Custom Business OS: Airtable + n8n - 170h/Mo

Want to know more about Jacobo?

Omnichannel AI agent that handles customers via WhatsApp and voice, checks stock in real time, and manages appointments autonomously.

How did we generate web content for every model and repair?

The Airtable CMS powered a programmatic website with unique pages per model and repair type — all automated with keyword research.

Airtable combines spreadsheet flexibility with relational database structure. For a business that iterates weekly, speed of change is critical. Adding a new field or view takes minutes, not days of development.

SaaS imposes its data model. When your business is multi-service (repairs + accessories), no vertical SaaS covers everything. The adaptation cost exceeds the build cost.

If your business fits a standard vertical (repairs only, no multi-service), use RepairDesk or similar. Building custom makes sense when your business differentiator is in the processes, not the product.

Airtable automations live inside the base itself, access data directly, and have no per-execution cost. For day-to-day business logic (50+ automations), that's unbeatable. Make is used as glue for specific SaaS integrations: new Google My Business review notifications (official integration), supplier webhooks and payment gateway syncs. n8n is used for Jacobo (the AI agent), where complex orchestration with language models and tool calling is needed.

By separating concerns: each base owns a clear domain and only syncs strictly necessary data with other bases via linked records. No duplicating everything — just linking what's needed. Combined with role-filtered views (technicians see their stuff, sales sees theirs), each user interacts with a manageable subset of the system.

Simple → native Airtable automations (zero cost per execution, but capped at 100,000 runs/month on the Business plan; hit the limit and they stop cold). SaaS glue → Make, fast and robust for integrations with external APIs (Google My Business, suppliers, payments), with purchasable credits if you need more capacity. AI orchestration → n8n for agents with LLMs and tool calling. Heavy computation → custom code. Rule: push logic as close to the data as possible.

Airtable generates recXXXXX for internal linking. Employees and customers use readable codes: OT-2024-04521, FAC-2024-01234. The separation prevents human errors without sacrificing referential integrity.

LAST_MODIFIED_TIME() on critical fields creates a queryable audit trail without external logging. Every change to a work order, invoice or quote is recorded with date and user.

Each base syncs only the necessary fields with other bases. Known trade-off: cross-base reporting requires intermediate views, but the alternative (duplicating data) creates inconsistencies that cost more than the workaround.

The same systems thinking, applied to LLMOps

The chatbot on this portfolio uses the same principles: agentic observability, automatic closed-loop, and 71 tests. From automating a physical business to automating an AI pipeline.

Got an operational problem that doesn't fit in a SaaS?

I built a system that managed 30,000+ repairs, automated 170h/month, and survived a business sale — intact. Whether it's for your own company, your platform's customers, or a team that needs someone to build from the inside — tell me the problem.

Yes, with caveats. On the Business plan (125K records per base), Airtable handles tens of thousands of records well. The key is designing bases with filtered views and not loading everything in a single view. As you approach the limit, consider periodic archiving or migration to Postgres.

Airtable Business (~$45/mo per user, 125K records/base plan) + integrations (YouCanBookMe, WATI, Make) ≈ $120-170/mo. RepairDesk is $99/mo but doesn't cover advanced CRM, complex automations, or the multi-service model. The real savings are in the 170h/month of eliminated manual work.

It's a real risk. The mitigation: periodic data backups and, most importantly, having the complete data schema documented. If migration were ever needed, the relational structure of the 12 bases is the real asset — it can be replicated on any platform.

Years of trial and error. But with a key pattern: each new module took half the time of the previous one, because accumulated learning accelerated everything. The system grew organically while the business operated — there was never a "stop everything and build" moment.

The buyer. I sold the business in 2025 as a going concern: brand, systems, and workflows included. The fact that the buyer acquired it as a running operation — without needing to rebuild anything — is the ultimate validation of the architecture.

The architecture (Airtable as SSOT with native automations) is replicable for any service business: workshops, clinics, agencies. What changes are the specific business rules and required modules.

Yes, with design discipline. Airtable isn't an out-of-the-box ERP, but its relational flexibility lets you build a custom one. The key is treating each base as an independent module with clean interfaces (linked records) and using native automations to orchestrate the flows. With 12 bases and 2,100+ fields, this Business OS handles everything a traditional ERP would — but with weekly iteration instead of monthly.

The main ones: 125K record limit per base on the Business plan (requires archiving if you grow significantly), pricing that scales fast with users, and vendor lock-in. The mitigation: intelligent data design with separate bases per domain, syncing only the necessary information between them, not everything at once. For this Business OS, the advantages (iteration speed, flexibility, friendly UI) far outweigh the disadvantages.

Airtable automations live inside the base itself, have no per-execution cost, and access data directly without intermediate APIs. For day-to-day business logic (50+ automations), it's the most efficient option. The limit is 100,000 runs/month on the Business plan: hit it and they stop, with no option to buy more. That's why external system integrations (Google My Business reviews, supplier webhooks) go through Make, where you can purchase additional credits if you need more capacity. n8n is used for Jacobo (the AI agent), where complex orchestration with language models is needed.