How I Automated Client Reports Using Claude API

I spent 11 hours every month writing client reports. Market updates, property performance summaries, buyer pipeline breakdowns — all formatted, all personalized, all done by hand. Then I connected the Claude API to my workflow in January 2026, and that number dropped to under 2 hours. That is not a rounding error. That is 9 hours back in my life every single month.

If you are a solopreneur — especially one in a service business with repeat clients — automated client reporting might be the single highest-leverage thing you can build this year. And the Claude API, specifically, is the tool I would choose again if I were starting from scratch today.

Here is exactly how I built it, what it cost, where it broke, and what I would do differently.

The Problem: Manual Reporting Was Eating My Business Alive

Running a one-person real estate consultancy in Madeira means I wear every hat. I find listings, negotiate deals, handle client communications, write property descriptions, and produce monthly market reports for active buyers and sellers. For years, my reporting process looked like this: open a Google Doc, stare at raw data from the local property portal, and write something personalized for each client. Multiply that by 14 to 18 active clients and you understand why it consumed most of a working day.

The reports themselves were not complicated. Each one needed: current market prices in the client’s target area, a summary of new listings that matched their criteria, any price movements since the last report, and a short personal note from me. Roughly 400 to 600 words per client. Repetitive structure. Variable data. Exactly the kind of work that should be automated.

I tried using ChatGPT via the standard interface in early 2023. It helped with drafting, but it was still manual — I had to paste data in, prompt, copy the output, paste it into my template. It cut maybe 30% of the time. Not good enough.

Why I Chose the Claude API Over Other Options

I evaluated three API options before committing: OpenAI (GPT-4o), Google Gemini API, and Anthropic’s Claude API. My decision came down to three things.

First, output tone. Claude’s writing sounds more like a thoughtful professional wrote it. When I tested the same property market summary prompt across all three, Claude’s version required the least editing before I would feel comfortable putting my name on it. That matters when the report is going to a client who has known me for three years.

Second, the context window. Claude 3.5 Sonnet (the model I use) handles long prompts well. My full client profile — their budget, preferred zones, purchase timeline, past notes from our meetings — can be 800 words or more. I needed a model that could hold all of that context and still write a coherent, personalized report. Claude managed this consistently.

Third, pricing. At the time I set this up, Claude 3.5 Sonnet was priced at $3 per million input tokens and $15 per million output tokens. For my volume — roughly 16 reports per month, each around 600 words of output — my monthly API cost has averaged €4.20 to €6.80. Trivial.

The Exact Setup I Built: Step by Step

I am not a developer. I want to be clear about that. I built this entire system using Make.com (formerly Integromat) to handle the automation logic, with Google Sheets as my data layer and Claude API as the writing engine. No code beyond a few simple HTTP request configurations.

Step 1: Build the Client Data Sheet

I maintain one Google Sheet with a row per active client. Columns include: name, email, target zone (e.g., Funchal centro, Calheta, Ponta do Sol), budget range, property type (apartment/villa/land), purchase timeline, and a “notes” column where I paste anything relevant from our last call — things like “wants sea view,” “flexible on timeline if right property appears,” or “concerned about building age.”

There is also a second sheet tab called “Market Data” where I paste the monthly numbers I pull manually from the local portals: median price per square meter by zone, number of new listings, average days on market, and any notable price reductions. This takes me about 20 minutes to update each month.

Step 2: The Make.com Scenario

Once per month, I trigger a Make.com scenario manually (I considered scheduling it, but I want to review the market data before reports go out). The scenario does the following:

• Reads each row from the client sheet
• Pulls the relevant zone data from the Market Data tab
• Constructs a prompt that combines client profile + market data + a base system prompt I wrote
• Sends that prompt to the Claude API via an HTTP module
• Writes the output back into a “Draft Report” column in the sheet
• Sends me an email with all drafts compiled in one message

The whole scenario runs in about 4 minutes for 16 clients.

Step 3: My System Prompt

The system prompt is the most important part. It took me about 3 weeks of testing to get right. The final version tells Claude: write in first person as Robson Penassi, keep the tone warm but professional, start with the most important market movement for this client’s target zone, mention 2 to 3 relevant listings if the data supports it, close with a personal observation, and never use real estate clichés. I also include explicit length guidance — between 450 and 550 words.

The prompt ends with a variable block where the client data and market data are injected dynamically by Make.com.

Step 4: My Review and Send Process

I do not send the reports straight from the automation. I spend about 90 minutes reviewing all 16 drafts, making minor edits — usually just tweaking the personal note section or correcting a detail I forgot to include in the data sheet. Then I send them manually from Gmail. This human review step is non-negotiable for me. These are real clients with real money decisions.

My Real-World Experience: January to April 2026

Let me give you the honest numbers from the first four months running this system.

In January 2026, I had 14 active clients on my reporting list. Before automation, producing those 14 reports took me an average of 47 minutes each — that includes pulling the data, writing, formatting, and doing a final read. Total monthly time: approximately 11 hours.

