 2026From the carousel · @lumi.estate
client_brief.yaml — preview
intent_stage:     "serious"
window_forcing:   "school start + lease end"
hard_constraints:
  - "3+ bedrooms (twins + nursery)"
  - "<40 min commute, ground floor"
soft_signals:
  - "paused at the kitchen window"
budget:           450k → 520k EUR
last_touch:       2026-04-20  ·  WhatsApp voice

The 7 fields below — what to write, why, and the Claude prompt that reads them.

The reframe.

Every empty field in your CRM is a missing ingredient in the prompt your AI runs every time you ask it to draft a message, summarise a deal, or decide what's next. AI doesn't fail because the model is weak. It fails because the brief is incomplete.

The shift happening in 2026 is quiet but consequential: top agents have stopped treating the CRM as where data goes to die and started treating it as the prompt that runs the day. They fill it in obsessively, with specific language, because they know every line they write becomes a line their AI quotes back to them in 48 hours.

The 7 fields below are the minimum viable context layer. None of them require a new tool — every modern CRM has a notes field. What changes is what goes inside.

“The agents who win in 2026 don't have better CRMs. They have briefs that AI can actually read.”

the brief

The 7 fields.

Each one with what to capture, the format that works, why it matters, the most common mistake, and what it unlocks for your AI. Read in order — they build on each other.

01
field 1 / 7

Intent stage

Where this person actually sits on the ladder from window-shopping to must-move. Not a vibe — a discrete level with behavioural markers.

what to capture

One of three levels — Browse, Serious, Urgent — plus the marker that put them there. Re-evaluate after every meaningful contact (it changes).

format & example
// 3 levels
browse   → no timeline, vague reqs,
           Saturday-habit replies
serious  → has a checklist, replies
           within 24h, asks process Qs
urgent   → forcing function exists
           (lease end, job, divorce)

// example
intent_stage: "serious"
intent_marker: "lease ends 2026-09-30,
              has saved 4 listings already"
why it matters

Intent drives every other decision: tone, cadence, channel, how many listings to send, whether to push for pre-approval today. Treating a Browse client like a Serious one burns out your follow-up cadence; treating an Urgent client like a Serious one loses them to whoever called back first.

common mistake

Marking everyone as 'serious' because it's the flattering assumption. In reality ~60% of new leads are Browse-mode. Be honest. Browse-mode clients are valuable — they convert in 6-18 months — but only if you don't burn them out with weekly nudges.

what this unlocks for AI

Claude can write three completely different message tones (informative for Browse, decisive for Serious, calendar-first for Urgent) — but only if it knows which one you're talking to. Without this field, every AI draft regresses to the safe middle and reads as generic.

02
field 2 / 7

Time-to-move window

Two dates and a forcing function. Not 'soon', not '6 months' — actual edges of the window and the reason it exists.

what to capture

Earliest possible move date, latest possible move date, and what's forcing the window (school start, lease end, baby due, transfer). All three.

format & example
window_earliest: 2026-07-01
window_latest:   2026-10-15
window_forcing:  "school start + lease ends 09-30"
why it matters

Determines urgency of showings, when to push for pre-approval, when to stop sending new inventory, and when to start a follow-up cadence around 'we know your window is closing'. Without dates, AI defaults to vague urgency that reads as pressure.

common mistake

Capturing only the latest date. The forcing function predicts behaviour — clients with a job-start trigger move 2x faster than clients with a 'we'd like to' trigger, even with the same end date.

what this unlocks for AI

Claude can write 'we have 8 weeks before your lease ends — here's what we should hit by week 4' messages. Specific, calendar-aware, low-pressure. The window field is what makes the math possible.

