All insights

State Modernisation

Citizen Charter Timelines and Agent SLAs

· 10 minute read

If the charter says thirty days for a certificate, an agent that answers in four seconds has not delivered the certificate. Map SLAs to the published service, then to the machine.

The vendor SLA arrived in a polished appendix: 99.5 percent chatbot uptime, four-second median token latency, twenty-four-by-seven multilingual NLU. The department's citizen charter, printed behind the enquiry counter, still said thirty working days for a caste certificate and fifteen for a correction in the ration record. Nobody put the two papers on the same table.

Citizen charters exist in many departments. They are uneven. Some are current. Some are framed and forgotten. Some sit only on a district site. They are still the public promise. An agent SLA that ignores them is a private promise to a purchase committee.

This guide is how to map those two clocks. It is not a claim that every Indian department has a perfect charter. It is not Sevottam certification advice. It is the working rule we use at Prcept AI: the citizen is owed the service in the charter; the machine is owed only the duties that make that service reconstructable.

Two clocks, only one of which the citizen can quote

Clock A is the charter or the statutory timeline: days to issue, days to first response, days to appeal. Clock B is the platform: uptime, retrieval latency, queue depth, human-review turnaround.

Clock B can be necessary. A down model on the last day of a scholarship window is an operational failure. Clock B cannot replace clock A. A fluent refusal in four seconds is still a refusal. A draft that sits with an officer for forty days still misses a thirty-day charter.

Write both clocks in the use-case paper. If clock A does not exist, say so, and do not invent one in the vendor contract that the public cannot enforce.

What to pull off the wall before you write an SLA

  • The latest citizen charter or service-standard table for that service, with the URL and the date you downloaded it.
  • Any statutory clock — appeal periods, RTI, guarantee-of-service laws in some states — checked with counsel, not guessed from memory.
  • The CPGRAMS or state grievance category that already exists for this service, if any.
  • Who is the competent officer for the speaking order, and their usual inbox latency in ordinary weeks, not in a demo week.
  • Language and channel the charter already promises: counter, portal, CSCs. An English-only agent is not a fulfilment of a bilingual charter.

Map the agent to the charter; do not replace the charter

For each service step, write whether the agent drafts, retrieves, routes, or is forbidden. Then write the clock that still binds the officer.

Example pattern, not a universal rule: intake completeness checks can be machine-timed in minutes; statutory scrutiny stays on the charter clock; the speaking order stays human; the SMS of the order is a channel, not a decision.

If the vendor wants a penalty only on uptime, add a penalty or at least a review trigger on human-queue age for agent-drafted files. Otherwise you will buy a chatty waiting room.

Illustrative mapping — replace with the department's live charter rows.
Charter promiseAgent mayAgent must notSLA that matters
First response in N daysDraft an acknowledgement with file numberSend a final rejectionDraft ready for officer before day N-2
Issue certificate in N daysRetrieve checklist and flag missing pagesInvent a missing document as presentOfficer queue age, not token speed
Appeal in N daysShow the original packet and policy versionRe-decide the appealPacket export time under thirty minutes
Counter service in local languageDraft in the charter languagesForce English for accuracyLanguage coverage tested on real forms

Grievance is not uptime

A citizen who is angry about a late certificate will not open a status page that says the model is healthy. They will use the counter, a helpline, CPGRAMS, or a political meeting. Design the agent so the grievance officer can see the same packet the original officer saw.

Do not let the vendor define resolved as the bot sent a link. Resolution is whatever the charter or the grievance manual already says. If you need a new definition, amend the charter in public, do not hide it in an AMS annex.

District variation is not a reason to hide clocks. One tehsil that is informally fast does not excuse another that is quietly slow. Log both against the same charter row. The agent should surface age, not hide it behind a cheerful greeting.

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.

Our charter is outdated, ignore it

Then update the charter through the process your department already uses. Do not use decay as a licence to invent a private SLA. Until it is updated, it is still what the board on the wall says.

