WhatsApp · AI agents · n8n · RPA · documents
Your business runs 24/7.
Your team doesn't.
We build AI employees that answer your WhatsApp, read your documents and run the back office — scoped to one process, quoted at a fixed price, delivered to your repository. Eleven years in enterprise IT, nine of them keeping other people's bots alive after the consultants left.
Start here
Most automation projects fail. You already suspected that.
Four numbers from independent research — not from us. If you have been pitched AI before and it went nowhere, this is why, and it is the reason we work the way we do.
of generative-AI pilots produced no measurable return on the P&L.1
MIT Project NANDA, 2025of companies scrapped most of their AI initiatives in 2025 — up from 17% the year before.2
S&P Global, n=1,000+of initial RPA projects fail outright, on EY's own observation.3
EY, Get Ready For Robotsof executives report high regret over their biggest tech purchase of the last two years.4
Gartner, n=1,12060-second self-check
Tick every one that happens in your business.
Nothing is sent anywhere. This is the same list we work through on the audit call — most teams tick four or more.
What we build
Six services. Every one with a price, a date, and a list of who it is wrong for.
We scope narrow on purpose. If a process is unstable or an off-the-shelf tool already covers it, we say so on the audit call rather than after the invoice.
WhatsApp AI AutomationMost asked for
A WhatsApp number that answers, qualifies and books without a person reading every message first. Business API provisioning and Meta verification, conversation design reviewed against your actual message history, human handoff with full context when the agent is unsure, CRM write-back, and two weeks of post-launch tuning.
AI Agent Development
Agents scoped to one job with defined inputs, a defined output, and a human checkpoint where the cost of being wrong is high. Process mapping with the people currently doing the task, model selection and prompting against your own data, confidence thresholds, refusal paths, audit logging, and an evaluation dataset you keep.
n8n Workflow Builds
Self-hosted on your infrastructure, versioned in your repository. Workflows exported as JSON to Git, error handling that identifies the exact step that failed, credentials managed separately from the workflows, and a runbook documenting the failure modes for each one.
UiPath & RPA
New bots where the rules are stable, and rescue work on bots that keep breaking when a UI changes. Process discovery and feasibility assessment, attended or unattended development in UiPath or Power Automate, selector hardening, Orchestrator setup with exception queues, and a handover that lists the known issues instead of hiding them.
Document Processing
Invoices, claims, KYC packets and contracts turned into structured data — with a confidence threshold and a review queue for everything below it. Extraction tuned on your documents rather than a demo dataset, per-field confidence scoring, validation against master data before anything is written, and accuracy measured on a held-out set.
RPA to AI Migration
Moving the brittle parts of an existing estate to AI models while leaving the parts that work alone. Audit ranked by maintenance cost, a keep / migrate / retire recommendation with reasoning for every bot, one bot at a time, parallel running until parity, and a licence reduction plan where one applies.
WhatsApp AI automation
WhatsApp already works for you. It is just completely manual.
It gets read. It gets replies. And every one of those replies costs a person's attention. Research on inbound leads has found average first-response times of around 42 hours, with roughly a quarter of enquiries never answered at all.5 An agent on your number closes that gap without pretending to be a human.
- Answers, qualifies and booksConversation design built from your real message history, not a generic script.
- Collects documents in the chatReads the photo the customer sends, spots what is missing, asks for exactly that.
- Hands over with full contextWhen it is unsure, a named person gets the thread and the reason it stopped — it does not guess.
- Templates handled properlyMeta verification, template categories and opt-in rules are part of setup, not your problem after launch.
One thing we will not tell you
Every WhatsApp agency quotes a 98% open rate. That figure originated in a vendor's own marketing and has never been published with a methodology or a sample size. Independent operator data puts real-world read rates closer to 60–80%, and far lower on unsegmented broadcasts.6 We would rather measure yours after launch than quote a number neither of us can check.
Built on the official WhatsApp Business Platform. Meta bills per delivered message; rates differ by country and India's marketing rate changed on 1 January 2026 — we walk you through your actual cost at audit.7
How an engagement runs
Four steps. The first one is free and the last one is you not needing us.
No discovery phase that bills for three months. No transformation programme. One process, taken all the way to done.
Audit
Forty-five minutes on your most painful process. You leave with a straight answer on whether automating it is worth doing — including when the answer is no.
