All insights

Compute & Cost

Measuring Cost per Resolved Grievance

· 9 minute read

Cost per resolved grievance is a valid unit only if resolved is your statutory or MIS definition, not a vendor close button. Build the fraction from Ledger A and last year's disposal count.

Every helpdesk vendor wants to sell you a falling cost per ticket. The trick is always in the denominator. If the vendor's agent can mark a ticket closed by sending a paragraph, cost per close will plunge and reopenings will climb. If the denominator is a disposal that your grievance rules already recognise — a speaking order, a transfer that the receiving office accepted, a refund that the treasury paid — the fraction is a grown-up metric. This framework insists on the grown-up denominator.

The numerator is not AI spend. It is the full year-two cash of the grievance cell that the agent touches, or a carefully allocated slice, compared to the same cell last year. A metric that only counts the model licence will always make the model look cheap. A metric that dumps the entire secretariat into the numerator will always make the model look pointless. Name the cell.

This is not a national benchmark of rupees per CPGRAMS item. We do not have that census. It is the arithmetic a secretary can renew in year two. Prcept will not invoice you per resolution. We will help you measure one.

Define resolved before you define cost

Use the definition your public grievance framework already uses. For many central submissions that is the CPGRAMS disposal logic. For a municipal cell it is the citizen charter or the state portal's closure code. For a campus it is the ordinance. Do not invent an AI-resolved status. Inventing a status is how you game your own fraction.

Split categories that are not comparable. A where-is-my-file query is not a land-title dispute. Report cost per resolved query and cost per resolved dispute separately, or the average will flatter the bot and hide the mess. Count reopenings. A disposal that returns in 30 days is not a resolution for this metric, even if the portal said it was. Put a reopening rate next to the rupee. Finance can stand a stable rupee. They cannot stand a cheap rupee that creates two files.

The fraction

Numerator, year T: cash of the grievance cell — officers (if you already load them), outsourced operators, postage, SMS, the agent's Ledger A slice (runtime, AMC, power allocation, SI hypercare, eval labour), and the MIS increment that exists only because of the agent. Exclude one-time discovery that will not recur, or show it as a separate first-year figure.

Denominator, year T: disposals that meet the definition, net of reopenings inside the window you chose (30 or 90 days). Source: the MIS export, not the vendor dashboard. Compare T to T−1 on the same definitions. The interesting number is the change, not the level. A cell that was already cheap and got slightly cheaper is a success. A cell that became ₹12 per AI-close from a dashboard is a fiction.

Do not monetise citizen time unless your finance manual already does that for other reforms. If you must show citizen time, show it as days, not as a rupee you invented from an average wage.

Framework sheet — one row per grievance class.
ClassResolved meansNumerator sliceWatch-outs
Information / statusCitizen accepted the factual reply; no reopen in 30 daysAgent + operator time on this classBots love this class; do not let it drown disputes
Service delivery (certificate, refund)The service actually occurred in the system of recordCell cash allocated by volumeThe agent did not resolve a refund the treasury did not pay
Appeal / disputeA competent officer's speaking orderOfficer time + agent draft supportNever an outcome-priced vendor event

What the number is allowed to decide

It may decide a renewal, a shrink, or a kill. It may decide whether the small model is enough. It may not decide a citizen's rights. A cheaper dispute-handling line is not a reason to skip a hearing the ordinance requires.

It may not become the vendor's invoice unit. If you pay per resolution, you have handed the denominator to someone who benefits from a soft definition. Keep the metric. Do not sell it. Publish it internally every month from week four. If it does not move by month nine, you have a year-two story already. Change the workflow or stop. Do not change the definition to rescue the fraction.

Prcept will hook drafts and packets to your MIS disposal codes. We will not give you a parallel AI closed funnel. If our draft did not become your disposal, it did not count. That is the only denominator we will put our name near.

Month nine is the point

If the fraction has not moved by month nine, do not wait for the AMC reminder. Redesign the workflow, shrink the model, or stop. Month nine is also when reopenings tell you whether you bought postage or redress. A falling rupee with rising reopenings is a fail. Write that rule before go-live so nobody invents a new definition in March.

Put the monthly card on the same distribution list as the existing grievance MIS report. If the secretary already reads days-to-disposal, the AI line should sit underneath it, not in a vendor PDF she has never opened.

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.

Without a rupee-per-ticket we cannot prove ROI

You can prove a change in a metric you already report, and a cash line you already sanction. That is ROI a PAC understands. A vendor rupee-per-close is theatre.

We should pay the vendor less if cost per resolution falls

That is outcome pricing on an official act. Share risk with SLAs, not with a success fee on disposal.

Reopenings are the citizen's fault

Some are. Measure them anyway. If they spike after the agent, the agent is part of the story.

Our MIS cannot export last year

Then you are not ready for this metric. Fix the export. Do not let the vendor become your statistician of record.

A thirty-day stand-up of the fraction

  1. Days 1–7: freeze the definition of resolved per class. Pull T−1 counts and reopenings.
  2. Days 8–14: draw the numerator from Ledger A and the cell's existing budget. Write allocations.
  3. Days 15–21: wire the agent to the MIS disposal code. Turn off any vendor close that does not write there.
  4. Days 22–30: publish the first internal card. Set the month-nine review that will feed the renewal note.

How this shows up in the file

Subject: Cost per resolved grievance — definition and first baseline.

Resolved, for this metric, follows [charter / portal / ordinance] codes [list], net of reopenings within [30/90] days. Classes are reported separately. Numerator is [cell cash + agent Ledger A slice], excluding [one-time items]. Source of counts: [MIS export], not a vendor dashboard. Vendor invoices will not use this unit. Month-nine review will recommend renew / shrink / stop.

This framework is not a national benchmark and not legal advice. Official disposal is your act, not a vendor close button.

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 P1 CIO/CTO searching “cost per grievance resolution” needs a number a CFO can defend, not a GPU brand. “Measuring Cost per Resolved Grievance” belongs in a cost model with people, power, idle time, AMC and the cost of a failed pilot.

Cost per resolved grievance is a valid unit only if resolved is your statutory or MIS definition, not a vendor close button. Build the fraction from Ledger A and last year's disposal count. 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 “Measuring Cost per Resolved Grievance” 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 P1 CIO/CTO, not “the vendor.”

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

Questions this usually raises

What is a good cost per resolved grievance?
The one that falls or holds while reopenings do not rise, on your definition. There is no honest national rupee we will quote.
Should we include officer salaries?
If your finance already loads them for other reforms, yes, consistently across T and T−1. If not, count cash you actually debit and show officer days as a second line.
Can CPGRAMS data give us the denominator?
If that is the system of record for those submissions, yes. Do not mix CPGRAMS disposals with a vendor close count.
Does a cheaper fraction justify a smaller model?
If the legally material fields still hold, yes. The fraction is a renewal tool, not a reason to drop Q8 on amounts.
Will Prcept price per resolved ticket?
No. Official disposal is your act. We will help you measure it. We will not meter it as our success fee.
Is this a national benchmark of rupees per ticket?
No. We do not have that census. It is arithmetic a secretary can renew in year two.

Sources