Per-Seat vs Per-Agent vs Per-Outcome Pricing

Pricing is a unit of account dressed as a product. A vendor who sells per seat wants a named officer in a drop-down. A vendor who sells per agent wants a multiplicity of little runtimes. A vendor who sells per outcome wants to attach an invoice to a public duty. All three can be honest. All three can be a trap. The question is whether the unit can be counted by an accounts officer who will not be employed by the vendor.

Government work makes outcome pricing harder than the SaaS blog assumes. A closed grievance is not always a resolved grievance. A scholarship decision is not a success if it is a wrong rejection. Paying for resolutions is how you buy speed at the expense of law. Paying for seats is how you buy a tool nobody shares. Paying for agents is how you wake up with seventeen containers and one user.

This comparison is for commercial negotiation and for the way you write the GeM or tender unit. It is not a price list. Prcept will sell you a deployable on-prem agent with a support envelope you can audit. We are sceptical of per-outcome meters on rights-affecting work, and we will say so in the room.

The three units, without the poetry

Per seat means a named or concurrent human. It maps to how you already buy office suites. It fails when the agent is a shared clerk on a counter, or when night-shift operators rotate. It also fails when the vendor counts every reader of a dashboard as a seat, including internal audit once a quarter.

Per agent means a running workflow or a deployed runtime. It can be clean if agent is defined as a blessed workflow with a version and a purpose tag. It is unclean if the vendor can mint agents for every prompt variant and invoice the minting.

Per outcome means a rupee attached to a measured event: ticket closed, form extracted, call summarised. It looks aligned. It is only aligned if the event is hard to game, hard to dispute, and not itself a legal decision. Public duties fail those tests more often than sales queues do.

Comparison — choose the unit you can count in the absence of the vendor.

Unit	Fits	Breaks when	Audit test
Per named seat	A small specialist cell with stable officers	Counters, rotas, shared kiosks, viewer inflation	Can we export the seat list and match it to eOffice IDs?
Per concurrent seat	Bursty helpdesks	The concurrency cap is a hidden queue, not a price	Can we see rejected logins and queue time?
Per blessed workflow (agent)	A handful of named productions	Vendor-defined micro-agents; env copies billed as prod	Is an agent a row in our register, or a container we never named?
Per technical unit (GPU-hour, token)	Bursts, IndiaAI, evals	Uncapped production chat; prompt bloat	Can accounts reconstruct the meter?
Per outcome	Narrow extractive tasks with a validator	Rights, money, resolution without a legal meaning	Can we dispute an outcome without the vendor's console?

Hybrids that do not lie

A platform fee for the on-prem runtime, plus a support envelope, plus optional GPU-hours for bursts, is a hybrid an accounts officer can follow. So is a per-workflow licence with a named cap and a written price for the next workflow. So is a per-document extraction price on a validator-backed field job, with a human-review queue that is not billed as a success.

A hybrid that does lie is free seats, paid outcomes where the outcome definition sits in a vendor wiki and can be changed by a release note. Freeze the definition in the contract. If they will not freeze it, you are buying a moving invoice.

GFR and GeM want a unit you can receive. You can receive a licence key, a support year, a GPU-hour, a completed extraction batch. You cannot receive transformation. Do not put transformation in the BoQ.

What we will not meter

Prcept will not attach our invoice to a citizen's grievance being resolved, a student being awarded, or an employee being cleared. Those are official acts. Officials sign them. A vendor who wants a success fee on an official act is asking to be a party to the act. That is the wrong geometry.

We will meter what we can show: a deployed workflow, a support period, a volume of extractive documents with a validator, a reserved burst. If that makes us look old-fashioned next to an outcome-priced competitor, so be it. The PAC is also old-fashioned.

Write the unit so GeM can receive it

A unit that cannot be received cannot be paid. GeM and a conventional tender both want a description an inspecting officer can tick. Licence key delivered, support year begun, extraction batch accepted against a validator, GPU-hours consumed on a bill. If your unit needs a vendor dashboard to exist, you have not written a unit. You have written a hope.