03
field 3 / 7

Hard constraints

Non-negotiables only. The list of things that, if missing, will kill the deal — not a wish-list of 'nice to haves'.

what to capture

3-7 items max. Each one with the reason. The reason is what tells Claude (and you) when a near-miss listing is worth pushing.

format & example
hard_constraints:
  - "3+ bedrooms (twins + nursery)"
  - "<40 min commute to Canary Wharf"
  - "ground floor (wheelchair parent)"
  - "no listed building (renovation needed)"
why it matters

An AI draft that recommends a 2-bed duplex to a family of four is an instant trust-killer. The hard-constraint list is what filters the listing universe before AI even drafts the message. Get this wrong and every recommendation feels off.

common mistake

Mixing hard constraints with preferences. 'Would prefer south-facing' is not a hard constraint. The test: would they walk away from an otherwise perfect place because of this? If no, it's a preference. Move it elsewhere.

what this unlocks for AI

Claude can rank inbound listings against the constraint list before you see them, surface only the matches, and explain near-misses with honest framing ('this one fails on commute by 12 min — worth it because of the garden?'). Saves ~2h/week on listing triage.

04
field 4 / 7

Soft signals

What they said or did that revealed underlying motivation. The thing you'd tell your partner about at dinner. This is the highest-signal field on the brief.

what to capture

Verbatim quotes when possible. Body-language observations. Reactions to specific rooms or features. Things they asked twice. Things they paused on. Things they told you about themselves that they didn't have to.

format & example
soft_signals:
  - "paused at the kitchen window —
     said \"this is where I'd make coffee\""
  - "both spouses looked at each other
     when we saw the spare bedroom"
  - "asked twice if neighbours had kids"
  - "told me her mother lives in Estoril"
why it matters

This is the field that separates 'competent assistant' from 'that agent gets me'. When your follow-up references the kitchen window, the buyer feels seen — not sold to. Most agents don't write this down because it feels too vague or too personal. That's exactly why it works when you do.

common mistake

Skipping it because 'I'll remember'. You won't. Three showings later, the soft signals from showing 1 are gone. Voice-note them on the walk back to your car — 30 seconds, before context decays.

what this unlocks for AI

Claude turns soft signals into specific reactivation lines: 'Saw a place this morning with the same kind of kitchen window you liked at the Gaspar showing — want me to pull it?' Specific. Personal. Not about the house. The 7-day silent buyer plays this card best.

05
field 5 / 7

Budget — comfortable + stretch

Two numbers, not one. The realistic ceiling and the absolute ceiling. Plus the conditions under which they'd stretch.

what to capture

What they said when you first asked, what they actually approved with the bank, and what they'd pay if a place hit specific notes (garden, top floor, original details). The middle number alone is misleading.

format & example
budget_comfortable: 450000
budget_stretch:     520000
flex_conditions:
  - "south-facing garden"
  - "more than 60m² interior"
  - "no major renovation needed"
currency: EUR
financed: true (mortgage pre-approved)
why it matters

Most clients quote ~20% below their stretch on the first call. If you only capture that number, you ghost them out of properties they would have bought. Two numbers + conditions lets you send 500-530k listings with honest framing: 'slightly above the comfortable line, but I think it's worth a look because it hits two of your stretch conditions.'

common mistake

Capturing only the first number they said. Or treating the stretch number as a secret. Or losing the conditions — you remember the price ceiling but forget that it only stretches if there's a garden.

what this unlocks for AI

Claude can write 'slightly above your comfortable line but here's why' messages that are persuasive without pushy. Without the conditions, every above-budget listing recommendation lands as 'agent doesn't respect my budget'.

06
field 6 / 7

Decision group

Who actually decides — not who you talk to. The shape of the decision, the people involved, and how they weigh in.

what to capture

One of: solo · couple-joint · couple-one-decides · family · guided (parents helping first-time buyer) · committee. Plus the names and roles. Plus the tell — what's the signal that the real decider has approved?

format & example
decision_group:
  shape: "couple-one-decides"
  primary: "Sofia"
  approver: "Carlos (financial sign-off)"
  tell: "Carlos always asks about
        commute and HOA fees last —
        when he stops asking, it's done"
why it matters

Decides whom to cc on the first email, whose schedule you optimise for showings, whose questions you preempt, and which spouse's hesitation kills the deal. Most agents assume husband = decider in hetero couples. Wrong 60%+ of the time. Default to 'unknown' and watch.

common mistake

Tailoring everything to whoever is the loudest in the conversation. The loud one is often not the decider — they're the buffer. Watch for who the loud one looks at when a question lands.

what this unlocks for AI

Claude can write parallel versions of the same update — one optimised for the primary's emotional language, one optimised for the approver's analytical language. Sent at the right time, to the right inbox.

