Start with the task, not the technology
Automation projects fail when they begin with a tool. They succeed when they begin with a specific task somebody does the same way every week: a person copying figures between two systems, a report assembled by hand from four exports, an enquiry that sits unanswered because nobody was assigned it. We measure how long that task takes today, automate it, then measure it again. If the saving does not justify the build, we say so before you spend anything, and that conversation happens more often than you might expect from a company selling automation. The reporting version of the same problem, where the output is a weekly figure rather than a record in a system, is covered in internal dashboards people actually use.
- Data entry between systems that do not talk to each other.
- Invoice, receipt and purchase-order processing from PDFs and scans.
- Scheduled reports assembled and delivered without a person.
- Enquiry routing, assignment and follow-up that never gets forgotten.
- Reconciliation between bank, accounting and sales records.
Document automation is where most of the money is
In Nepal especially, an enormous amount of skilled time goes into retyping documents: customs paperwork in Birgunj, purchase orders in Biratnagar, patient forms in Bharatpur, invoices everywhere. Modern extraction handles this well: read the document, pull the fields, populate the record, and flag anything that does not match for a person to check. The value is not only the hours saved but the transcription errors that stop happening, and those are the expensive ones, because a mistyped figure is usually discovered by a customer or an auditor rather than by the person who typed it. Extraction is a model making a judgement, so the interface has to show its confidence and make correction quick, which is the argument in shipping AI features users actually trust.
Following up on Viber and WhatsApp without forgetting anyone
For most Nepali businesses the follow-up that decides the sale happens in a chat app, and it is the single most common place work gets dropped. Somebody replies at nine in the evening, the thread scrolls away, and nobody chases it. We build the routing rather than the conversation: enquiries logged from every channel into one place, assigned to a person, escalated when a reply time passes, and templated messages available so a follow-up takes seconds. The human still writes the message that matters. The system makes sure it is not the message nobody remembered to send. Where the follow-up ends in a payment, which rails to integrate and in what order decides how short that last step can be.
Integrations between what you already run
Most businesses do not need new software; they need the software they already have to stop being islands. We connect accounting packages, CRMs, e-commerce platforms, messaging channels and spreadsheets so a change in one is reflected in the others, using documented APIs where they exist and carefully built adapters where they do not. That second category is common in Nepal, where a widely used local accounting or billing package may have no API at all, and the honest answer is a scheduled export and a reconciliation step rather than a live integration that would be fragile. Where the goal is a weekly number rather than a record written back into a system, data science is the better starting point.
A human stays in the loop
Every automation we build has an exception path: when the system is unsure, it stops and asks rather than guessing. That is what makes automation safe to point at invoices and payments, and it is why our automations tend to still be running two years later. The exception queue is also the best diagnostic you will get, since what lands in it tells you exactly where your process is genuinely ambiguous rather than merely undocumented. Where the judgement genuinely needs a model rather than a rule, that is AI and machine learning, and we will say which one you are actually buying.
What goes wrong in automation, and how we avoid it
- Automating a broken process faster.
- We map the process before automating it and frequently find steps that exist because of a system replaced years ago. Removing a step is cheaper than automating it and we look for that first.
- A workflow held together by one person's login.
- Integrations run on service accounts owned by the business, with credentials in a secret store. Automations built on a personal account break the week that person leaves or changes their password.
- Silent failure.
- Every automation logs what it did, alerts when it cannot proceed, and falls back to the manual path. Work that disappears without an error is far more damaging than work that visibly stops.
- Extraction accuracy assumed rather than tested.
- We run the extraction against a sample of your real documents, including the bad photocopies and the handwritten annotations, and report the measured accuracy before you commit to the full build.
How we run automation work
- 01
Measure
We watch the task being done, time it, and count the volume. That baseline is the business case, and occasionally it is what tells us not to proceed.
- 02
Map
The process written out including every exception the person handles without thinking about it. The exceptions are where automation goes wrong, and they are never in the original description.
- 03
Pilot
The automation runs alongside the manual process on real work, with both outputs compared. Nothing is switched over on the strength of a demo.
- 04
Cut over
The manual path stays available while the exception queue is monitored and the rules are tuned. Alerting and logging go live at the same moment, not afterwards.
