AI Automation Services in Nepal, Built on n8n
Someone in your office copies enquiries from Facebook into a sheet, chases them on WhatsApp by hand and builds the same report every Monday. This is the work that turns those repeated jobs into n8n workflows that run on their own, in accounts you own.
Animated n8n-style workflow: a trigger node fires, a data packet travels through a parser, splits across two parallel steps, merges into a final action and reaches the output, with every node ticked as it completes, showing AI automation as a workflow that runs itself node by node.
I'm Shashikant Shah, a digital marketing expert in Kathmandu and a software engineer. AI automation on n8n is the building half of what I describe on my AI automation expert page (and the business case is in AI automation for business in Nepal): you bring one process, I scope it, build the workflow, test the awkward cases and hand it over. It's for businesses in Nepal with a repeated job that follows rules: lead handling, follow ups, reporting, moving data between tools. Most often it automates the leads and reports my SEO and ads work produces.
Short answer: AI automation with n8n in Nepal means one defined workflow at a time, running in your own n8n instance, connected to your CRM, WhatsApp, email and sheets, with an OpenAI step only where language is involved and a person approving anything before it reaches a customer. You get the workflow, the exports and the documentation, and you can run it without me.
When you need this
A college in Kathmandu gets enquiries from a website form, a Facebook lead form, WhatsApp and email. They sit in four places until someone checks, and by the time the call goes out the student has spoken to two other colleges. Reply speed decides it, and one person can't be that fast across four inboxes.
The other version is the report. Every Monday someone opens Google Ads, Meta, Analytics and the sales sheet, copies numbers into a slide and sends it by eleven. It takes three hours, it's wrong once a month, and nobody reads it until Wednesday.
Then there's publishing: the same offer posted to the Business Profile, Facebook, Instagram and a WhatsApp broadcast, by hand, every time. None of these jobs need judgement. They need someone who never forgets.
The problems I see most often:
- Enquiries in four inboxes, answered in hours or days, when the first reply should go out in minutes.
- Zapier or a no code tool that worked until volume grew and the monthly bill went past what the job is worth.
- An automation someone built that failed silently in Falgun, and nobody knew until the leads stopped.
- A model sending replies to customers with nobody checking, and one confidently wrong answer that cost a client.
- Nothing documented, so the person who built it is the only one who can touch it, and they've left.
What I actually do
- Lead intake to your CRM: website form, Facebook lead forms, WhatsApp and email land in one pipeline, deduplicated, labelled and routed to the right person, with a first reply inside minutes. It ties into the CRM work when you don't have one yet. You get one list, not four.
- Follow ups on WhatsApp and email: reminders, quotes, and the second and third message most businesses never send, drafted for a person to approve where the message matters. You get fewer leads going cold.
- Report generation: Google Ads, Meta, GA4 and your sales sheet pulled into one scheduled report, with the odd numbers flagged in plain words. The tracking underneath is the analytics work. You get Monday morning back.
- Scraping and data to sheets: prices, listings, stock or enquiries collected on a schedule into Google Sheets, validated before they land. You get data that's current without anyone typing it.
- OpenAI steps with human review: classifying enquiries, pulling fields out of invoices and CVs, drafting replies, each checked against a schema and queued for a person before anything with consequences happens. You get the speed without the embarrassment.
- Reliability built in: retries, timeouts, duplicate protection, run logs and a failure alert to your phone, so if a workflow stops you know that day. Colleges can see how this fits on the education page.
- Documentation and handover: workflow exports, the credentials used, schedules, recovery steps and a walkthrough, so another developer can run it without me. You keep everything.
How it works
First call and scoping. The first thirty minutes are free. You describe one process: what triggers it, which tools it touches, how many times a month it runs and the exceptions a person handles today. I tell you whether it's a good candidate, and some aren't. If you go ahead, you get a written scope: the boundary of the workflow, the approval points, the acceptance checks and what's not included.
The work. One narrow workflow first, built end to end over two to four weeks, in your n8n instance, self hosted or n8n cloud, on your own provider accounts. I test the normal cases, bad input, a provider outage and the human approval path before it goes live. Where a node can't do it, I write the code inside the workflow rather than leave the platform.
