Sizing a Citizen Helpline: Seats, Peaks, Abandonment

Executive Summary

A seated helpline is staffed for a number someone forecast, and it is judged on a monthly average that the busiest hour does not respect. This guide is for the nodal officer or state IT department that runs a helpline under a seat-based contract and needs to answer three questions: how many calls can be on the line at once, what happens to the rest, and how that changes when an automated layer answers the first part of every call.

Executive Callout Three separate limits decide what a citizen hears at the peak: the telephone lines the carrier provides, the seats that are staffed, and how long a caller will wait. Each fails differently, and only one of them shows up in the usual report. The worked example below is illustrative and calculated from the Erlang C formula. It is not measured data from any Indian helpline.

Introduction: The Average Hour Does Not Exist

Seats are bought against a forecast of daily volume. Volume then arrives in clumps: after a deadline, an announcement, a result, a season. The monthly report averages the clumps away.

Public figures show the problem. A written reply in the Lok Sabha, reported by Millennium Post, put calls received at the Kisan Call Centres at 47.87 lakh in 2021-22 and 35.22 lakh in 2022-23. That is about 13,100 and 9,650 a day on average. Neither number says anything about the busiest hour of a sowing season, and we found no published hourly or half-hourly figure for any Indian government helpline. The numbers that size a helpline are the ones its contract usually does not ask for.

This guide sets out what to ask for, how to turn it into a seat count, and what changes when part of the call is answered by software. The change is to the helpline's own workflow, covered on the AI call centre automation for government helplines page: seats stop being sized to the peak and start being sized to the calls that need a person.

Three Places a Call Is Lost

Where the call fails	What the caller experiences	How it is modelled	Where it shows up
Telephone lines	Busy tone or a dropped connection	Erlang B (blocked calls)	Often nowhere in the helpline's own system; the carrier's records hold it
Seats	Hold music, a queue	Erlang C	Queue time in the call-distribution report
Patience	The caller hangs up and redials	Abandonment, Erlang A	Abandoned count, but the redial is booked as a new call

The first row is the one departments miss. A caller who meets a busy tone never enters the queue, so a report built from the queue shows a calmer helpline than the one citizens experience. The third row inflates the volume it appears to measure: the WFM Labs reference on Erlang C notes that callers who abandon may retry, which inflates apparent volume, and that the standard model does not represent retrial behaviour.

The Arithmetic of a Seat

Telecom and contact-centre planners size seats with the Erlang C model, named for the Danish engineer A. K. Erlang. It takes three inputs for an interval: the number of calls, the average handle time (AHT), and the number of seats on the phone at the same time.

Offered load in Erlangs is calls per hour multiplied by AHT in hours. 2,400 calls an hour at six minutes each is 240 Erlangs, the amount of work arriving every hour, expressed as seats kept busy.
Chance of waiting C(N, A) comes from the load A and the seat count N. Seats must exceed the load or the queue grows without limit.
Service level, the share of calls answered within t seconds, is 1 − C(N, A) × e^(−(N − A) × t / AHT).

Two consequences matter more than the formula.

First, no seat count can carry more work than it has seats. If 240 Erlangs arrive and 67 seats are staffed, at most 67 of the 240 can be served in that hour. At least 72% of the work is lost to hold, abandonment or a busy tone before any redials are counted.

Second, the model assumes every caller waits as long as it takes. The WFM Labs page says this makes Erlang C overstate staffing, which is safe for planning, and that at very high occupancy it overstates waits. For a helpline with real abandonment, the same page points to the Erlang A extension. Treat Erlang C as a planning ceiling and confirm it against a live pilot.

A Worked Example (Illustrative)

Take a state helpline with a normal hour of 600 calls and a busiest hour four times that, 2,400. Four times is an assumption for the example, not a finding. Take AHT at six minutes. For the service target, borrow a reference level from outside government: TRAI's quality-of-service benchmark for telecom customer care is that at least 95% of calls are answered by an operator within 90 seconds, as listed in its December 2025 performance indicators report. That is a telecom standard, not a government-helpline one. We found no published equivalent for government helplines to use instead.

Hour	Calls	Load (Erlangs)	Seats on the phone for 95% in 90 s	Occupancy
Normal	600	60	67	90%
Busiest	2,400	240	249	96%

Staffing to the normal hour leaves the busiest hour at 67 seats against 240 Erlangs. Staffing to the busiest hour means 182 seats sit idle in a normal hour. Both are seats on the phone at the same moment, not headcount: breaks, leave and absence come on top, and a 96% occupancy is not one that people sustain. The table shows why a flat roster cannot serve a spiky line, which is a structural problem and not a failure of the people working it.