- 05
Verify
We re-measure the task after go-live and report the actual saving against the baseline, including where it fell short of what we projected.
What automation costs
Automation is quoted as a fixed scope per workflow, priced against the baseline measurement we take first, so the business case is visible before you commit. Where the work amounts to a connected system across several departments rather than a single task, it is a Growth engagement. Where a business wants a standing capacity to keep automating and integrating as things change, the Dedicated Team tier is the cheaper arrangement.
Growth
Software behind the front door.
- Nepal, from
- Rs 320,000
- About $2,290 at Rs 140 to the dollar
- International, from
- $4,200
- A$6,400 · €3,900
6 to 12 weeks
Companies that need logins, payments, dashboards or an app, not just pages.
Dedicated Team
Your engineering bench, offshore.
- Nepal, from
- Rs 145,000per engineer, per month
- About $1,040 at Rs 140 to the dollar
- International, from
- $2,200per engineer, per month
- A$3,400 · €2,050
Monthly, 3-month minimum
Teams in Australia, the US, the UK, the EU or Nepal outsourcing a squad of developers, ML engineers or DevOps people who work to your roadmap.
What pushes the price up
- Documents that arrive as photographs, faxes or poor photocopies.
- Systems with no API, where data has to be exported and reconciled rather than requested.
- Handwritten or Nepali-language content, where accuracy has to be established case by case.
- Approval chains with several roles and conditional routing.
- Anything touching payments, which needs a stricter review path and an audit trail.
What brings it down
- One workflow first, chosen because it takes the most hours.
- Digital source documents rather than scans.
- Systems that already expose a documented API.
- Accepting a daily batch rather than instant processing.
- Every figure here is a starting point, not a quote. You get a fixed price against a written scope before anything is built.
- Nepali clients are quoted in NPR at local rates, with a VAT invoice against our PAN. International clients are quoted in USD, AUD, EUR or GBP.
- You own the code, the domain and the hosting from day one, on every tier.
Who this is for, and who it is not
Worth a conversation if
- Tasks done the same way at least a few hours a week, by somebody whose time is worth more.
- Businesses processing a steady volume of invoices, orders or forms on paper or as scans.
- Teams losing enquiries in chat threads and inboxes with no single place they land.
- Companies running several systems that require the same data to be typed into each one.
Probably not us if
- Work that takes under an hour a week, where a better form or a clearer procedure wins on cost.
- Processes that change every month. Automating a moving target produces maintenance, not savings.
- Anyone hoping to remove the review step entirely. Our automations are designed around a person checking the exceptions.
- Cases where the real problem is that nobody owns the task. Software will not assign accountability.
Frequently asked questions
How do we know automation is worth it for our business?
Count the hours a week the task takes and who does it. If it is under an hour a week, or the process changes constantly, a better form or a clearer procedure usually beats automation and we will tell you that. We take the baseline measurement before quoting precisely so the answer is a number rather than an opinion.
Will this replace our staff?
In practice it moves them off retyping and onto work that needs judgement. Our automations have exception queues because a person is still the final check, and the businesses that get the most from this are the ones that redeploy the time rather than trying to bank it.
Can you automate work involving Nepali-language documents?
Often, though accuracy depends heavily on document quality and layout. We test on your real documents, including the poor photocopies, and report the measured accuracy before you commit to a full build. Where accuracy is not good enough, we say so rather than shipping something that quietly mistypes numbers.
Our accounting software has no API. Is that a dead end?
Not usually, but it changes the design. Where a system cannot be queried, we work from scheduled exports with a reconciliation step, which is less elegant than a live integration and considerably more reliable than a scripted workaround that breaks at the next update.
What happens when the automation breaks?
It alerts, logs what it was doing, and falls back to the manual path rather than silently dropping work. We agree a support arrangement for fixes, and the logs mean a fix starts from evidence rather than from someone's recollection of what happened.
Can you connect this to Viber or WhatsApp?
We build the routing around those channels rather than trying to automate the conversation itself: enquiries logged in one place, assigned, and escalated if nobody replies within an agreed time. The message that closes a sale is still written by a person, and it should be.
How long does an automation take to build?
A single well-defined workflow is usually two to four weeks including the pilot period, most of which is running it alongside the manual process rather than writing code. Document extraction takes longer because accuracy has to be established on real samples first.
