Run a full eCommerce process with one supervisor, not twenty-two seats.
Akontec allocates a live eCommerce support process to your company. It is delivered by a trained fleet of virtual AI agents running around the clock on your own workspace, and operated from a panel your existing team leader can run on day one. Your floor keeps doing what it already does. This runs beside it.
The conversation your clients are about to start.
Within the next eighteen months, your eCommerce clients will ask what your AI-delivered pricing looks like. They will ask whether you are ready or not, and some of them will ask a competitor first.
Building that capability yourself means a platform, a model layer, guardrail engineering, an evaluation harness and an operations console — eighteen months and a budget that has to be justified before a single rupee comes back.
This is the other route. Akontec has built it, runs it in production, and allocates the client volume. You take one process, on one workspace, with one supervisor, and you are operating an AI delivery line in twenty-one days at the cost of roughly two months of a mid-size campaign's salary bill.
This does not replace your seat business. It sits alongside it, consumes no floor space, and gives you something to show the next client who asks.
A BPO project you already know how to price.
Before any of the technology, understand the work. This is ordinary eCommerce back-office and customer support for sellers trading across Indian marketplaces. Six work streams, measurable output, an SLA, a quality bar.
Catalogue operations
New listings, attribute completion, category mapping, image and copy compliance, bulk edits, marketplace-specific field rules, listing health repair.
Pricing operations
Competitor tracking, rule-based repricing inside a margin floor, promotion loading, price-parity checks across channels.
Inventory operations
Stock synchronisation across channels, reorder-point flags, dead-stock reporting, oversell prevention, warehouse reconciliation.
Order operations
Order confirmation, dispatch tracking, delay chasing, cancellation handling, marketplace case filing for lost or damaged shipments.
Customer support
Buyer messages across marketplace inboxes, email, chat and the seller's own store. Pre-purchase queries, order status, complaints, escalation triage.
Returns & refunds
Return authorisation, pickup chasing, inspection triage, refund eligibility, claim filing and reconciliation against marketplace settlements.
Full scope, volume bands, SLA matrix and quality thresholds are in the project proposal.
What this process costs on seats.
Run the numbers the way you always do. Standard-band volume on a conventional staffing model needs somewhere around twenty-two productive agents once you cover three shifts, plus supervision and quality.
| Line | Conventional seat model | AI eCommOps |
|---|---|---|
| Productive headcount | 18–22 agents across three shifts | 1 reviewer |
| Supervision & quality | 1 TL, 1 QA, part trainer | Your existing TL, part-time |
| Floor and infrastructure | 22+ seats, systems, power, connectivity | None |
| Ramp to full production | 45–60 days including hiring | 21 days |
| Night and weekend cover | Staffed and paid, or unstaffed and breached | Fleet runs continuously |
| Attrition exposure | Continuous, on your training investment | None on the AI layer |
| Output consistency | Varies by agent, shift and month | Same rules applied every time, sampled weekly |
| Capital at risk to start | Seats, systems, licences, hiring, ramp salaries | ₹3,45,000 one-time |
We are not claiming an AI fleet is a person. It is not. It is faster and more consistent on rule-bound, high-volume work, and it is useless on judgement calls — which is exactly why one experienced reviewer stays in the loop, and why anything touching money or a published claim never leaves human hands.
Ten trained agents at go-live. Twenty-five at ceiling.
The fleet is not one chatbot with ten names. Each agent owns a defined slice of the process, holds its own tool permissions, and carries an autonomy level that decides whether it acts or only proposes.
| Agent | Owns | Autonomy |
|---|---|---|
| Inventory AI | Stock sync, reorder points, dead-stock flags | Supervised |
| Listing AI | Copy, attributes, category mapping, images | Draft only |
| Pricing AI | Competitor scrape, repricing, margin guard | Supervised |
| Support AI | Drafts replies across five channels | Draft only |
| Refund AI | Eligibility checks, claim filing, reconciliation | Draft only |
| Returns AI | Pickup chasing, inspection triage | Supervised |
| Compliance AI | Policy and claim screening before publish | Draft only |
| Ads AI | Campaign pacing, bid and placement moves | Supervised |
| Forecast AI | Demand projection, replenishment planning | Supervised |
| Fraud AI | Order and return abuse detection | Draft only |
What the fleet is never allowed to do
Enforced in the action handler as code, not written as an instruction a model could be talked out of. This is the part every operator asks about at three in the morning, so it is the part we publish.
