14 AI agents in production at Albus Factory: case study
The full system of 14 agents currently running at Albus Factory. By function, with real activity numbers, the tools used, and what they concretely changed for a 9-person SME.
Albus Factory is a 9-employee SME that today operates like a team of 50. 14 AI agents in production, integrated with Hubspot, Slack, Gmail, Apollo, Drive. Here is the full system, function by function, with the real volumes processed in April 2026.
Sales function - 5 agents
SDR Orchestrator
The outbound prospecting orchestrator. It reads the Hubspot queues (SDR, Email, LinkedIn, SMS, Human), decides the action of the day for each prospect, coordinates the multi-contact account strategy, and delegates writing to the SDR Message Writer. Manages a 6-touch sequence over 80 days per prospect.
Volumes: 20 new prospects injected per day from Apollo, 100 outbound emails sent per week, 6 to 12 qualified meetings per month.
SDR Message Writer
Sub-agent of the SDR Orchestrator. Drafts a calibrated message (email, LinkedIn, SMS) per touch and per prospect seniority. Touch 1 = social proof. Touch 2 = new angle. Touch 3 = curiosity. Touch 4 = value. Touches 5-6 = wave 2. Each message is unique, no template.
Conversation Handler
Every 2 hours, detects incoming replies (prospects, clients), analyzes content, drafts the reply on the same channel, notifies Slack with preview + Hubspot link, and schedules the send 1h30 out (3h for meetings). A human can intervene on the preview before sending. Effective response time: under 4 hours during business hours.
Account Manager Orchestrator
Daily loop on existing clients. Reads Claude Account tasks, loads client context (deals, Gmail emails, Hubspot notes), assesses relationship health, then drafts personalized messages (email or SMS) in partner tone. No cold outreach. Detects upsell signals and weak signals of disengagement.
Email Sender
Execution agent. Sends the emails prepared by the SDR Orchestrator and the Conversation Handler from the Claude Email Hubspot queue. Verifies pre-send conditions (reply received in the meantime, manual cancellation), sends via Gmail, updates Hubspot. Tracks the daily counter (max 100 outbound emails/day to avoid spam).
Growth function - 3 agents
Content Blitz Engine (SEO)
Produces 5 to 10 optimized articles per week, published directly to Webflow via native MCP. Each article: 1,500 to 2,500 words, integrated SEO brief, automatic internal linking, generated images.
SEO Keyword Dominator
Tracks 300+ keywords continuously. Detects position drops, recommends article updates, identifies new opportunities. KPI report delivered every Thursday at 8am.
Google Ads Orchestrator
Coordinates 6 sub-agents: Account Auditor, Weekly Optimizer, Campaign Builder, Performance Reporter, Ad Copy Lab, Budget Redistributor. The budget is redistributed each week based on real performance. Ad copy generated and tested on demand.
Operations function - 4 agents
Meeting Report
Turns each meeting recording into a structured minutes document in 5 minutes. Identifies decisions made, action items, owners. Automatically creates Hubspot tasks from notes. Minutes shared to participants without human intervention. Volume: 15 to 25 meetings per week processed.
Handoff Automator
As soon as a deal is signed in Hubspot, generates the full spec doc (scope, milestones, risks, contacts) and creates the Asana project with all tasks structured. Time: 10 minutes vs 3 hours before. Zero information loss between sales and delivery.
Task Audit
Daily supervisor of the sales system. Audits the 9 Hubspot queues (SDR, Email, LinkedIn, SMS, Account, ABM, Nurture, Partnership, Human), detects anomalies (delays, duplicates, missed replies, ghost contacts, CRM inconsistencies), cleans duplicates, reinjects orphan contacts. Safety net that ensures no opportunity slips through the cracks.
CRM Enrichment
Apollo enriches contacts (email, direct phone, intent signals). 200 contacts scored on 5 ICP criteria each run. Anomalies are detected and corrected automatically.
Communication function - 2 agents
Content Calendar + LinkedIn Creator
Every Sunday at 8am, orchestrates the editorial week. The LinkedIn Content Creator produces 3 posts + Canva visuals. The Brand Voice Enforcer validates each piece before publication. Calendar ready Monday, week published without intervention.
Email Routing
Automatic inbox prioritization, classification by request type, drafting calibrated to brand voice. Only high-value messages bubble up to the human for validation.
Aggregated system numbers
14 - Agents in prod
9 → 50 - Humans → output
57 h/week - Automated
< 50K EUR - Invested in system
Concretely, these 14 agents automate about 57 hours of human work per week, or 1.5 FTE. Combined with the productivity boost of remaining humans (fewer repetitive tasks, more advisory), it's the equivalent of 4 FTE avoided over 14 months.
Total system infrastructure cost (Claude API + Apollo + Hubspot + Gmail + servers + secondary tools) is below 4,000 EUR per month. Less than 50,000 EUR over 14 months. The ROI is direct: 4 FTE at 50K EUR loaded = 200K EUR saved per year against 50K EUR spent. 4x factor.
The 3 pitfalls avoided at Albus
1 · Don't chase total automation
Some agents have a human in the loop (Conversation Handler, Email Sender). It's not a flaw, it's a choice. A 100% autonomous agent on a sales-critical decision can lose a 50K EUR opportunity. The human costs 15 minutes per day but secures critical trade-offs.
2 · Don't build before you have the memory
Albus's first agents were bad because they lacked context. 40% of the transformation time was invested in formalizing the business knowledge (ICP, brand voice, offers, objections, history). This memory is what makes the agents relevant. Without it, they're chatbots.
3 · Don't aim for perfection on the first run
Each agent writes its own lessons after each run and re-reads them on the next run. In 4 months, 13 lessons files accumulated. The system learns. An agent in production isn't a delivery, it's a starting point.
The real question isn't how much it costs to deploy this system. It's how much it costs not to. For an SME losing 2 points of margin per year to better-equipped competitors, inaction is the most expensive choice.