Reporting, and what you keep. After go live you get a run log you can read and a short monthly note on runs, failures and cost per run. You keep the n8n instance, the workflow exports, the custom code and its repository, the API keys and the hosting, all in your name. Ongoing maintenance is optional, and never a condition of handover.
The first pilot, and where it stops
I start small on purpose. The pilot is one workflow with one trigger, the tools it has to touch and one approval point. The written scope sets its boundary before I build anything:
- What starts the workflow, and what it does on a normal run.
- What it does with bad or missing input.
- Who approves what, and what never goes out without a person.
- The checks it must pass before it goes live.
- What is not included.
Anything outside that boundary is a second workflow. I quote it separately, after the first one has run on real work and you have seen what it does.
An example: one enquiry, from form to follow up
This is an illustration of the pattern, not a client result. It shows what a lead routing workflow does with a single website enquiry. CRM stands for customer relationship management: the tool where your team tracks leads and customers.
- Trigger: a visitor submits the enquiry form on your website.
- Validation: the workflow checks that the name, the contact detail and the message are present and usable. An incomplete entry goes to a review queue; it is never dropped.
- Duplicate check: it looks for the same phone number or email in your CRM, so one person who writes twice does not become two leads.
- Record: it creates or updates the CRM record, with the source and the page the enquiry came from.
- Notification: the person who owns that kind of enquiry gets a message with a link to the record.
- Exception: if the CRM is unreachable or a step fails, the run retries, then alerts a person with the original enquiry attached. Nothing is lost silently.
Workflow management after the build
An automation that nobody watches breaks quietly. So part of every build is the management around it: an error alert to your phone or Slack when a run fails, a log you can read without me, version control for the workflow itself, and a one page document that says what it does, what it needs and who to call. On a retainer I check the runs, the failures and the cost of the AI steps once a month, and we decide together whether to extend the workflow or leave it alone. If you have someone in house, I hand the management to them and stay on call.
Proof from AI automation work
Goenka Gold, Red Mud Coffee, Shree Ridhi Sidhi Jewellers and FoodMario are on my case studies page with automated publishing in the scope I delivered, alongside profile setup and offer campaigns. That work was automated publishing, not lead routing in n8n, and I list it as exactly that: a repeated job done by a workflow instead of a person.
This website shows how I build and test. Its build pipeline runs 15 automated checks on every deploy, and I built that pipeline myself. It is a build pipeline, not an n8n workflow, so read it as evidence of method. I haven't published an n8n case study with numbers yet, and I won't invent one. When I do, it will show the hours before, the hours after and the date, like every other figure on that page. How I decide what to automate first is in my article on how automation saves businesses 20 hours.
What AI automation costs
An n8n workflow is quoted per project after the first call, because the scope decides the price. Your accounts and the agreed project assets stay under your control. My published retainer for ongoing work is USD 1,800 a month, about NPR 245,000, and the audit and sprint prices are on the pricing page. The first thirty minute call is free, and if a Zapier template or a shared sheet does the job at your volume, I'll say so and you won't need me.
A workflow also has running costs after the build. You pay each one directly to the provider, from your own account, and I add no markup:
| Running cost | What it is | Who you pay | What moves it |
|---|---|---|---|
| n8n hosting | Your own server, or an n8n cloud plan | The hosting provider or n8n | The plan you choose and how often the workflows run |
| Language model usage | OpenAI or another model, only on steps that read or write language | The model provider, on your own key | How many runs use a model and how much text each one handles |
| Messaging | The WhatsApp or email sending service, if the workflow sends messages | The messaging provider | How many messages go out |
| Other subscriptions | The CRM, sheets and other tools the workflow connects | Each provider | The plans you already have |
| Maintenance | Checking runs, failures and cost each month | Me, only if you want it | The scope |
I work the running cost out during scoping. I count how often the workflow runs in a month and which steps call a model or send a message, then write each line into the scope at the provider's current published price. After go live, the monthly note shows the real cost per run next to that estimate.
n8n in Nepal: who builds it and what custom automation means
Who builds n8n workflows in Nepal?