- Never publish a medical, health or performance claim. Blocked at draft, routed to human review.
- Never price below the client's margin floor. Hard refusal plus an alert, every attempt logged.
- Never issue a refund above ₹5,000 without a named human approval.
- Never send a reply that references legal action. Escalated to the reviewer.
- Never act outside the marketplace accounts assigned to your workspace.
- Every decision is logged with its inputs and retained for 24 months. QC re-checks 200 random decisions each week.
Your supervisor runs it from here.
One console for the whole operation: the fleet and its autonomy states, the task centre, the human review queue, marketplace health, the AI operating balance and the earnings ledger. No engineering skill required — if your team leader can run a CRM, they can run this.
The panel is the reason a fifteen-minute demo closes what a fifty-page document cannot. Watch a fleet work on live volume, open the review queue, and see exactly where a human still decides.
| Agent | Status | Autonomy |
|---|---|---|
| Support AI | Running | Draft only |
| Inventory AI | Running | Supervised |
| Pricing AI | Running | Supervised |
| Refund AI | Running | Draft only |
| Returns AI | Running | Supervised |
Illustrative rendering. No client data shown.
What you keep, not what you bill.
Most proposals quote a payout and leave you to discover the cost base in month one. This is the whole position, with every running cost already deducted.
| Position at standard band | Monthly | Note |
|---|---|---|
| Gross payout to your company | ₹1,30,000 | Volume-linked, against verified delivery |
| Platform & panel licence | − ₹9,500 | Fixed |
| Workspace VPS (8 vCPU / 16 GB / 200 GB) | − ₹5,200 | Fixed |
| AI consumption, prepaid balance | − ₹16,000 | Varies with volume |
| Compliance, audit retention, misc | − ₹3,300 | Fixed |
| Your income, after everything | ₹96,000 | Setup recovered in month 5 |
| Across the 12-month term | Income | Cumulative |
|---|---|---|
| Setup, paid at day zero | — | −₹3,45,000 |
| Months 1–2, ramp band | ₹50,000 / mo | −₹2,45,000 |
| Months 3–5, standard band | ₹96,000 / mo | +₹43,000 |
| Months 6–12, standard band | ₹96,000 / mo | +₹7,15,000 |
| Net across year one | — | approx. ₹7,15,000 |
Capacity expands out of earnings
Five more agents cost ₹1,75,000 and add roughly ₹80,000 of income after their own running cost — a block repays itself in a little over two months. You never buy capacity before the volume exists to fill it. At the twenty-five agent ceiling the workspace earns around ₹3,01,000 a month net.
What we commit to in writing
Billable volume allocated within 45 days of go-live, or the setup fee is credited back pro-rata. That is a service-level commitment on our side of the contract. It is not a guaranteed return, and we will not describe it as one.
This is a fit if
- You run a registered BPO or ITES company with GST and can sign a service agreement.
- You have one experienced team leader you can put on this part-time from day one.
- You want an AI delivery line to show clients without funding the build yourself.
- You are comfortable with volume-linked revenue rather than a fixed per-seat invoice.
- You can carry two months of running cost while allocation ramps.
This is not a fit if
- You are looking for passive income with nobody watching the operation.
- You want a fixed monthly figure regardless of delivered volume.
- You expect to bring your own eCommerce clients — allocation comes from Akontec.
- You want the AI to run entirely unattended, including on refunds and published claims.
- You need the setup returned inside three months.
Fifteen minutes on the live panel will tell you more than this page.
Book a walkthrough with an Akontec delivery manager. We open a working workspace, run the fleet, show you the review queue and answer the awkward questions. No obligation and no sales deck.