We sell a prospecting platform, so people often ask how we prospect ourselves. Here's the honest answer: with an AI assistant, a scheduled prompt, and Goava's data underneath.
I (Anton, CEO at Goava) run this flow every weekday. So do our two sales reps. When we open our inboxes in the morning, short, personal email drafts are waiting for the right people at the right companies. We read them, tweak a word or two, and press send.
We keep the send volume low and relevance high, so that we can send using our own domain and don't have to use any third party sender infrastructure.
Below is the exact recipe, simplified so you can copy it. And because more and more teams try to build something similar with only an LLM and the open web, we'll also share why we run it on structured data, and what tends to go wrong without it.
Every weekday morning, it finds new sales opportunities and prepares a short email draft for the right people. If nobody answers, it also prepares one follow-up draft.
"Every weekday: get yesterday's opportunities from Goava. For each company, check our CRM. Skip anyone contacted in the last 60 days. Pick one or two of the recommended contacts who have an email address. Write a short email draft in my style: mention the signal, ask one question about how they find new customers, add two lines about Goava, and suggest a 30-minute meeting. Save it as a draft in my email. Don't send it. If there's no reply after two working days, save one follow-up draft in the same thread. Then give me a summary of which drafts are ready."
Notice how short the prompt is. Finding the companies, spotting the signal and picking the right person all happen in Goava before Claude even starts. Claude's job is the part AI is great at: checking, writing and organising. When you copy the prompt above I'd recommend adding "ask me questions before you set up the scheduled task " so your agent can clarify things for your specific workspace.
You could ask an AI assistant to do all of this on its own: search the web for companies, read news for signals, guess who the decision maker is. For a one-off question, that works fine. For a flow that runs every day for a whole team, it gets expensive, fragile and risky.
1. LLM costs you can't predict. When an LLM has to find signals and data on its own, every run means searching, reading and re-reading web pages. How many tokens that takes depends on the day, the companies and the web. It's hard to know what next month will cost. With Goava, the signals are already found and the data is already structured. The agent fetches a short, clean result, and contact details use data credits at a known price.
2. Stability and maintenance. A homemade scraper or a long chain of web searches tends to work great in the demo. Then a site changes, a source disappears or a prompt drifts, and the flow quietly gets worse. Often only one person knows how it works, so nobody can say what the upkeep really costs. With Goava, we maintain the sources and the data. Your part stays a short prompt that anyone on the team can read.
3. Legal risk. Scraping the web and processing personal data brings GDPR exposure unless it's done right: lawful basis, information to the people concerned, deletion and more. Building that yourself is a project in its own right. Goava's data is collected and processed in line with GDPR, so your team can focus on selling.
You don't need to copy our flow exactly. Here's how to build your own version.
"Using Goava, build a segment of B2B companies in Sweden with 20–200 employees in IT services and staffing. Show me the filters you used."
Your AI assistant is already a capable prospector. Give it the data to match.
Connect to your Claude, ChatGPT, or Copilot.
PS. Built something clever with Goava MCP? We'd love to see it. Send it to us, and we might feature it in our next post.