AI will make us faster than the charter

Good. Write the stretch as an internal target. Do not publish a four-hour promise you cannot keep in a flood week. Faster than the charter is a gift. Slower than the charter is a grievance.

Uptime is all finance will pay penalties on

Then at least report charter clocks in the monthly review even if they are not liquidated damages. What you do not report you will not staff.

Districts have different informal practices

The charter is the public floor. Informal speed in one tehsil is not a reason to hide delay in another. Log both.

Four weeks to put both clocks on one page

  1. Week 1: download every charter and service-standard PDF the department still publishes. Date the download. List services you actually intend to touch.
  2. Week 2: for each service, write the four columns: charter promise, agent may, agent must not, SLA that matters. Strike any vendor SLA that has no citizen meaning.
  3. Week 3: measure current officer-queue age on paper files or the existing MIS for twenty closed cases. That number, not a demo latency, is your baseline.
  4. Week 4: put charter clocks into the acceptance tests and the monthly review template. Add a line that CPGRAMS or the state portal remains the grievance path.

How this shows up in the file

Subject: Alignment of proposed agent service levels with the departmental citizen charter.

Agent uptime and latency are operational controls. They do not replace published service timelines. Each proposed workflow shall map to a named charter row or shall state that no charter row exists. Speaking orders remain with the competent officer. Grievance remains on the existing public path.

A vendor SLA that is silent on officer-queue age will be treated as incomplete. This note is an internal aid. It is not legal advice.

Election, flood and transfer weeks are part of the SLA

A charter that assumes an officer is at the desk every working day is already optimistic. Add a line in the operational SLA for what happens when the competent officer is on election duty, when the SDC floods, or when transfer orders empty the section. The honest design is a halt state that still issues acknowledgements, plus a queue that ages in public to the collector, not a model that starts deciding because humans are away.

Language coverage is a charter fact. If the wall promises service in a scheduled language, an English-only draft is a breach even if it is fast. Test on real forms from that district, not on staff English. A translation that invents a legal phrase is worse than a slower human draft.

Prcept AI will put charter rows into acceptance. We will not sell you a 99.9 that ignores the board on the wall. If you want a stretch target, we will put it in the monthly review, not on the citizen SMS.

  • Print the mapped four-column table and clip it to the purchase file.
  • Name the officer whose queue age will be reported. If you cannot name them, you do not have a gate.
  • Keep CPGRAMS or the state portal as the grievance path even if the agent can draft a reply.
  • Disable any auto-send of refusals before the charter clock is mapped.

What the next file must contain

“Citizen Charter Timelines and Agent SLAs” earns a line in the noting only if a P1 CIO/CTO can attach proof of “citizen charter service timeline.” 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 “citizen charter service timeline.”
  • 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.

This article is informational field guidance for Indian public institutions, not legal, procurement, security-accreditation, engineering or board advice. Confirm against the live Gazette, GFR, GeM term, CVC instruction, CERT-In direction, UIDAI regulation, DPE guideline, DPDP text, departmental charter, plant safety manual and your counsel before you file it.

Questions this usually raises

Does every department have a binding citizen charter?
Many departments publish charters or service standards; quality and currency vary. Some services also have statutory clocks. Do not assume a uniform national SLA. Pull the paper that applies to the service you are automating.
Can an agent SLA legally override the charter?
Treat the charter and any statutory timeline as the public promise unless counsel tells you otherwise after reading the actual instrument. A private AMS annex is not how you amend a public standard.
Is CPGRAMS the same as a charter?
No. CPGRAMS and state portals are grievance paths. Charters are service standards. Design for both. Do not call a chatbot reply a closed grievance.
What if the charter is silent on digital channels?
Silence is not a licence to invent a harsher or a looser public clock. Keep the existing timeline and add channel-specific operational SLAs underneath it.
Should we publish the vendor's four-second latency to citizens?
No. Publish what the charter already promised. Keep token latency as an internal operational control. Citizens cannot enforce a GPU metric.

Sources