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.

10-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

Sunday 7 PM:
the 10-minute pipeline review.

Most agents spend Monday morning rebuilding their week from scratch. Open the CRM, scroll the pipeline, decide who to call. By 11 AM half the day is gone and the wrong calls got made. The agents who close consistently do the deciding on Sunday. AI sorts the pipeline; the agent makes 8 decisions; Monday starts in motion.

10-min readUpdated April 2026Pack 26 of 30 · @lumi.estate
sunday_review_19_00.txt — example output
Pipeline review · week of April 27

HOT — move on Monday
  - Sofia & Carlos Ferreira — Mon 9:30,
    answer Carlos's HOA question (280€/mo)
    and lock Saturday 11am viewing.
  - Diego Almeida — Mon 14:00, send the
    pre-approval intro to Banco Santander
    (he's window 60 days, lender chosen).
  - Aisha Rahman — Tue 10:00, open-house
    follow-up on Lapa property (asked
    about the kitchen, didn't ask about
    schools — bring it up).

COLD — needs reactivation
  - João Silva — silent 11d. Open with
    "saw a place this morning that fits
    your top-floor + view ask" — WhatsApp.
  - Maria Costa — silent 9d. Open with
    "the Estoril building you liked just
    dropped 4%" — SMS.
  - Henrique Pinto — silent 14d. Open
    with "school catchment update for
    Ajuda Primary — moved by 600m" —
    email (his preferred channel).

AT RISK — could lose this week
  - Ana & Tiago Mendes — competitor agent
    showing them Carcavelos properties.
    Rescue: lock the Sintra walking tour
    you promised 3 weeks ago, this Sat.
  - Miguel Almeida — open objection on
    HOA fees unresolved 18 days. Rescue:
    send the comparison vs his current
    HOA (saves 90€/mo) by Tuesday.

8 clients on the board this week.
3 hot, 3 cold, 2 at risk.

Three buckets. Eight clients. Each entry is one specific action. Nothing about “follow up”.

Why weekly reviews die.

Every productivity book recommends a weekly review. Almost no agent does one consistently. The reason is not discipline — it's architecture. The classic weekly review asks the agent to (1) sit down with their CRM, (2) scan every open client, (3) re-evaluate priority, (4) decide actions for the coming week. That's 90 minutes of cognitively expensive work, on a Sunday evening, with no clear stopping point. Of course it dies.

The 10-minute version moves the cognitive expense to AI. The model scans every client, sorts by signal, surfaces the 8 that matter, and drafts the one specific action for each. The agent's job collapses to: read, decide, drag onto Monday's calendar. 10 minutes, bounded surface area, clear stopping point. The protocol holds because the architecture lets it.

The 3-bucket structure is doing most of the cognitive lifting. Hot, cold, risk are not arbitrary categories — they map to three different agent behaviours: execute, reactivate, rescue. Each behaviour has its own playbook. The bucket assignment tells the agent which playbook to run, before they've thought about it.

“Weekly reviews don't fail because agents are lazy. They fail because the format demands an hour of cognitive work in a 10-minute slot.”

the structure

Three buckets. Three behaviours.

Each bucket exists because it prompts a different action. If two buckets prompt the same behaviour, they should be one bucket. If one bucket prompts no clear behaviour, it should be cut.

01

Bucket 1 · HOT — move on Monday

These are the clients where one specific action this week determines whether a deal moves or stalls. The 3-cap forces prioritisation: there are always more than 3, but only 3 of them are truly leverage moves. The rest can wait.

02

Bucket 2 · COLD — needs reactivation

These are clients who went silent on a promised next-step. They're not lost yet — but every additional day of silence makes the recovery harder. The bucket forces you to send the one specific opener that re-opens the conversation, instead of letting them slide into the cold-revisit queue.

03

Bucket 3 · AT RISK — could lose this week

These are deals you might actually lose by Friday if you don't act. The cap of 2 is on purpose: if more than 2 deals are at risk simultaneously, your real problem is upstream (in capacity or response time) and the buckets are a symptom, not the cure.

the sunday protocol

Five steps. Ten minutes.

The whole loop fits between your second pour and the start of dinner. The protocol is designed to be quick because it has to be — anything longer and it competes with the rest of your Sunday and loses.

  1. 01

    Sunday 7 PM. Coffee or wine. Phone, no laptop.

    The protocol works because the trigger is a specific time on a quiet evening. Not Friday afternoon (you're tired), not Monday morning (too late). 7 PM Sunday lands when most agents are doing nothing useful with their week-prep window. Phone is enough — the brief is short.

  2. 02

    AI generates at 6:55 PM. Pushed to your inbox.

    The cron runs 5 minutes before you sit down. You don't generate — you read. The whole protocol depends on the brief being there waiting; if you have to trigger it manually, you'll skip the trigger half the time.

  3. 03

    Read top to bottom. Make 8 decisions. Total: 10 minutes.

    Each entry is one decision: am I doing this Monday, or am I delegating, or am I dropping it? No re-analysis, no second-guessing. The work of analysis already happened upstream — the brief is decision-prompting, not analysis-prompting.

  4. 04

    Drag each accepted action onto Monday's calendar.

    If the action lives only in the brief, it dies in the brief. Each accepted action becomes a calendar event with a time, a person, and the one specific thing to do. The brief is the source; the calendar is the operating system.

  5. 05

    Close the brief. Stop thinking about work until tomorrow.

    This is the hidden benefit. The reason the protocol holds long-term is that it gives the agent permission to stop thinking about pipeline. Once the 8 decisions are made, the rest of Sunday is yours. The 3-bucket structure works partly because it bounds the surface area you have to think about.

four ways the protocol breaks

The failure modes.

