Workflow automation with n8n and AI

We build workflow automation with n8n. It connects the apps you already pay for (forms, email, spreadsheets, CRMs, accounting tools, AI models) and runs the steps between them on a trigger or a schedule. n8n bills its paid plans per execution, and an execution is "a single run of your entire workflow", however many steps it has (n8n pricing, checked 1 November 2026). Below: what we build, Cloud versus self-hosting, the licence, how a project runs and what you own.

What we build

Every project ends with working workflows plus what you need to run them without us:

• The workflows, live in your n8n account and exported as JSON files, the format n8n saves workflows in (n8n docs).
• An error workflow for each production workflow. It starts with n8n's Error Trigger node, runs when an execution fails and sends a named person the failed step and a link to the run. That link only works if n8n saved the execution, so failed runs stay saved (n8n docs).
• Credentials connected in your own accounts and named plainly.
• A runbook: what each workflow does, what starts it, who gets alerts, how to pause it and how to re-run a failed execution.

Hypothetical examples of the shapes this work takes (none of them is a case study):

• Forms into your tools. A quote request on your website creates a CRM contact, posts a summary to your sales channel and emails the customer a confirmation.
• Invoices into your books. An AI step reads supplier PDFs from a shared inbox, the workflow checks the lines add up to the total, and each invoice becomes a Google Sheets row or a draft bill. Failures go to a person.
• CRM updates. A deal marked "won" creates the onboarding tasks, updates the revenue sheet and tells the delivery team.
• Scheduled reports. Every Monday at 8am, last week's orders, refunds and low-stock items arrive as one short email.
• AI steps. Sorting email, pulling fields from documents, drafting replies. n8n can make an AI agent wait for a person to approve an action, and the request can reach them in Slack, Microsoft Teams, Gmail or WhatsApp. We would use that for anything that sends a message or changes records (n8n docs).

For the AI steps we bring rules that held up in the Python LLM backends we've shipped. Ask the model for JSON only and parse it defensively, because models wrap JSON in code fences or cut it off. Check every field in code before passing it on. Treat an empty or truncated answer as a failure. Retry rate limits and server errors with a growing delay, and don't retry a request the API refused for any other reason.

A workflow that stops and alerts someone does less damage than one that quietly writes bad data into your books.

For a chatbot that answers customers from your documents, see AI chatbot development. If the process needs its own database, screens and user roles, it's probably custom software with n8n around the edges.

Who it's for

Founders, operations managers and small teams in the US, the UK, Europe and the Gulf whose tools don't talk to each other, with someone moving data between them by hand every week. Also teams who have outgrown per-task billing elsewhere, or want to self-host to control where their data sits.

Pakistani businesses too. If your orders arrive on WhatsApp and get typed into Google Sheets by hand, n8n can sit in between. Meta charges per delivered message on its WhatsApp Business Platform, by message type and the customer's country. Messages from customers are free, and each business number gets 1,000 free service messages a month (Meta pricing, checked 1 November 2026). Those costs go in the plan.

n8n Cloud or self-hosted

n8n's own docs set out the trade-off (Choose how to use n8n, checked 1 November 2026). On n8n Cloud, n8n handles "hosting, updates, and scaling". Self-hosted, the Community edition is free "with almost the complete feature set", but you "must provide and manage infrastructure" and maintenance is "your responsibility".

If you self-host, three things from n8n's docs matter. By default n8n keeps workflows, credentials and past executions in SQLite; it also supports PostgreSQL, versions 16 to 18 when we checked on 1 November 2026 (n8n docs). It creates a random encryption key on first launch, saves it in the ~/.n8n folder and uses it to encrypt credentials (n8n docs), so that key belongs in your backups. And upgrades are yours: n8n 3.0 shipped on 27 October 2026 with breaking changes, and n8n asks you to run its migration tool first (release notes, checked 1 November 2026).

What we would do: start on Cloud unless a self-hosting reason applies. Worked example with assumed numbers: a shop with 40 orders a day, one run per order, makes 1,200 executions in a 30-day month. Add an hourly stock sync (720) and a daily report (30) and the total is 1,950, inside Starter's 2,500. Two Starter limits matter too. It runs at most 5 executions at once, so check that against your busiest hour. And it keeps execution logs for 7 days (Pro keeps 30), so someone has to look at a failed run within the week (n8n docs, checked 1 November 2026).

The n8n licence in plain terms

Self-hosted n8n comes under the Sustainable Use License, which n8n calls its Community license. Its FAQ is short (n8n License FAQ, checked 1 November 2026):

• It is free and applies to self-hosted n8n only. n8n Cloud is a paid subscription with its own terms.
• "If workflows are only created or modified by you or people in your organization, you can use n8n under the Community license for free."
• Allowed: building "automations for your clients on your instance, as long as your clients do not have the possibility to create or edit them", and charging "for workflow creation, setup, and maintenance services".
• Not allowed: hosting n8n as a service where your customers build workflows, letting outside users configure workflows through your product, forking n8n's code to launch your own automation product, white-labelling n8n, or using licensed enterprise features without an Enterprise license.

In a typical project your business runs n8n for its own operations and we're paid to build and maintain the workflows. If your customers will build automations inside your product, n8n asks you to talk to sales@n8n.io about an Enterprise license. Settle that first. This is our reading of n8n's FAQ, not legal advice. n8n takes licence questions at license@n8n.io.

