Onboarding 3 workflow builds for Q1 2027
Check availabilityAgentic operations studio · India
We design, ship, and run AI workflows for Indian businesses — WhatsApp lead follow-up, invoice processing, support triage, and the hundred small ops that quietly eat your week.
GuardrailThe agent drafts. A human approves anything with money or a promise attached.
The gap
Most Indian businesses don’t have a software problem. They have a coordination problem: a lead arrives on WhatsApp, gets copied into a spreadsheet, chased by a human, invoiced in Tally, and reported in a deck. Every handoff leaks time.
The goal isn’t fewer people. It’s fewer things a person has to remember.
Hiring is the default answer to more volume. Margins flatten as the team scales, and the founder becomes the routing layer for every exception.
A chatbot that works on a laptop rarely survives real data, permissions, edge cases, and the one customer who replies in Marathi at 11pm.
Off-the-shelf SaaS assumes a workflow you don’t have. Custom glue is fragile, undocumented, and leaves when the developer does.
What we build
Not “AI transformation.” A specific process, a clear owner, a number that moves. Each of these can be scoped as a three-week pilot.
Before: leads sit in WhatsApp and a spreadsheet until someone remembers. After: qualified, routed, and followed up in minutes — CRM clean.
Before: every ticket waits for a human to read, tag, and route it. After: order lookups answered, refunds routed, edge cases escalated with context.
Before: PDFs downloaded, line items retyped, PO mismatches found at month-end. After: extracted, matched, flagged, and drafted into Tally for approval.
Before: a delayed order is discovered by the customer. After: delays, stock-outs, and dispatch failures detected, explained, and actioned.
Before: three spreadsheets, two WhatsApp groups, one person who knows the formula. After: numbers pulled, reconciled, narrated, and sent for approval.
Before: 400 resumes, 60 unread messages, interviews booked by hand. After: screened against the role, scheduled, and kept warm — human decides.
Outcome ranges are from illustrative composite engagements. Your baseline is measured in the audit.
Browse all 12 workflow specsWays to start
No packages. We start with a free audit, then quote hours after we’ve seen the work — with a cap you approve before we start.
01 · Free
Bring one workflow. We map the process, volume, tools, and exceptions, and tell you if it’s worth automating.
Google Meet · IST
No deck. No written roadmap. An honest go/no-go.
02 · Written
If it’s a fit, we send a one-page estimate: hours, rate, scope, exclusions, and the ceiling.
No charge for the proposal
The cap is the fixed number. The hours are the estimate.
03 · Weekly
We bill hourly against the cap, with a weekly update. If we hit the ceiling, we stop and re-scope.
4-hour minimum · weekly billing
You can stop at any weekly block; the cap is not a commitment to spend it.
The audit is free. Paid work is ₹6,000/hour + GST, billed weekly against a written cap. WhatsApp/Meta usage, AI model APIs, and third-party subscriptions are external and not included. We’ll tell you if the ROI isn’t there — before you approve the work.
Illustrative outcomes
Three composite examples, built from the kinds of workflows we scope most often. Names and numbers are illustrative; the mechanics are real.
Inbound messages were answered by whoever was free. The agent now replies in under three minutes, checks stock, quotes, and routes high-intent buyers to a human.
Two accounts people spent three days a month retyping invoices. Extraction, PO matching, and exception flags now run before 9am; a human approves the entries.
The helpdesk was a queue of “where is my order?” and “how do I reset?”. The agent answers from docs and order data, tags the rest, and hands over with full context.
These are illustrative composite scenarios for design and discussion — not claims about named clients. We will publish verified case studies, with permission, as engagements complete. Ask us for references on a call.
How we work
We sit with the people doing the work. Map the triggers, handoffs, exceptions, and the cost of each run. Pick one workflow with a clear baseline.
Build the smallest version that can run in production. Real data, real users, real permissions. One workflow, not a platform.
Add evals, approvals, retries, cost caps, and logs. Break it on purpose. Document what happens when the model, the API, or the human fails.
We watch the runs, you watch the numbers. Monthly care plan: monitoring, fixes, new edge cases, and a quarterly re-score of the ROI.
Guardrails
An agent without guardrails is a liability with an API key. Every workflow we ship has a human checkpoint, an eval set, and a rollback path.
Read the approachDiscounts, refunds, contracts, and anything a customer could screenshot get a checkpoint. One tap, no login.
A golden set of real cases, including the ugly ones. We measure accuracy, cost per run, and escalation rate — not vibes.
Scoped API keys, service accounts, and revocable credentials. No shared passwords, no “just give it admin”.
Inputs, tool calls, outputs, and approvals are logged. When something looks wrong, we can reconstruct exactly what happened.
Model routing, retries, queues, and a cheap deterministic path for the common case. No surprise API bills.
Data minimisation, retention rules, DPAs, and India-region hosting where it matters. We don’t train models on your data.
