I Let Claude Run My Discovery Calls for a Week—Here’s What I Learned

I almost lost a client last February because I forgot to ask him one question during our first call. Not a minor question — the one about timeline. He wanted to relocate from Lisbon to Madeira by April, and I spent three weeks showing him properties in the wrong price range because I assumed he had flexibility. He didn’t. That mistake cost me roughly 12 hours of wasted work and very nearly cost me the commission.

That’s when I started taking client discovery seriously as a process, not just a conversation. And that’s eventually what pushed me to try something I wasn’t sure would work: letting Claude handle a significant part of my discovery workflow for a full week.

This isn’t a review of Claude’s general features. I’ve written about that elsewhere. This is specifically about what happened when I used Claude’s tools — particularly its ability to process information, generate structured questions, and analyze responses — to run my client intake process across seven consecutive business days in March 2026. Here’s exactly what I did, what worked, and where it fell apart.

The Problem I Was Trying to Solve

My discovery calls were inconsistent. After 14 years in Madeira real estate, you’d think I’d have a locked-in system. I don’t. Or I didn’t. The problem with doing everything yourself is that your “system” is really just muscle memory — and muscle memory gets lazy when you’re tired, distracted, or on your fourth call of the day.

I was missing information. I’d finish a call, open a new property search, and realize I hadn’t confirmed whether the client needed parking, or whether they were buying in their own name or through a company (which matters enormously for Golden Visa eligibility). Small things. But in real estate, small things become expensive problems.

I needed a consistent discovery framework. And I wanted to see if Claude could help me build and run one — not as a chatbot talking to my clients directly, but as a tool I used before, during, and after each call to make sure nothing fell through the gaps.

What I Actually Set Up Before the Week Started

I spent about two hours on a Sunday evening building what I called my Discovery Stack. Three components:

1. A Pre-Call Briefing Prompt

I gave Claude everything I knew about an incoming client before each call — their name, where they found me, any email threads we’d had, what they said they were looking for. Claude’s job was to generate a tailored set of discovery questions specific to that person, ordered by priority, with notes on why each question mattered.

For a client coming from the UK asking about NHR tax regime implications, the questions Claude surfaced were different from those it generated for a Portuguese family relocating internally. That customization was the whole point.

2. A Post-Call Debrief Template

Immediately after each call, I typed a rough brain dump into Claude — everything I remembered, in whatever order it came out. Claude then structured it into a client profile: budget confirmed, timeline, must-haves, dealbreakers, decision-making process, who else is involved in the decision, and any red flags I noted.

3. A Gap Analysis Check

This was the most useful piece. I had Claude compare my post-call debrief against a master checklist of 22 things I should know before starting a property search for any client. It would tell me exactly which items were still missing and suggest how to get that information — either by sending a short follow-up email or asking at the next touchpoint.

Setup cost: $20 on my Claude Pro subscription, which I already pay monthly. Zero additional spend. Time to set up the prompts: about 2 hours total.

My Real-World Experience Running This for 7 Days

I had 11 discovery calls that week. That’s slightly above my average — I usually do 6 to 8 — but March is when the spring market picks up in Madeira, so the timing was useful for a real test.

Before each call, I fed Claude whatever I had on the client and got back a prioritized question list. This took me between 4 and 8 minutes per client depending on how much background information existed. I printed the questions or kept them open on my second monitor. I didn’t read them like a script — I used them as a checklist I could glance at to make sure I wasn’t drifting.

One call stands out. A couple from Germany — he worked remotely, she was a freelance translator. They’d emailed asking about properties “around €400,000, somewhere quiet, maybe with a garden.” Vague. The pre-call brief Claude generated flagged something I probably would have forgotten to ask: whether their remote work income was EU-sourced or from non-EU clients, because that affects NHR tax calculations in ways that matter for their effective cost of living in Portugal. I asked it. Turned out one of their main clients was American, which changed the conversation completely and saved us from going down the wrong tax-planning path for the next 6 weeks.

After each call, the debrief process took me roughly 10 minutes instead of the 20 to 25 minutes I’d normally spend writing up notes manually. Over 11 calls, that’s approximately 165 minutes saved on post-call admin alone. About 2.75 hours in a single week.

The gap analysis check caught missing information in 8 out of 11 calls. The most common gaps: rental vs. purchase intent confirmation (3 clients had been ambiguous), parking requirements (I had apparently been ignoring this question habitually), and decision timeline with a specific move-in date rather than a vague “sometime this year.”

For 6 of those 8 clients with gaps, I sent a short follow-up email the same day using a template Claude drafted. Response rate was 5 out of 6 within 24 hours. The information I got back let me start property searches with actual confidence instead of educated guessing.