45 min · freeScope and fixed price
One process, one price, one delivery date, written down before any work starts. If the scope changes, the price conversation happens before the work, not after.
fixed, in writingBuild in the open
Repository access from day one and a working demo every week. You watch it being built, so there is no reveal at the end that turns out to be a slide deck.
weekly demosHandover and exit
Documentation and a runbook as standard deliverables, not upsells. Thirty days of defect fixing included. Ongoing support exists but is optional — by design.
30 days includedThree columns, one honest look
Doing it manually. Hiring a typical AI agency. Working with us.
The middle column is not a strawman — it describes how a large part of this industry actually operates, and it is why so many of these projects end up abandoned.
| Where it shows up | Manual, today | A typical AI agency | Agentic AI Automation |
|---|---|---|---|
| Who owns the code | Nothing to own — it lives in people's heads | Their platform, their account, their terms | Your repository, your cloud, your credentials |
| What it costs as volume grows | Another hire, and another after that | Metered per task; the bill climbs quietly | Self-hosted, no per-task metering from us |
| Price certainty | Predictable, and permanently high | Time and materials, scope creeping | Fixed price and a date, before work starts |
| Reply to an enquiry | Next working day, if someone spots it | Instant, but a scripted bot that annoys people | Seconds, from your real message history |
| When it is unsure | A person asks a colleague — eventually | Guesses confidently and you find out later | Confidence threshold, refusal path, review queue |
| Documents extracted | Typed by hand from a photo on a phone | OCR on a demo dataset; you re-check everything | Tuned on your documents, accuracy measured on held-out data |
| When something breaks | Somebody notices, days later | Silent failure — no error, nothing logged | Step-level error handling and a runbook per workflow |
| Proof it worked | Nobody is measuring | Anonymous case studies and round numbers | Measurement agreed at scope; named case studies only |
| The day it ends | The person leaves and the knowledge goes too | You lose access to what you paid for | Documentation, runbook, and everything already yours |
Ask any vendor these
Ten questions worth asking before you sign anything. Ours are answered below.
Print this list and put it to every agency you speak to, us included. The answers tell you far more than a portfolio does.
Where we have done this before
Six sectors, and the one process we would start with in each.
Banking, healthcare, law, logistics and manufacturing — the sectors behind eleven years of enterprise IT and nine years of RPA across India, the UK and the Middle East.
Healthcare & clinics
E-commerce & D2C
Law firms
Manufacturing
Logistics
Professional services
What it costs
Published, because you should not have to sit through a call to learn the range.
Every engagement is quoted at a fixed price against a written scope. These are the bands almost everything lands inside.
A single workflow taken end to end. WhatsApp automation and most n8n builds sit here. Typically delivered in 1–3 weeks; a narrow single-process build can be 7–10 days.
Several connected processes across a function — document pipeline plus review queue plus system write-back, or a bot estate migration. Typically 4–12 weeks.
Monitoring, tuning and change requests. Genuinely optional — 30 days of defect fixing is already included, and the runbook is written so your team can take it on instead.
Typical delivery is four weeks. Model and messaging usage is billed to you directly by the provider at their published rates — we take no margin on it, and we walk you through the real monthly figure on the audit call before you commit to anything.
We have no case studies published. That is deliberate, and it is temporary.
Our rule is that a case study names the client, states how the result was measured, and carries written approval from that client before it goes online. Two engagements are in that approval process now. Until they clear it, this page stays empty rather than filling up with anonymous "a leading NBFC" stories and round numbers nobody can check.
What we can do instead: on the audit call we will talk through comparable deployments in detail under NDA, including the parts that went badly. Given that 95% of AI pilots return nothing measurable1 and more than half of tech buyers regret their last big purchase,4 we would rather earn the first hour of your scepticism than spend it.
Who you are actually hiring
A small senior team in Faridabad. No branches, no bench.
Not a staffing agency, not a reseller, not a large consultancy. We take a limited number of engagements at a time because the work is done by the people you meet on the audit call.
Remote delivery with overlapping working hours by arrangement. One office, in Faridabad, Haryana.
- 11 yrsin enterprise IT — banking, healthcare and logistics.
- 9 yrsbuilding RPA and automation for banks, hospitals, law firms and logistics operators across India, the UK and the Middle East.
- ToolsUiPath, Power Automate, n8n, and the AI model layer on top of them.
- WhyMost of those nine years went on the unglamorous half of automation: keeping bots alive after the consultants left. This company exists because that half is where projects are actually won or lost.
- We say noTo unstable processes, to work an existing tool already covers, to anonymous case studies, and to anything that would make you permanently dependent on us.
Straight answers
The questions that come up on nearly every call.
Will this replace our staff?