I do. I'm Shashikant Shah, a software engineer in Kathmandu, and I build n8n workflows for businesses across Nepal and abroad. People search for a top n8n expert in Nepal, but nobody ranks n8n builders. Ask anyone you shortlist to show a workflow diagram, how errors get reported and who owns the server afterwards.
Is this an AI automation agency or company in Nepal?
No. People search for an AI automation agency in Nepal or an AI automation company, and I'm neither. I'm one consultant who scopes, builds and maintains the workflow myself, which keeps the handover simple. If you need a team working on many workflows at once, an agency is the better fit and I'll say so on the first call.
What does custom AI workflow automation mean on n8n?
Custom AI workflow automation means the workflow follows your own steps, not a template. On n8n I connect your form, CRM, sheets and messaging, add code where the logic is specific to you, and call a language model only on the steps that need it. You host it, so you control the data and the running cost.
How do you track what an AI workflow costs to run?
Every run is logged with the steps it used. For AI workflow cost tracking I record how many runs called a model, how much text each one handled and what the provider charged. The monthly note shows the real cost per run next to the estimate from the scope, and you see the provider's bill on your own account.
Which workflow automation tools do I build with?
I build on n8n, a workflow automation tool you can host yourself, with OpenAI or Claude on the steps that need reading or writing. Zapier or Make can suit a simple two step job, and I'll say so. I choose n8n when you want to own the server, keep the data in your own account and control the running cost.
Do you sell workflow management software?
No. I design and build workflows for your business. I don't sell or resell a workflow management system. If you only need to assign tasks and see who is late, an off-the-shelf task board does that and a custom build isn't worth it. A build pays when the same data has to move between your form, CRM and messages every day.
Can n8n automate lead follow up and WhatsApp replies?
Yes. An n8n workflow can put every enquiry in one list, send a first reply and remind your team when nobody has followed up. WhatsApp replies go through the official WhatsApp Business platform on your own number, and messages about prices, complaints or refunds go to a person, not the workflow.
Questions about AI automation
What is n8n, and why use it instead of Zapier or Make?
n8n is a workflow tool you can host yourself, put your own code inside, and run at volume without paying per task. If your process is simple and runs a few dozen times a month, Zapier or Make is often the better answer, and I'll tell you that rather than build something heavier than the job.
What happens when the AI gets something wrong?
It will, sometimes. That's why every model output is checked against a schema before anything acts on it, and anything that touches money, customers or public posts waits for a person to approve. The workflow is built so a wrong answer gets caught, not sent.
Is my data sent to OpenAI?
Only the fields a given step needs, and only if that step uses a language model at all. Most of a workflow is plain rules with no model involved. Where your data can't leave your own servers, that step stays local or the model is dropped from it.
How long does one workflow take?
Scoping takes about a week, then two to four weeks to build and test the first workflow. Lead intake is the quickest because the tools are standard. Anything that reads documents or needs a new integration takes longer, and the written scope says how long before you pay.
How is this priced?
Per project, after the first call, because the boundary of the workflow decides the price. You pay your own n8n hosting, OpenAI usage and subscriptions directly, so there's no markup on tools. My published retainer for ongoing work is USD 1,800 a month (about NPR 245,000 a month), and the pricing page has the rest.
Which process should I automate first?
The one that repeats most, follows clear rules and costs you money when it runs late. For most businesses that is enquiry handling: getting every lead into one list and a first reply out quickly. Start with one narrow workflow, run it on real work, then decide on the next.
When are you not the right person?
If you want the whole business automated in one go, no. Those projects fail, and I'd rather build one workflow that saves a specific team a specific number of hours a week. If the process changes every week or needs judgement on every case, a person is cheaper than a workflow, and I'll say so.
Pages related to AI automation
All services · AI automation expert in Nepal · AI chatbot and agent development · CRM development · How automation saves 20 hours · Pricing.
Tell me about your workflow
Send me one process: the trigger, the tools it touches and how often it runs. I'll tell you whether it's worth automating and where the human approval belongs, whether or not you hire me. The first thirty minutes are free.
I read every message myself and reply within one working day. If you would rather talk, call +977 9845451010.