After setting up the Claude API workflow, my February reporting cycle took 1 hour 55 minutes total. That breaks down as: 20 minutes updating the market data sheet, 4 minutes running the Make.com scenario, and 91 minutes reviewing and sending all drafts. For 14 reports.

By April, I was at 17 active clients. The total reporting time that month was 2 hours 10 minutes. Without automation, that would have been closer to 13.5 hours at my old pace.

One moment that stuck with me: in March, a longtime client named Helena — she has been searching for a villa in Calheta for almost two years — replied to her report within 20 minutes of receiving it. She wrote “This is exactly the kind of update I needed, very clear.” She had no idea it started as an AI draft. The personal notes section I wrote for her specifically mentioned a property that had just reduced its price by €35,000 in her target zone. She booked a viewing that week. That deal is still in negotiation.

The API costs for those four months totaled €22.40 — roughly €5.60 per month. My Make.com plan, which I use for other automations too, costs €16 per month. Even if I allocated the full Make cost to this workflow, I am paying under €22 per month to recover 9+ hours of time. At my consulting rate, those 9 hours are worth considerably more than that.

The reports have also gotten more consistent. Before automation, I will be honest — the reports I wrote at the end of a long day were noticeably worse than the ones I wrote in the morning. The Claude drafts are uniformly solid regardless of when the scenario runs.

Where This System Falls Short

I want to be direct about the failures, because there were real ones.

The biggest limitation: Claude has no access to real-time data. This sounds obvious, but it bit me harder than I expected. In March, I forgot to update one zone’s median price in my Market Data sheet before running the scenario. The reports for three clients in that zone went out with February’s numbers. One client noticed and asked about it. Embarrassing. The system is only as good as the data I feed it. Garbage in, confidently written garbage out.

Second issue: the first-person voice is not always right. Occasionally a draft will use a phrase or sentence structure I would never write — slightly too formal, or a word choice that feels off for Madeira’s market. Nothing that damages trust with a client, but it requires attention during the review phase. This is not something you can fully engineer away. The review step earns its 90 minutes.

Third: setting this up took real time. Between learning the Claude API documentation, building and debugging the Make.com scenario, and refining my system prompt through 20+ test runs, I spent roughly 9 hours on the initial build. That investment paid back within the second month. But if you have no experience with Make.com or HTTP API calls, budget more time than you think.

Cost and Time Comparison: Manual vs. Automated Reporting

Factor	Manual (Before)	Claude API Automated (After)
Time per month (14-17 clients)	~11 hours	~2 hours
Monthly tool cost	€0	~€6 API + €16 Make.com
Report consistency	Variable (time-of-day dependent)	Consistent quality
Personalization	High (fully manual)	High (data-driven + manual review)
Risk of data errors	Low (I saw the data directly)	Medium (depends on my data entry)
Setup time required	Zero	~9 hours one-time
Scalability	Degrades with more clients	Handles 30+ clients at same cost

What I Would Do Differently If Starting Today

Build the data validation step first. Before I ever run the Make scenario, I now have a simple Google Sheets formula that flags any zone with data older than 28 days. I built this after the March mistake. Takes 15 minutes to set up, prevents a lot of awkward client conversations.

Start with 3 clients before you run it for all of them. My first live run went to everyone simultaneously. Two of those reports had a tone issue I would have caught if I had tested on a small batch. Send a pilot group, review carefully, then open the floodgates.

Invest more time in the system prompt upfront. I spent too long debugging the Make.com scenario when the real problem was a vague system prompt. The clearer and more specific your instructions to Claude, the less time you spend editing outputs. I now treat my system prompt like a junior employee’s onboarding document — every edge case covered, nothing assumed.

Consider Claude 3 Haiku for the first pass if cost is a concern. Haiku is significantly cheaper, and for a structured report with good data inputs, it performs decently. I still use Sonnet because the tone is noticeably better and the cost difference at my volume is under €3 per month. But for a solopreneur just starting out, Haiku is worth testing.

Is This Worth Building for Your Solopreneur Business?

If you produce any kind of structured, recurring written report for clients — market updates, performance summaries, analytics breakdowns, project status reports — yes. The Claude API is particularly well-suited because the output sounds like a human professional wrote it, not a template generator.

The setup requires some patience and a willingness to learn Make.com basics (or hire someone for a one-time build — I have seen this done for under €200 on Fiverr). Once it runs, it runs. I have touched my core scenario twice in four months, both times to add a new data field.

My overall rating for this approach in a real estate solopreneur context: 9/10 — it recovered 9 hours per month immediately and has held that result across four months without meaningful degradation in report quality, with the one-point deduction for the data dependency risk that requires discipline to manage.

The Anthropic API documentation is available at docs.anthropic.com — it is genuinely readable even if you are not a developer. Start there, then build your Make.com scenario around the HTTP module. The whole thing is achievable in a weekend.

If you want to see the exact system prompt structure I use, or have questions about the Make.com setup, drop a comment below. I check them weekly and respond to the specific questions.