The math
A directional model, not a quote. Move the sliders to see the annual value of the hours your team spends on work that follows rules — and an indicative hourly fee.
Your operationsDirectional estimate
Salary + benefits + overhead ÷ productive hours. ₹500/hr ≈ ₹8.7L CTC.
If a new person could be trained on it in a week, it’s a candidate.
One workflow is usually 50–70 hours; a 3-workflow build 120–160.
Estimated annual value
of reclaimable capacity, at your loaded cost
Indicative fee = build hours × ₹6,000. The actual cap is agreed after the free audit. Excludes GST and external costs.
“We’re not here to sell you a model. We’re here to remove a recurring headache — and leave you owning the system.”
Founder, kumoworks · Bengaluru
The studio
kumoworks is a senior-only agentic operations studio. We don’t sell transformation programmes or seat licences. We pick a workflow that is costing you money, build it, prove it, and then decide whether to scale it.
We’ve spent our working lives inside product, operations, and engineering teams — close enough to know that the last 20% is where automation projects die. So we ship the last 20%: the exception handling, the approvals, the audit trail, the handover doc.
We work with a small number of clients at a time. That’s not scarcity marketing; it’s capacity.
More about the studioWhat clients say
We stopped losing leads in the WhatsApp inbox. The agent replies in under three minutes, and my team only sees the ones worth a call.
The invoice workflow paid for itself in the first month. What took two people three days now takes one person an afternoon.
What I appreciated was the honesty. They told us not to automate two of the five workflows we asked for.
Sample content. These quotes and personas are placeholders to show the layout. Replace them with verified client quotes (with written permission) before launch. We don’t publish invented endorsements.
Pricing
The first 30 minutes are free. After that, you pay for hours — with a written cap, not a package.
One workflow, quick map, feasibility, rough hour range, go/no-go.
Design, build, integrations, evals, handover.
A written ceiling based on the estimated hours.
WhatsApp/Meta, AI models, SaaS, hosting, payments.
No packages. The cap is the fixed number; the hours are the estimate. External costs are separate. All prices exclude 18% GST.
See what’s includedQuestions
Still unsure? Ask on the call. We’d rather talk you out of a bad automation than sell you one.
Book a callWe’re India-based and India-first. Most of our work is with Indian SMBs and mid-market teams. We also take remote work from the Gulf, UK, and Singapore. Pricing is in INR for Indian entities; international contracts are in USD.
No. It removes the parts of their job they shouldn’t be doing. Every workflow we ship has a human checkpoint for money, promises, and people decisions. The goal is capacity, not a smaller payroll.
We design for failure: approval gates, deterministic fallbacks, retries, and an audit log. If the model is unsure, it routes to a human. You can replay every run. We measure error rates before go-live and monitor them after.
Book a free 30-min audit now; paid work can start within a week of the signed statement of work. A one-workflow pilot is typically 3–4 weeks if your systems have APIs. If a critical tool has no API, we’ll tell you before you pay — and suggest a workaround or a different workflow.
Usually not. We work with WhatsApp Business, Zoho, Tally, Razorpay, Shopify, HubSpot, Google Workspace, Freshdesk, and most systems with an API. We can also build a small internal tool where the gap is real.
The 30-minute workflow audit is free. Paid work is ₹6,000/hour + 18% GST, billed weekly against a written not-to-exceed cap. There are no packages or tiers. The cap is agreed after the audit. External costs (WhatsApp/Meta, AI model APIs, SaaS/hosting) are separate.
Fixed pricing is more common and often converts better, but a fixed price before the audit is a guess. The cap gives you a fixed ceiling; the hourly rate keeps the scope honest. If the audit shows a stable scope, the cap is the fixed number.
No. Our fees cover design, build, testing, documentation, and handover. WhatsApp/Meta message and conversation charges, BSP platform fees, AI model API usage, CRM/ERP subscriptions, hosting, and other third-party services are external. You pay them directly where possible, or we bill at actuals with your approval. WhatsApp is not free; Discord, Signal, and Telegram have no per-message platform fee, but hosting or bridge costs may apply.
Yes. We build on WhatsApp Business, Discord, Signal (via bridge), Telegram, email, web forms, Slack, and voice. Signal support uses a bridge and a dedicated number; reliability depends on the bridge. We’ll tell you the trade-offs before you commit.
We use least-privilege access, minimise the data we touch, and sign an NDA and DPA. We don’t train models on your data. Where it matters, we host in India or inside your cloud account. You can revoke access at any time.
Both are useful. ChatGPT is a great assistant; it isn’t an operations system. No-code tools are great for simple triggers. The hard part is the last mile: permissions, exceptions, approvals, cost control, and monitoring. That’s the part we build.
3 slots open for Q1 2027
A free 30-minute audit. No deck, no jargon. We map one process, estimate the hours it costs you, and tell you honestly whether an agent is the right fix.