If someone does it by copying and pasting, it can be automated

Invoices someone types by hand, the Monday report, emails classified and forwarded one by one, data copied from one system to another. That starts running on its own: it alerts you when something fails, retries what can be retried, and logs what it did.

2,000,000 active users served · 10 countries in operation

n8n
Document processing
AI extraction
Integrations
Alerts
Retries

Built on

An engineering team that has already been on the other side.

Ninety seconds: who we are, how we work and what you get at the end.

An agent that takes the event, decides and leaves a record

An email, an order or an alert comes in. The agent queries your systems, decides by your rules, asks for approval when needed and executes. Pick a case and watch it run.

Example: a customer asks what their policy covers. The system answers with the exact page of the contract and opens the case.

01Event arrives

Does it cover theft from a public car park?

context
product = motor_premium

02Plans

Queried · policy 88213

Read · clause 4.3

Decided · open a claim

03Queries the systems

Policy_wording_2026.pdf84 pp

Motor_exclusions.pdf12 pp

Claims_email1,240

ERP · policies38,400

84 pp → 612 chunks

04Human approval

Asked for approval · human

ApproveReject

05Executes and logs

Queried · policy 88213
Read · clause 4.3
Decided · open a claim
Asked for approval · human
Executed · created the case in the ERP

log · 5 steps · who and when

Swipe to follow the flow →

Map the process as it is today

With the person who runs it, not with the org chart. We routinely find three undocumented exceptions, and those are exactly what break badly built automations.

Build with the exceptions inside

An automation that handles 80% of cases and lets the other 20% through unannounced still forces someone to check everything by hand, which was the whole point of automating.

Monitoring and alerts

When something fails, and eventually it does, someone finds out the same day, with detail on what was processed and what is pending, and the process can be retried without duplicating.

Documentation and handover

Written down: what it does, how to modify it, who to call. The goal is that your team can adjust it without depending on us.

Formats

Scopes and timelines

01 2–3 weeks

One process

One circuit end to end: supplier invoices land in the accounting system, the Monday report builds and sends itself, or incoming email is classified and forwarded to whoever owns it.

02 4–6 weeks

Several connected processes

One event triggers the whole chain: the web order draws down inventory, issues the invoice, notifies the customer and alerts the team. Every step is logged and retried when the system on the other side does not answer.

03 +1–2 weeks

With intelligent document processing

Added to either of the two above when what comes in has no fixed shape: invoices from a hundred different suppliers, photographed delivery notes, scanned PDFs. The model reads them; there is no template to maintain.

Common cases

What we get asked for most

Invoices and documents typed by hand

Someone keys them into an accounting system or a spreadsheet. Today they are read automatically, even scanned or photographed on a phone.
Recurring reports

Someone assembles the same report every week from three different sources.
Email classification and routing

Quotes go to sales, support to support and invoices to accounting, with the key data already extracted.
Syncing systems that do not talk

The online store with inventory, the CRM with billing, web forms with the database.
Alerts and follow-ups

Expirations, outstanding payments, low stock and contracts up for renewal, which today depend on someone remembering.
Bulk data loading

Files that arrive from suppliers or customers in a different format every time.

How to tell if it applies

The three-minute test

Does someone do it more than once a week, always the same way, moving information between two places? If the answer is yes, it can almost always be automated, and it usually pays for itself in under a year.

What AI changed is which tasks make that list. Data used to have to arrive in the same shape every time. Today the model reads invoices with different layouts, emails written the way people write them and scanned documents, pulls out the data and checks that it has the right shape before passing it on. That adds tasks that could not be automated three years ago.

We work from Medellín with companies across Colombia and remotely with teams in the United States and the rest of Latin America. The process is mapped over video with whoever runs it: what we need to see is the screen where the copying and pasting happens today, not the office.

This is where the data comes in

An AI system is worth what its sources are worth. These are the standard connectors; anything with an API or a database connects the same way, and what has no API is handled by file.

SAP

Enterprise ERP

ERP

Oracle

ERP and database

ERP

NetSuite

Cloud ERP

ERP

Salesforce

CRM and service

CRM

HubSpot

CRM and marketing

CRM

PostgreSQL

Database and pgvector

Databases

Fourteen sectors, and in each one the measurement changes

Insurance

Policy wordings, exclusions, claims

Says what a policy covers, quoting the clause and the version in force that day. If the wording changed in March, it answers with March’s.

AI for Insurance →

An example

— Am I covered if my car is stolen from a public car park?

Yes, with a 10% deductible. 2026 policy wording, clause 4.3.

A RAG system in production: three years, two million users

We built and operated the assistant for a loyalty platform serving more than two million active users across ten countries.

We built the full pipeline: document ingestion and normalization, chunking, embedding generation, vector store on Azure AI Search and Pinecone, and retrieval with grounded generation on LangChain.

We held it above 99% availability for three years.

RAG systems →

Frequently asked questions

What is the difference between n8n, Zapier and custom development?

Zapier is the easiest and the most expensive at scale: it charges per task executed and the bill climbs fast at volume. n8n does the same but can be self-hosted at fixed cost and handles considerably more complex logic. Custom code is for processes with rules no visual tool expresses well. For most mid-market companies self-hosted n8n is the sweet spot, which is why it is our default.

Do we have to change the systems we already use?

Almost never. The automation connects to what you have through APIs, databases or even exchanged files. If one of your systems is completely closed with no way to connect, we tell you on the first call before you invest in a proposal.

What happens if the automation fails one day?

It will fail eventually: an API changes, a file arrives in a strange format, a service goes down. That is why monitoring is not optional. When something fails, someone on your team gets an alert the same day with detail on what was processed and what is pending, and the process can be retried without duplicating anything.

Does this replace people?

In practice, almost never. It removes mechanical work so the same people handle what requires judgment. Most of our clients automate because their team is underwater, not because they want to shrink it. If your goal is headcount reduction, say so upfront: it changes how the project is designed and what has to be documented.

How reliable is AI document reading?

Very good on documents with reasonable structure, and much better combined with schema validation and business rules. What matters is designing what happens with low-confidence cases: those go to human review rather than into the system. A flow that accepts everything blindly is where automation becomes a problem instead of a solution.

What does it cost to maintain?

With n8n on your own server, roughly USD 20 to 60 monthly in infrastructure for typical volumes. If there is AI document reading, model consumption is added on top depending on volume. We do not charge a subscription on the automation: it is yours and documented so your team can maintain it.

What is the difference from an AI agent?

An automation follows a defined path: if this happens, do that. An agent decides the path based on the goal and context. Automation is cheaper, more predictable and easier to audit, so when the process has clear rules it is the right answer. The agent comes in when the path cannot be enumerated in advance.

Where should we start?

With the process that consumes the most hours and requires the least judgment. Not the most visible one, and not the one the executive asks for: the one that makes someone on your team lose half a day a week to something mechanical. That one usually pays back fastest and builds the confidence to keep going.

Fifteen minutes. A concrete answer.

It gets fixed, it gets built, or it is not worth it. And if the diagnostic does not reach three actionable findings, it is not charged.