07
field 7 / 7

Last meaningful touch

The one field that gets the least attention and matters the most for AI. Timestamp + medium + content + outcome + the next promised step.

what to capture

When you last actually moved the relationship forward (not 'sent listings' — those don't count). What you said. What they said back. What you committed to do next, with a date.

format & example
last_touch:
  at:       2026-04-20 14:30
  channel:  "WhatsApp voice note"
  content:  "follow-up after Rua da Prata
             showing — flagged the stairs"
  response: "loved balcony, worried about
             second-floor stairs for kids"
  next:     "send 2 ground-floor + balcony
             options by Thursday 14:00"
why it matters

This is the context for every future message. AI without it writes generic. AI with it writes 'wanted to follow up on the balcony question — here are two ground-floor options I think you'll like'. The whole follow-up game runs on this field.

common mistake

Logging 'sent follow-up' without the content. Or logging the content without the outcome. Or — most common — logging the exchange but not the next step you committed to. That's the breach of trust agents don't realise they're committing.

what this unlocks for AI

Claude can write the next message in the thread, in your voice, that picks up exactly where you left off. The 'I'll send you 2 thoughts tomorrow at 10' commitment becomes a calendar reminder, a draft message at 9:55, and a sent follow-up at 10:00 sharp.

copy · paste

The template.

Drop this into a Notion page, a Follow Up Boss custom-fields block, a kvCORE note, or a Google Doc per client. Field order matters — Claude reads top to bottom.

client_brief.yaml
# ── client brief · the 7 fields ───────────────────────
client:           "Sofia & Carlos Ferreira"
ig_or_referral:   "ref from Anna G., Q1 2026"

# 1 — INTENT STAGE
intent_stage:     "serious"
intent_marker:    "lease ends 2026-09-30,
                   has saved 4 listings"

# 2 — TIME-TO-MOVE WINDOW
window_earliest:  2026-07-01
window_latest:    2026-10-15
window_forcing:   "school start + lease end"

# 3 — HARD CONSTRAINTS
hard_constraints:
  - "3+ bedrooms (twins + nursery)"
  - "<40 min commute to Canary Wharf"
  - "ground floor (wheelchair parent)"
  - "no listed building"

# 4 — SOFT SIGNALS
soft_signals:
  - "paused at kitchen window —
     'this is where I'd make coffee'"
  - "asked twice if neighbours had kids"
  - "Sofia's mother lives in Estoril"

# 5 — BUDGET BAND
budget_comfortable: 450000  # EUR
budget_stretch:     520000
flex_conditions:
  - "south-facing garden"
  - "more than 60m² interior"
  - "no major renovation needed"

# 6 — DECISION GROUP
decision_group:
  shape:    "couple-one-decides"
  primary:  "Sofia"
  approver: "Carlos (financial sign-off)"
  tell:     "Carlos asks about commute &
             HOA fees last — when he
             stops asking, it's done"

# 7 — LAST MEANINGFUL TOUCH
last_touch:
  at:       2026-04-20 14:30
  channel:  "WhatsApp voice note"
  content:  "post-Rua-da-Prata follow-up"
  response: "loved balcony, worried about
             stairs for the kids"
  next:     "send 2 ground-floor + balcony
             options by Thu 14:00"
the prompt that uses it

What to feed Claude.

This is the system prompt that turns the brief into useful output. Tested against Claude Haiku and Sonnet — works on either. Feed the brief as the user message; ask the question at the end.

system_prompt.md
You are a senior real-estate agent's second brain.
Your inputs: a 7-field client brief.
Your outputs: short, specific, in the agent's voice.