Not in the engagements we run. We automate the copy-paste half — the retyping, the chasing, the "did anyone reply to this" half. Surveys put manual data entry alone at nine-plus hours per employee per week, with around 60% of people reporting burnout from repetitive data work.9 That is the part we take. Judgement work still needs the person, and it gets more of their attention once the retyping stops.
We tried automation before and it broke. Why is this different?
Because that is the problem we were hired to fix for nine years before starting this. The failure mode is almost never the build — it is silent failure after handover: a schedule that stops firing, a token that expires and returns a polite empty response, a screen that changes and a selector that no longer matches. Every build we deliver has step-level error handling, an exception queue and a runbook that names the failure modes per workflow. Nothing goes out with only the happy path tested.
Do we have to replace our existing software?
No. Agents work on top of what you already use — your CRM, ERP, accounting software, shared drives and your existing WhatsApp number. Replacing a working system is usually the most expensive possible way to fix a process problem, and we will say so.
What if the agent gives a customer wrong information — who is liable?
You are, legally, and that is exactly why we build the way we do. In February 2024 a tribunal in British Columbia rejected an airline's argument that it was not responsible for what its own chatbot told a passenger.10 Every agent we ship has a confidence threshold, a refusal path and a review queue, and the scope of what it is permitted to state on its own is agreed with you in writing before launch.
How long does it take?
Typical delivery is four weeks. A narrow single-process build can be 7–10 days. UiPath work runs 3–8 weeks and a bot-estate migration 6–12. You get the specific date in the scope document before any work begins — for context, Deloitte's survey of 400 firms found 63% of RPA projects missed their delivery deadlines,11 which is precisely why we commit to one in writing.
Is n8n really better than Zapier or Make?
For some things, and not for others. Self-hosted n8n makes sense when you are paying meaningfully per task, when you need data to stay on your own infrastructure, or when you want the workflows in your Git repository. For a simple two-step integration, a stock SaaS connector is cheaper and we will tell you to use one.
Can you work with our data protection and compliance requirements?
It is settled at scoping, not discovered afterwards. Everything runs in your accounts, in a region you pick, with access scoped to the specific systems a workflow needs and every action logged. We work across India, the UK, the US, the UAE, Singapore and Saudi Arabia, so residency questions are a normal part of the conversation.
What does the free audit actually involve?
Forty-five minutes on one process — the one that hurts most. We map how it really works, not how the process document says it works. You leave with a straight recommendation, including the times the recommendation is "don't automate this, fix the process first" or "the tool you already pay for does this."
Start with one process
Show us the job nobody in your team wants to do.
Forty-five minutes, no charge, no deck. You will get a straight answer on whether it is worth automating — and if it isn't, you will get that answer too.
Follow the build
We publish what we learn, including the failures.
Workflow teardowns, honest post-mortems, and the odd rant about per-task pricing. Pick your channel.
Sources for every figure on this page
- MIT Project NANDA, The GenAI Divide: State of AI in Business 2025 — 95% of generative-AI pilots showed no measurable P&L return. Coverage
- S&P Global Market Intelligence, 2025, n=1,000+ — 42% of businesses scrapped most AI initiatives, up from 17% in 2024. CIO Dive
- EY, Get Ready For Robots — "we have seen as many as 30 to 50% of initial RPA projects fail." PDF
- Gartner, survey of 1,120 executives — 56% report high regret over their largest tech purchase in the last two years. Coverage
- Oldroyd, McElheran & Elkington, The Short Life of Online Sales Leads, Harvard Business Review, 2011 — 2,241 US companies audited; average first response 42 hours; 23% never responded. HBR
- Analysis of the "98% open rate" claim, which originates in vendor marketing with no published methodology; operator data across 40 campaigns puts real read rates at roughly 60–94% depending on list quality. Analysis
- Meta, WhatsApp Business Platform pricing — per-delivered-message billing since 1 July 2025; rates vary by country; India marketing rate increase effective 1 January 2026. Meta docs
- Randomised trial, 54,066 visits across 25 primary-care clinics — no-show rate 4.4% with two reminders vs 5.3–5.8% with one (P<.001). AJMC
- QuestionPro for Parseur, July 2025, n=500 US professionals — 9+ hours per week on manual data transfer; ~60% report burnout from repetitive data tasks. Release
- Moffatt v. Air Canada, BC Civil Resolution Tribunal, 19 February 2024 — airline held responsible for information its chatbot provided. CBC
- Deloitte Global Robotics Survey, 400 firms — 63% did not meet RPA delivery deadlines. Forbes