Put the definition in the pre-bid clarification, not only in the negotiation. Bidders who cannot live with a frozen unit will tell you early, which is a gift. Bidders who smile and change the wiki later are why the file exists.

Objections you will hear — and what to do with them

These are the lines that stall the file. Answer them in the room, then put the answer in the note. A spoken answer without paper will be forgotten by the next officer.

Outcome pricing aligns incentives

Only if the outcome is the legal outcome and you can measure it without the vendor. Alignment to a dashboard is not alignment to the Act.

Per-seat is what GeM understands

GeM understands any unit you specify clearly. A bad unit that looks familiar is still a bad unit.

We want the vendor to share risk

Share risk with warranties, SLAs, retentions and milestone holds. Do not share the official act.

Our SI will mint many agents; we should get volume pricing

Volume on a unit you have not defined is a discount on fog. Define the workflow first, then ask for a declining block.

A two-sitting commercial choice

1. Sitting 1: list the humans, the workflows and the events you can already count in your MIS without the vendor.
2. Sitting 2: pick the unit that matches that list. Write the definition in one paragraph. Send it as a pre-bid clarification. Refuse any quote whose unit is not that paragraph.

How this shows up in the file

Subject: Commercial unit for [system].

The unit of account shall be [named seat / concurrent seat / blessed workflow / technical meter / extractive batch]. Definition: [paragraph]. Events that are official acts (sanctions, speaking orders, grievance disposal under the applicable framework) shall not be vendor-invoiced outcomes. Meters must be exportable to a department store.

This note is a procurement aid, not legal advice. Official acts are not vendor outcomes.

This article is informational field guidance for Indian public institutions, not legal, procurement, tax, accounting, tariff or engineering advice. Confirm against the current Gazette, GFR, GeM term, SERC tariff order, IndiaAI portal rule, CAG mandate, DPDP text, departmental finance manual and your counsel before you file it. Figures are methods and order-of-magnitude illustrations, not a dataset of real deployments and not a substitute for a live quote.

How to put this in the finance note

A P2 Procurement searching “AI pricing model comparison” needs a number a CFO can defend, not a GPU brand. “Per-Seat vs Per-Agent vs Per-Outcome Pricing” belongs in a cost model with people, power, idle time, AMC and the cost of a failed pilot.

Seats punish sharing. Agents punish architecture. Outcomes punish any duty you cannot meter without gaming. Pick the unit you can audit. IndiaAI subsidy, if you use it, is a live notice — not a permanent discount. On-prem TCO includes ops headcount. Do not invent Rs/hour. Cite the source of every rupee.

• Separate capex, opex, and one-time cleanup.
• Show utilisation, not just peak GPUs.
• Price the human fallback, not only inference.
• Date every tariff and subsidy assumption.

Close this loop before the next CAB

Put “Per-Seat vs Per-Agent vs Per-Outcome Pricing” on the next change-advisory or bid-opening agenda as a single line item with an owner. If it cannot earn a line item, it will not earn a control. The owner should be a P2 Procurement, not “the vendor.”

Revisit the item when the model, the GeM term, the region, or the SI changes. “AI pricing model comparison” is not a one-time workshop. It is a watch item. Date the last check. Unsigned watch items are souvenirs.

What the next file must contain

“Per-Seat vs Per-Agent vs Per-Outcome Pricing” earns a line in the noting only if a P2 Procurement can attach proof of “AI pricing model comparison.” A heading is not proof. A vendor slide is not proof. A workshop photograph is not proof.

Write three dated sentences: what was decided, who owns it after the next posting order, and when it will be re-checked. If you cannot write the three sentences, you are not ready to buy, to sell, or to go live.

Leave unsourced percentages out of the note. DPDP is not a blanket localisation statute. The November 2025 AI governance text is guidance, not an Act. CERT-In’s 28 April 2022 directions still set specified incident and log clocks. A PAC, when lawful, lives in GFR Rule 166.

• Name the designation that owns “AI pricing model comparison.”
• Attach one artefact a stranger can open next year.
• Record the instrument you are actually using.
• Revisit when the model, the SI, the notice or the posting changes.