If you try this protocol for two weeks and it stops sticking, it's almost always one of these four.

failure mode

Treating it as a comprehensive review.

The brief is intentionally incomplete. It surfaces 8 of your 30 active clients, not all 30. If you find yourself thinking 'but what about Maria?' the answer is: she's not on the board this week. Trust the prioritisation. The review next Sunday will surface her if she belongs.

failure mode

Editing the AI's bucket assignments.

The temptation is to move a client from cold to hot because you feel like it. Resist. The AI's classification is based on the data — last touch, days silent, missed promises. Your gut feeling is often the same data, processed less rigorously. Override sparingly, and only when you have specific knowledge the data lacks.

failure mode

Reading it Friday instead of Sunday.

Friday review feels productive but reads with a tired week behind it. The actions get diluted by 'I'll think about it over the weekend' deferrals. Sunday is the right trigger because Monday is right after, and you can't defer the actions any further.

failure mode

Skipping the calendar drag step.

The single biggest failure mode. The brief identifies the 8 actions; the calendar is what makes them happen. If accepted actions live only in your head or in the brief itself, the protocol degrades to a Sunday-evening worry session. The calendar is the bridge.

the prompt that writes it

What to feed Claude.

The system prompt that turns the full pipeline snapshot into the 3-bucket review. Tested against Claude Sonnet — the prioritisation logic benefits from the larger model on this one (Haiku tends to over-fill the buckets).

weekly_review_system_prompt.md
You are a real-estate agent's weekly pipeline
analyst.

INPUT
You will receive an array of all active clients
in the agent's pipeline. For each:
  - name, intent_stage, window_earliest,
    window_latest, last_touch.at,
    last_touch.next_promised, days_since_touch,
    open_objections, soft_signals (latest 3),
    listings_shown_count, properties_offered_on
  - any open todos linked to this client
  - any conflicts on the agent's calendar
    that block the next promised step

OUTPUT
A 3-bucket review, plain text, designed to be
read at Sunday 7 PM in 10 minutes. Each bucket
holds at most 3 clients (cap matters — see
rules).

STRUCTURE — exactly 3 buckets, in this order:

  Bucket 1 · HOT — move on Monday.
    Up to 3 clients where the next concrete
    action is overdue or due this week, AND
    intent_stage is serious or urgent. For
    each: name, the one action, the deadline.

  Bucket 2 · COLD — needs reactivation.
    Up to 3 clients where days_since_touch ≥ 7
    AND intent_stage is browse or serious AND
    last_touch.next_promised was missed. For
    each: name, the one specific opener
    (referencing a soft_signal), the channel.

  Bucket 3 · AT RISK — could lose this week.
    Up to 2 clients where there's an open
    objection unresolved >14 days, OR a
    competitor agent is showing them properties,
    OR the agent missed a promised next-step
    >7 days ago. For each: name, the risk,
    the rescue move.

RULES (non-negotiable)
1. Cap each bucket strictly. If there are
   more than 3 hot, pick the 3 highest by
   (window_urgency × intent_stage). The
   point is forced prioritisation, not
   completeness.
2. Each entry is ONE concrete action. Not
   "follow up" — "call Sofia at 9:30 Mon
   re: HOA fees".
3. Sort within each bucket by when the
   action must happen (earliest first).
4. If a bucket has zero entries, say so
   explicitly. "No hot moves needed
   Monday" is a valid output and a real
   signal.
5. End with a 1-line week-shape summary:
   "8 clients on the board this week.
    3 hot, 3 cold, 2 at risk."

ANTI-PATTERNS (never produce these)
- A bucket with 5+ clients (lose
  prioritisation; defeats the purpose)
- Generic actions like "check in" or
  "send some listings" — every action must
  be specific
- A separate "FYI" or "watch" bucket — the
  three buckets exist because they each
  prompt different behaviour. Don't add
  more.
- Editorial commentary ("things look great
  this week!"). Just the buckets.

Voice: clinical, brief, decision-prompting.
The agent is reading this with their feet up
on a Sunday — they need clarity, not
analysis.
Open Claude →

Paste the prompt above as a system message. Feed in your full pipeline snapshot as the user message. Set up a Sunday 6:55 PM cron once and forget it.

what changes after 8 weeks

The compounding effect.

None of these numbers move in week 1. By week 8 the entire shape of the agent's week is different.

metric
before protocol
after 8 weeks
Time spent in CRM Monday morning
60-90 minutes scrolling, deciding, switching contexts
0 minutes — the decisions were made Sunday
Clients you actively touched this week
12-15 (the loud ones, the recent ones, the easy ones)
8 (the leverage ones, AI-prioritised)
Deals lost to silence (>14 days untouched)
3-5 per quarter — the ones who quietly slipped
≤1 per quarter — the AT RISK bucket catches them at week 2
Reply rate to your reactivation messages
8-12% (template-driven)
30-45% (specific opener per cold-bucket entry)

The metric that moves slowest but matters most: deals lost to silence. The protocol exists because the AT RISK bucket catches the deals that would otherwise quietly slip — and once an agent goes a full quarter without losing one of those deals, the protocol becomes load-bearing. They don't skip Sundays anymore.

built around the Sunday-night pipeline review

The brief is step one.
Showing up Sunday at 7 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 David Allen's GTD weekly review and the agency-ops playbook. The 2026 shift: AI surfaces and prioritises, the agent only decides. Our slice: Sunday-night pipeline triage in 10 minutes over coffee.

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