The work your team repeats, done by a system.
Copying a lead from a form into the CRM. Chasing an approval. Building Monday's report from four exports. If a person does it the same way every time, it should run itself, with a human stepping in only where judgement is needed.
What it is
Workflow automation is the plumbing between the tools you already use. A form is submitted, a message arrives, an invoice lands in the inbox, a row changes in a sheet; the automation picks it up, enriches it, decides what to do by rules you set, pushes it into the right place and tells the right person. It runs on a schedule or on an event, it retries when a tool is down, and it logs every step so you can see exactly what happened to any item.
The "AI" part is used where it earns its place: reading a messy email or PDF into structured fields, classifying an enquiry, drafting a reply for a person to approve, matching records that do not quite match. It is not sprinkled on for its own sake. Most of a good automation is still careful, boring logic.
What we build
- Lead routing. Form, ad or WhatsApp enquiry → enrichment → scoring → the right rep is notified, the CRM record is created, and the follow-up sequence starts.
- Invoice and document processing. Extract fields from PDFs and photos, match to purchase orders, route mismatches to an approver, post the rest to your books.
- Reporting. Pull from ads, CRM, billing and ops, reconcile, and deliver the numbers to Slack, WhatsApp, a sheet or a dashboard, on time, every time.
- Onboarding and offboarding. New customer or employee → accounts created, documents sent, tasks assigned, reminders chased until done.
- CRM hygiene. Deduplication, stage updates from real activity, stale deals flagged, follow-ups scheduled so nothing sits forgotten.
- Scheduled jobs and syncs. Two systems that should agree, kept in step without anyone exporting and importing.
How it works
- Map the task with the people doing it
In Discover we watch the work as it happens today and write down every trigger, tool, decision and failure. The map is a one-page document you can correct before anything is built.
Week 1 - Design the flow and the exceptions
We decide which steps are rules, which need AI, and which need a person. Exceptions get an explicit path: a queue, a Slack message, a review screen. Scope and price are fixed here.
Week 1–2 - Build with weekly working demos
The flow goes live on a test copy first. Each week you see real items moving through it and we adjust the rules together.
Week 2–4 - Go live, monitor, hand over
Switch over with the old process still available, watch it for the first fortnight with alerts on every failure, then hand over with documentation and a recorded walkthrough.
Week 4+
Typical deliverables
- The automation itself, running in your n8n, Make or cloud account
- A run log and alerting so a failed item is never silent
- A review queue or approval step wherever a person must decide
- Written documentation of every rule, plus a recorded walkthrough
- Test cases for the edge cases we found in Discover
- All credentials, accounts and code owned by you
Who it is for
- Teams of two or more people repeating the same cross-tool task weekly or more
- Businesses where leads, orders or documents arrive in volume and get lost or delayed
- Owners who want the Monday numbers without a person building them
Who it is not for
- Tasks that happen a few times a year and take an hour; a checklist is cheaper
- Processes nobody can describe yet; Discover can help, but the process has to exist first
- Anyone expecting the automation to make the hard judgement calls on its own
Tools we use
What people askabout automation.
Do we have to change the tools we use?
Almost never. The point of automation is to connect what you have. We only suggest replacing a tool when it has no way to connect and that is the real bottleneck.
What happens when something breaks?
Every automation we ship logs each run and alerts a named person on failure, with the item and the step that failed. Items are retried where safe and parked for a human where not. Nothing fails silently.
How is "AI" used, and can it make mistakes?
We use language models for reading and classifying messy input, never for irreversible decisions on their own. Anything with money or a customer commitment attached goes through a person or a confidence threshold, and we test the model on your real examples before go-live.
What does it cost to run?
Usually a few thousand rupees a month in tool subscriptions and API usage, which you pay directly. We tell you the expected running cost in the proposal.
Can our own team maintain it?
Yes. Flows are built in tools with visual editors where possible, documented rule by rule, and handed over with a walkthrough. A retainer is available if you would rather we keep watch.
Put a number on the manual version.
People × hours × cost × the share we can automate. The calculator gives you a yearly figure from your own inputs in under a minute.