What Changes With an Automated Answering Layer

The layer answers first. It resolves the call reasons that have a determinate answer, such as a status read-back or a document list, and it transfers the rest to a seat with the conversation so far attached. Seats then see only the calls that need a person.

The table repeats the busiest hour from above. The containment share is an assumption chosen to show the sensitivity, not a result from any deployment. Because the easy calls leave first, the calls that remain are longer, so human AHT is set at 7.5 minutes for every row after the first. That is also an assumption.

Share contained by the layer	Calls reaching seats per hour	Human AHT	Load (Erlangs)	Seats for 95% in 90 s	Occupancy
0%	2,400	6.0 min	240	249	96%
40%	1,440	7.5 min	180	190	95%
60%	960	7.5 min	120	130	92%
80%	480	7.5 min	60	68	88%

At 40% containment, seats fall by about a quarter, from 249 to 190, because the remaining calls are longer. At 60% they fall by roughly half, and at 80% by about three quarters. The gain is steeply non-linear, which is why a containment target set before anyone has read the call log is a poor basis for a contract. Pick the call reasons first, from the last 90 days of dispositions, and let the percentage follow.

Two limits of the layer itself belong in the same sizing exercise.

It has a concurrency ceiling. The number of simultaneous automated calls is set by the telephony channels purchased and the limits of the speech and language services behind them. It can be bought up, but it is a number, and the vendor should state it in writing.
It does not fix the line limit. If the carrier provides 100 lines and 300 callers dial at once, 200 meet a busy tone whatever answers behind the lines. Size the lines first.

The human-in-the-loop guide covers what a good transfer looks like. A transfer that makes the caller start again adds handle time to the seats you were trying to relieve.

What to Pull From the Last 90 Days

Ask the operator and the carrier for these. None needs new tooling.

Half-hour intervals, not days. For each interval: calls offered, answered and abandoned, with abandoned calls split by seconds waited.
AHT by call reason. A single aggregate hides the mix shift that automation causes. The KPI guide lists the measures worth tracking beside it.
Carrier records of attempts against connected calls on the helpline number, which is the only place busy-tone blocking appears.
Repeat callers. The share of numbers calling again within a day, using hashed numbers, since caller numbers are personal data.
A calendar of what caused each peak: deadline, announcement, result, season. A peak that can be dated can be forecast.

If the operator cannot supply item 1 or item 3, that is the first finding of the project.

What Changes in the Helpline's Workflow

	Today	After
How seats are set	Contract fixes a seat count from a daily forecast	Seats set from escalated calls in the busiest half-hour; lines and automated concurrency set from the full peak
What the monthly report shows	Calls received, answered, average handle time	Service level and abandonment by half-hour, line-blocked attempts, AHT by call reason
What happens at the peak	Hold, abandonment, redials, busy tone	The layer takes the call; a seat joins only when needed
What the review meeting asks	Was the monthly average met?	Was the busiest half-hour met, and what was the automated layer's peak concurrency?

The buyer is the department or the state IT unit that owns the helpline contract. The workflow it runs today is a seat count fixed from a forecast. What changes is the unit of measurement, from the month to the half-hour, and the thing being bought, from seats to answered calls.

Writing the Peak Into the Contract

Measure service level at the busiest half-hour of each day, not the monthly average, and publish it.
Report abandonment with seconds waited, and require the carrier's attempted-versus-connected figures alongside it.
Require a stated concurrent-call ceiling for the automated layer, and test it at the forecast peak and above it before go-live.
Count transferred calls and their AHT separately from fully handled ones, so the seat count follows what seats actually do.
Keep a fallback route to the existing menu and queue, so an outage of the layer sends callers to what they use today instead of to silence.

Risks and Mitigation

Risk	Mitigation
Containment assumed instead of measured	Start on overflow only: the layer takes calls that would otherwise queue or meet a busy tone. Measure for 30 days, then widen.
Lines undersized, so the layer never sees the peak	Get the carrier's blocking data first and size lines to the forecast peak
Transfer surge when the layer fails	Fallback route to the existing queue, tested before launch
Erlang C used as a forecast	Use it as a planning ceiling, compare with pilot data, and re-run after each change in call mix
Peaks not dated, so the forecast is a guess	Keep the event calendar and update it each season

Key Takeaways

A helpline loses calls in three places: lines, seats and patience. The usual report shows only the second and part of the third.
No roster can serve a load larger than its seat count. Sizing to the median hour guarantees loss at the peak; sizing to the peak buys idle seats.
Automation changes the seat count by the share of calls it fully handles, and the remaining calls are longer, so the gain is not proportional.
Ask for half-hour data, AHT by call reason and the carrier's blocking records before any design.
Write the busiest half-hour, the automated layer's concurrency ceiling and a fallback route into the contract.

