The 5 Claude AI agents that generated 565 meetings in 2025 and how to build them
We built 5 AI agents with Claude. They have been running in production for our acquisition and our clients' acquisition for more than a year and a half.
More than 2.5 million operations per month. 565 prospect meetings generated in 2025.
These 5 agents do not operate separately. They are connected: each one filters for the next, from a raw LinkedIn signal to a meeting in the calendar. Here is how each one works, with practical use cases and the keys to building your own.
Why Claude rather than GPT or another LLM?
- More reliable for classification and structured scoring: it follows output formats such as JSON, scores out of 100 and strict categories without inventing data
- Better understanding of nuances in French-speaking LinkedIn profiles
- More restrained responses: it writes like a person, not like AI trying to impress
- Prompt caching cuts costs by 90% on recurring prompts, a major difference across thousands of leads per month
This is not an ideological choice. It is based on 12 months of production data.
Agent No. 1: the persona classifier
Leads from LinkedIn signals include decision-makers, influencers, operational contacts, competitors and irrelevant profiles. Sorting them manually takes 45 minutes for 10 leads.
The agent analyzes the job title, headline, summary, company, company size and industry. It then assigns a persona defined for each client:
- Persona A, direct decision-maker: CEO, Managing Director or founder of a company with 10 to 500 employees
- Persona B, influencer: Sales Director, VP Sales or Head of Growth
- Persona C, operational contact: Sales Manager or Business Developer, useful for nurturing
- Persona D, outside the target: student, freelancer outside the industry or identified competitor
The key: rules based on patterns, not keywords. "Founder and Manager," "Co-founder" and "President" are treated as equivalent, while the CEO of a two-person company does not have the same value as the CEO of a 200-person company. Claude reasons by elimination: competitor, outside the industry, then decision-making level.
30% of leads automatically removed before any contact
More than 95% of leads classified correctly
A persona list specific to each client
Agent No. 2: the maturity scoring agent
A lead who only accepted your invitation is not the same as one who visited your profile, liked 3 posts and downloaded your gated content. The agent assigns a score from 0 to 100% by combining:
- Prospect data: persona, company size, industry and location
- Engagement signals: profile visits, likes, comments, follows, gated content, landing pages and interactions with competitors or thought leaders
- Context signals: job change, funding round and hiring related to your offer
The key: it is not a point count. Claude analyzes the combination of signals. Three profile visits in 7 days, a 50-person SaaS company and a comment on a competitor's post suggest active research, so the score rises above 80%. A perfectly targeted CEO with no signal remains at 40 to 50%.
Only leads scoring above 60% are contacted
40% higher average reply rate
One client reached a 91% reply rate among leads scoring 80% or more
Agent No. 3: the icebreaker generator
"Congratulations on your new role" and "your profile caught my attention" no longer work. The agent starts with the prospect's real activity: posts, comments, job change and company news.
Strict constraints: 3 sentences maximum, no generic flattery, no disguised pitch and a message a person could send without editing.
For a client selling digital services to construction companies, the agent audits the prospect's online presence before any message. It reviews the website, landing pages, tracking, social networks, Google listing and active advertising, then assigns a digital maturity score out of 10. For a timber construction company in Toulouse: landing pages but no tracking, an excellent Google listing rated 4.8 out of 5 from 19 reviews, and no advertising. Score: 6 out of 10.
"Hello, here are the points I noted: your website has landing pages for timber construction and renovation, but GTM is not detected, which prevents you from tracking quote requests. You have an excellent 4.8 out of 5 rating from 19 Google reviews, but posting completed projects weekly could increase your local click-through rate by 25 to 35%. Are you planning to implement these improvements this year?"
This is no longer a prospecting message. It is a mini assessment the prospect cannot ignore. It is then automatically added to a campaign in the platform.
More than 50% reply rate, compared with 5% for templates
Every icebreaker is unique, with no sense of repetition
Agent No. 4: the LinkedIn content creator
People recognize an AI post in 3 seconds. The agent therefore learns each client's unique voice from 10 to 15 of their best posts: sentence length, vocabulary, structure, tone and the concerns of their personas.
The client provides a topic, a content pillar and rough notes. Claude co-creates the post like a ghostwriter: structure, writing in the client's voice, alternative hooks and an appropriate format.
The key: the voice prompt captures the client's editorial identity, including natural and prohibited phrases, typical structure, positioning and results that can be cited. It improves with every round of feedback.
No generic AI posts
Engagement maintained or improved
Creation time reduced by 80%, from 45 minutes to about 10 minutes of review
The client approves and adjusts every post
Agent No. 5: the AI appointment setter
The prospect replies, then problems arise: a next-day response, an overly aggressive message, a mishandled objection or a forgotten follow-up. Appointment setting is where 80% of leads are lost.
The agent analyzes the level of interest and the tone of the reply, then drafts a response that you approve before it is sent.
- Never force a call
- Match the length of the prospect's message
- Ask no more than one question per message
- Always provide something useful
- Address objections without pressure
Example: the prospect replies, "Interesting, but we already use Lemlist."
"Lemlist does a good job of sending sequences. The question I often see is how to supply Lemlist with genuinely qualified, mature leads rather than cold lists. That is where this approach changes things. If this is relevant to you, I can show you how we structure it. It takes 20 minutes."
Reply-to-call conversion significantly higher than manual appointment setting
Writing time reduced by 90%
No leads lost because of a poorly calibrated message
A person approves every reply
What this changes in practice
You no longer need to spend hours on Google, pay a fortune for enrichment credits or copy and paste information. You automatically collect the information needed to personalize messages and qualify hundreds of prospects in minutes.
Integrated into the Allbound platform, the result is a 40 to 50% reply rate in prospecting. One of our clients even reached 91%.
Want these 5 agents working on your acquisition?
See how the Allbound platform can adapt to your market.
Book a call