How a project runs

Here is how we would run an automation project, with the same four steps as everything we build.

1. Scope call. We walk through the process as it runs today, exceptions included: the invoice with no number, the form submitted twice. We ask for a few real examples, personal details removed, because the awkward cases live in the data.
2. Fixed written plan. Each workflow, its trigger, the tools it touches, what counts as done and what happens when a step fails, with milestones, price and dates. Running costs (n8n plan, AI usage, message fees) are listed separately, so you see your future bills before you approve.
3. Weekly builds you can try. Workflows run against test data or a test inbox, and you can open the execution list and see every run. Live systems are connected only once test runs look right. Each step that writes data checks first whether the record is already there, so a re-run doesn't create a duplicate invoice or contact.
4. Handover. Code, workflow files, accounts, keys and documentation.

What you own

Everything, from the first day:

• The n8n Cloud account or the server, in your name and billed to you.
• Every workflow, in your n8n account and as JSON files for your own repository.
• Every credential and API key: your OpenAI or Anthropic account, your CRM, your email, all in your name.
• For self-hosting, the database, backups and encryption key.
• The runbook and documentation.

Exported workflow files include credential names and IDs. n8n says the IDs aren't sensitive but the names could be, so we name credentials to give nothing away. It also warns that an HTTP Request node imported from a cURL command can carry an authentication header, so we check for those before a file goes into your repository (n8n docs).

For ongoing changes after launch there is our monthly product retainer; otherwise anyone who knows n8n can take over from the handover.

Stack and tools

n8n does most of the work. Around it, depending on the job:

• n8n on Cloud, or self-hosted with Docker and PostgreSQL.
• OpenAI and Claude APIs for extraction, sorting and drafting, on your own keys.
• Python with FastAPI, or Node.js, when a step is too heavy or custom for an n8n node: awkward file formats, long-running jobs, logic that needs proper tests.
• Supabase, Firebase or PostgreSQL when a workflow must remember things, such as which invoices it already processed.
• AWS or Cloudflare for hosting, webhooks and file storage.
• Next.js for a small form or review screen in front of a workflow. See web development.

Timelines

Rough sizes to plan around, not quotes. The written plan gives firm dates.

• One workflow across two or three tools, with error alerts and testing on real samples: about 1 to 2 weeks.
• 3 to 6 related workflows, or one with an AI extraction step that needs a test set: about 3 to 6 weeks.
• A self-hosted setup with backups and monitoring: add about a week.

What moves these numbers: whether each tool has a usable API (some only export CSV files), how quickly someone grants API access, enough sample documents to test an AI step on the awkward ones, and the exceptions. The main path is quick. The invoice in another currency takes the time.

Working with us from abroad

Pakistan Standard Time is UTC+5 all year, with no daylight saving since 2009 (Wikipedia, checked 1 November 2026). Our hours are Monday to Friday, 10am to 7pm PKT:

For US clients we also take calls from 6pm to 10pm PKT, Monday to Friday. On the US East Coast that's 8am to 12pm in winter (EST) and 9am to 1pm in summer (EDT).

UK clocks change on the last Sundays of March and October (GOV.UK, checked 1 November 2026), so the gap to London is 4 or 5 hours.

The Gulf shares our whole working day, and the UK and Europe share the morning and early afternoon, enough for a weekly call and live questions. US projects get live calls and urgent replies in the US morning, through the 6pm to 10pm PKT window, alongside the weekly build. Wherever you are, you get a build to try every week, and email (hello@zoftaa.com) or WhatsApp messages usually get a reply within one business day. More on hiring developers in Pakistan.

When we're not the right fit

• Your automation is two steps and a ready-made template already does it. Use the template.
• You want to sell a product where customers build their own n8n workflows. Talk to n8n's sales team about an Enterprise license first.
• Your procurement rules require an n8n partner or certified consultant. We are neither.
• You need people watching workflows around the clock. Alerts reach your team at any hour, but we work Monday to Friday in PKT hours.
• The process still changes every week. Automating it freezes it in place, so run it by hand until it settles.
• The data falls under rules that require audited, certified vendors. We hold no certifications.

Frequently asked questions

Is n8n free for commercial use?

Self-hosted, yes, when only your own organization creates or edits the workflows, per n8n's licence FAQ. n8n Cloud is a paid subscription. Not legal advice.

Should we use n8n Cloud or host it ourselves?

Cloud, unless your data must stay on your infrastructure, your setup is customised or your volume is high. Workflows export as JSON files, so moving later is possible.

Can you take over workflows someone else built?

Yes. We would export them, read recent failed executions and document each one, then put the fixes into a written plan.

Can n8n workflows use OpenAI or Claude models?

Yes. n8n's pricing page lists agents and AI nodes on every paid plan. We connect them to your own OpenAI or Anthropic account, with a human approval step before any AI step that sends something or changes records.

What happens when a workflow fails?

Its error workflow alerts the person named in the runbook with the workflow, the failed step and a link to the run. You can load a past execution's data to debug it, then re-run once fixed. Steps check before they write, so a re-run doesn't create duplicates.

Are you an n8n partner?

No, and we hold no n8n certification. n8n's expert programme for system integrators is a closed pilot with a waitlist (n8n partners), and we are not in it.

How do we start?

Send a few lines through the contact page: the process, the tools involved, how often it happens and what goes wrong today. We usually reply within one business day and set up the scope call.