RULES (non-negotiable)
1. Never invent facts not in the brief.
   Mark uncertain inferences as [unverified].
2. Match tone to intent_stage:
   browse  → informative, low-pressure
   serious → decisive, calendar-aware
   urgent  → calendar-first, short
3. Hard constraints are filters, not preferences.
   If a recommendation breaks one, name it explicitly
   ("this fails on commute by 12 min — flagging").
4. Reference at least one soft_signal in any
   client-facing message. Specific > generic.
5. Budget framing:
   - within comfortable: no caveat needed
   - between comfortable and stretch: name the
     stretch and which flex_condition it hits
   - above stretch: don't recommend unless the
     agent override-flagged it
6. Address messages to the decision_group.primary.
   When the answer requires the approver, say so.
7. Every output ends with the next concrete step,
   with a date or time. No "let me know".

Default response shape:
  - 1 line: framing
  - 2-4 lines: substance
  - 1 line: next step with date/time

Voice: warm, brief, no hype, no emoji.
Open Claude →

Copy the prompt above, open Claude, paste into a new chat as a system message, then send the brief as your first user message.

same client · same week

Before vs. after the brief.

Same buyer. Same week. Same prompt: “draft a follow-up after the showing”. The only thing that changes is what AI has to work with.

before · empty CRM

“Hi Sofia! Thanks again for taking the time to see the apartment yesterday. It was a beautiful property, wasn't it? Let me know if you have any questions and we can definitely look at more options. Looking forward to hearing from you!”

Generic. Not bad — just unmemorable. The buyer reads it as a template and doesn't reply for 9 days.
after · brief filled

“Sofia — quick one. The Rua da Prata stairs were on your mind, fairly. I've pulled two ground-floor options with balconies (one of them in the same area as your mother's place in Estoril, by the way). Slightly above the comfortable line on the second one, but it has the south-facing garden you mentioned. Sending both Thursday by 14:00 like I said. — A.”

Specific. References two soft signals. Honest about budget. Names the next step. Buyer replies in 90 minutes.
if you only have 5 min

The minimum viable brief.

You don't have time to fill all seven for every client today. Fine. The 80/20 is three:

  1. 1

    Intent stage

    Without it, every AI message tone is wrong. Spend 30 seconds on this one alone.

  2. 2

    Window + forcing function

    The math your AI uses to write urgency-aware copy. Two dates and a sentence.

  3. 3

    Last meaningful touch (with next step)

    Without this, you're starting every conversation from scratch. Even if the field has nothing else, this one carries the relationship.

Add fields 3-6 over the next two weeks. By the end of the month every active client has a complete brief, and you've stopped writing follow-ups from a blank page.

the weekend exercise

Retrofit your top 20.

One sitting. ~90 minutes. Open your top 20 active clients and fill the brief — what you remember, marked [from memory] where you're inferring.

Two things will happen:

  • About a third of the clients you thought were “waiting” will reveal themselves as Browse-mode. Stop nudging them weekly. Move them to a slow cadence (one quality touch every 6-8 weeks). You'll get hours back.
  • About a quarter of them are more urgent than you remembered — there's a forcing function in their window you didn't capture. Call them today. Some are 6 weeks from a deal you didn't know was happening.
  • The rest will surface as “I have no idea what's actually going on with this person”. Reach out with one specific question — not a generic check-in. The brief will tell you which question to ask.

That single weekend tends to shift more deals than three months of new-lead sourcing.

built around this exact workflow

Reading this template is step one.
Living inside it is step two.

Lumi is the app that is this template. You speak after a showing — Lumi captures the soft signals. You forward an email — Lumi updates the constraints. You pull up a client — the brief is already there, ready to feed Claude.

  • Voice → 7 fields, automatically
  • No forms. No CRM data entry.
  • Free for agents in EU · LatAm · MENA
Join the waitlistSee how it works
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

Adapted from a thesis circulating in the vibe-marketing community in early 2026 (Greg Isenberg, Alex Finn, MarTech) — the idea that workflows collapse into prompts and prompts get better with context. Translated to real-estate agent reality.

Want the carousel version of this brief? @lumi.estate on Instagram. Comment PROMPT on the post and we'll DM you.