Finance Ops
- Invoice matching and exception handling
- Month-end reconciliation prep
- Expense policy checks
AI agents for enterprise operations
Your best people are still copy-pasting between systems.
I design and build AI agents that take the repetitive work out of finance, revenue and customer operations. They work inside the tools you already use, and a person approves every decision that matters.
One call to map one workflow. No deck, no pitch.
- 09:02:11read41 new supplier invoices in AP inbox
- 09:02:13match37 matched to PO + goods receipt in ERP
- 09:02:14post37 queued for payment run, audit trail written
- 09:02:15flag4 exceptions: price variance above 2%
- 09:02:19draftSupplier queries drafted for 3 of 4
- 09:02:20hold1 invoice over approval limit → Finance lead
- 09:02:20waitAwaiting human approval
- 09:14:03doneApproved by Finance lead. Supplier queries sent.
§01 The problem
The work isn’t hard. There’s just too much of it, and it lives between systems.
- 01
Swivel-chair work
Skilled people re-keying the same data from inbox to ERP to CRM to spreadsheet. Every hour spent there is an hour not spent on judgment.
- 02
Pilots that never ship
The demo looks great. Then security review, edge cases and integration stall it. MIT’s 2025 “GenAI Divide” study found about 95% of enterprise gen-AI pilots had no measurable impact on P&L.
- 03
Another tool won’t fix it
Another platform means another login and another migration. The gap is between the tools you already have, not a missing one.
Put a number on it.
Your numbers
What it costs you each year
$243,000
4,416 hours of skilled time, or about 2.4 full-time people, spent moving information between systems.
An agent rarely takes all of this. The call is about finding the part it can take reliably.
§02 Where agents fit
Agents that do the work, not slides about agents.
The best first workflow is repetitive, rule-heavy and spans more than one system. These are typical starting points.
Revenue Ops
- Inbound lead research and routing
- CRM clean-up and de-duplication
- Call notes turned into CRM updates
Customer Ops
- Ticket triage with drafted replies
- Order, refund and status look-ups
- Escalation summaries for leads
Procurement
- Vendor onboarding and document chasing
- Quote comparison against specs
- Contract renewal tracking
Compliance & Risk
- Document review against checklists
- KYC / KYB data gathering
- Audit evidence collection
People Ops
- Candidate screening summaries
- Onboarding tasks across tools
- Policy Q&A over internal docs
Not on the list? If your team does it the same way every day, it’s worth a conversation.Book a call →
§03 How it works
Four steps from “what if” to running in production.
Start with one workflow and prove it on real work. Autonomy is earned one step at a time.
- 01Free
Mapping call
We pick one workflow. I ask how it really runs today, who touches it and which systems it crosses. You leave knowing whether an agent is worth building.
- 02Fixed price
Audit and fixed-price scope
I document the workflow step by step, mark what an agent can own and what stays with your people, and quote a fixed price. No open-ended retainer.
- 03Human approves
Build in review mode
The agent runs on real work but only drafts. Your team approves or corrects every output, and those corrections make it better.
- 04You own it
Narrow autonomy, then handover
Once a step is reliable, the agent is allowed to act on that step alone. You get documentation, monitoring and the code.
§04 Built for enterprise
The questions your security and finance teams will ask, answered up front.
Your data stays yours
Agents can run inside your own cloud account. Model providers are used under no-training terms, and nothing leaves the systems you approve.
People own the decisions
Anything that moves money, touches a customer or changes a record of truth has an approval step until you decide otherwise.
Every action is logged
A full trail of what the agent read, what it decided and what it did, ready for your auditors.
No rip-and-replace
Agents work with your existing ERP, CRM, helpdesk, email and spreadsheets through their APIs. Your team keeps its tools.
Model-agnostic
Claude, GPT or open-weight models, chosen per task on accuracy, cost and your data rules. Swappable later.
No lock-in
The code, prompts and workflows are yours at handover. No platform subscription to me.
§05 Who you’ll talk to
You talk to the person who builds it.
I’m Shreyas. I build software for a living. I’ve founded and shipped my own products, including ThreeCal, Hermi and Mumbai Social Network, and now I build AI agents for operations teams.
There are no account managers and no handoffs. The person on your first call is the person who maps the workflow, builds the agent and supports it after launch.
I take on a small number of clients at a time, so every engagement gets my direct attention.
§06 FAQ
Straight answers.
What does it cost?
Every build is fixed-price, scoped after the audit, so you know the number before any work starts. Running costs (model usage and hosting) are estimated up front too.
How long until something is live?
The first workflow is deliberately narrow so it can run in review mode within weeks, not quarters. Wider rollout follows once it has earned trust on real work.
Is our data safe?
Agents can be deployed in your own cloud. Access is scoped to exactly the systems and fields the workflow needs, model providers are used under no-training terms, and every action is logged.
What happens when the agent gets something wrong?
In review mode, nothing goes out without a person approving it. After that, uncertain cases are routed to a person instead of guessed at, and the log shows exactly what happened.
Do we need to replace any of our tools?
No. Agents sit between the systems you already use. If a tool has an API, or even just an inbox or a shared drive, an agent can usually work with it.
We tried AI before and it didn’t stick.
That’s common. It usually happens because the scope was too broad or the pilot never connected to real systems. Starting with one narrow, measurable workflow is the fix.
What happens after handover?
You get documentation, monitoring and the code. Ongoing support and new workflows are optional and agreed separately.
§07 Next step
Pick one workflow. Let’s see if an agent can run it.
- Bring the workflow that eats the most hours.
- I’ll tell you honestly whether an agent fits, and where it doesn’t.
- You leave with a rough scope, whether or not we work together.
Calendar not loading?Open the booking page ↗