By Friday afternoon I had more complete client profiles than I’d had after any previous week of calls. I’m not exaggerating when I say my Monday search sessions the following week were notably faster because I wasn’t starting from incomplete briefs.

What the Numbers Looked Like Across the Week

Metric	Before Claude Stack	During the Week
Calls handled	6–8 avg per week	11
Post-call note time	20–25 min per call	~10 min per call
Complete client profiles after call	~40% fully complete	~73% fully complete
Gaps identified and followed up	Ad hoc, often missed	8 of 11 flagged, 6 resolved
Pre-call prep time	5–10 min (unstructured)	4–8 min (structured output)
Additional tool cost	—	$0 (existing Claude Pro sub)

Where Claude Struggled and What I Had to Work Around

I want to be direct about the failures because they’re real and they matter if you’re thinking of doing this yourself.

Claude has no memory between sessions unless you explicitly feed it context every time. This is the single most frustrating limitation for ongoing client work. I have clients I speak with over weeks or months. Every time I started a new Claude conversation about an existing client, I had to paste in the previous profile, the call history summary, everything. If I forgot to do that — and twice I did — I got generic questions that weren’t calibrated to where that client and I actually were in the process.

I ended up keeping a running Google Doc for each client and copy-pasting the relevant section at the start of each Claude conversation. It worked, but it added 3 to 5 minutes of friction per session that I hadn’t budgeted for. A proper CRM integration would solve this. I don’t have one set up yet that connects cleanly to Claude.

The other limitation: Claude doesn’t understand Madeira’s market the way I do. When it suggested questions around “school districts” for a family with young children, I had to remind myself to filter its output through local knowledge. In Madeira, the relevant question is which municipality the family wants to be in — Funchal, Câmara de Lobos, Santa Cruz — because school infrastructure and commute patterns vary significantly. Claude generated good general real estate discovery questions, but I was always the one adding the geographic specificity. It cannot replace local expertise. It can only help you deploy your expertise more consistently.

How This Compares to Other Approaches I’ve Tried

I’ve tried using a static Google Form for pre-call intake (clients ignored it about 60% of the time). I’ve tried rigid call scripts (felt robotic, clients sensed it). I’ve tried recording calls and transcribing them with Otter AI for post-call notes (good transcription, but still required manual structuring).

The Claude approach sits in a different category from all of these. It’s not a form clients fill out. It’s not a transcript tool. It’s a thinking partner that helps me think more clearly before and after a human conversation — without replacing the conversation itself.

The closest comparison is using a checklist, but a checklist doesn’t adapt to individual clients. Claude does. That adaptability is what made the German couple’s NHR question surface when a generic checklist would have missed it entirely.

What I’d Do Differently If I Started Over

Three changes I’m making going forward:

Build a persistent client context file from day one. Every new client gets a Google Doc immediately. After every touchpoint, I update it with a timestamped note. When I go to Claude, I paste the entire doc. This adds maybe 2 minutes per client but eliminates the memory problem almost entirely.

Create a Madeira-specific knowledge layer in my prompts. I’m building a standing prompt appendix that tells Claude the relevant local factors — municipalities, typical commute patterns, property types common in different zones, NHR implications for different income sources. This way I don’t have to filter as much; the output starts closer to where I need it.

Add a second gap analysis pass 48 hours after follow-up emails. If I still have missing information after the follow-up, I want a prompt that drafts a second, softer nudge. I didn’t build this during the week and I should have.

My Honest Assessment: Is This Worth Your Time?

If you run a solo operation where client relationships are long-cycle and high-value — real estate, consulting, financial advisory, anything where a bad first call costs you weeks of wasted work — yes. The setup time is low and the compounding benefit is real.

If you do high-volume, short-cycle sales where discovery is quick and standardized, you probably don’t need this level of customization. A good CRM template would serve you better.

I’d give this workflow a 4/5 — it materially improved my client profile completeness rate from around 40% to 73% in a single week, which is the most concrete evidence I have that it’s working, but the lack of native memory between sessions is a real limitation that requires workarounds I shouldn’t have to build manually.

Claude Pro costs $20/month. I was already paying it. The marginal cost of this specific use case was zero. The time I recovered — roughly 3 hours across one week, plus the avoided cost of another February-style mistake — makes it an easy decision to keep running.

Where to Go From Here

If you want to try this yourself, start small. Pick your next five discovery calls. Before each one, give Claude everything you know about the client and ask for a prioritized list of questions you should ask. After each call, dump your notes in and ask Claude to structure them. See if the output catches things you missed.

You don’t need a complex setup. You need a habit of using it consistently, and a willingness to paste in context every time until you’ve built a system that makes that automatic.

If you’re running a solo real estate operation or any other high-stakes client business and want to talk through how I structured the prompts, drop a comment below. I share the actual prompt templates I use with readers who ask — no opt-in required, just ask.