Conclusion

A helpline cell does not need a better forecast to start. It needs the 90-day pull, a calendar of its own peaks, and a service level measured where the citizens feel it. With those, the seat count, the line count and the automated layer's concurrency can each be set from a number the department owns.

Government leaders exploring AI-powered citizen engagement can begin with a focused pilot in one department or constituency to validate impact before scaling statewide. Aisewak helps public institutions deploy multilingual Voice AI solutions designed specifically for Indian governance. For a layer that sits in front of seats you already run, start with the AI call centre automation for government page. For the comparison with menu-based systems, see AI voice agent vs. IVR.

FAQ

What is concurrency in a government helpline?

It is the number of calls in progress at the same moment. Seats set the human ceiling, telephone lines set the carrier ceiling, and an automated layer has its own ceiling set by purchased channels and service limits. The busiest of these moments, not the daily average, decides what citizens experience.

How many seats does a citizen helpline need?

It depends on calls in the busiest interval, average handle time and the service level you commit to. In the worked example, 2,400 calls an hour at six minutes each needs about 249 seats on the phone for 95% answered within 90 seconds. Your own half-hour data replaces the example's assumptions.

What is Erlang C?

It is a queueing model that gives the chance a caller must wait, given the call load and the number of seats. Contact-centre planners use it to find the seats needed for a service level. It assumes callers wait as long as it takes, so it overstates staffing for lines with real abandonment.

How do we measure abandonment honestly?

Count abandoned calls with the seconds waited, and add the carrier's attempted-versus-connected figures so busy-tone calls are visible. Report repeat callers separately, since a redial is the same citizen counted twice.

Does an AI voice layer remove the need for seats?

No. It reduces the calls that reach seats by the share it fully handles, and the remaining calls are usually longer. In the illustration, 60% containment roughly halves the seats needed. Calls with discretion, distress or disputed records still need an officer.

What service level should we set?

The department decides, per call type. TRAI's benchmark for telecom customer care, at least 95% of calls answered within 90 seconds, is one outside reference. We found no published standard specific to government helplines, so write the target and its measurement interval into the contract.

Can the automated layer take unlimited calls at once?

No. Its limit is the telephony channels bought and the capacity of the services behind it. Ask the vendor for the ceiling in writing and test it at and above the forecast peak.

What data do we need before we start?

Half-hour call data for the last 90 days, AHT by call reason, abandoned calls split by seconds waited, the carrier's attempted-versus-connected records, and a dated list of what caused each peak. Caller numbers are personal data, so hash them before analysis.

Schema Markup Suggestions

Article: headline, description, datePublished and dateModified of 2026-10-21, publisher AiSewak.
FAQPage: the eight Q&A pairs above.
Table: the worked example tables, with the assumptions stated in the surrounding text.

Suggested External References

WFM Labs, Erlang C — wiki.wfmlabs.org/wiki/Erlang_C, formulas and assumptions, read 21 October 2026
TRAI, Quarterly Performance Indicators Report — December 2025 release, quality-of-service benchmark for calls answered within 90 seconds
Millennium Post, Centre to revamp Kisan Call Centers — millenniumpost.in, 2 August 2023, reporting a written Lok Sabha reply

A helpline staffed for the normal hour loses most of the busiest one. A guide for helpline cells: the three places a call is lost, a worked Erlang C example, and what changes in the seat count when software answers first.

LinkedIn Executive Summary

A seated helpline is bought against a daily forecast and judged on a monthly average. Neither describes the hour in which citizens fail to get through.

A worked example makes the gap plain. With 600 calls in a normal hour and 2,400 in the busiest, a six-minute average handle time and a 95%-within-90-seconds target, the normal hour needs 67 seats on the phone and the busiest needs 249. A roster set for the first can carry at most 67 of the 240 Erlangs of work that arrive in the second.

Calls are also lost in two places the usual report misses: at the telephone lines, as busy tone, and in redials, which are booked as new calls. The first check for any helpline cell is whether the operator and the carrier can produce half-hour data.

An automated answering layer changes the seat count by the share of calls it fully handles, and the calls left for seats are longer. In the illustration, 40% containment cuts seats by about a quarter and 60% by about half. These are assumptions, not results from any deployment.

The contract should measure the busiest half-hour, state the layer's concurrency ceiling and keep a fallback route.

AI Search Optimization Summary

Entities: Erlang C, Erlang B, Erlang A, A. K. Erlang, TRAI, Kisan Call Centre 1551, Lok Sabha, WFM Labs, Aisewak

Topics: government helpline capacity planning, concurrent calls, peak-hour staffing, call abandonment, line blocking, Erlang C worked example, automated answering layer and seat count, helpline contract service levels

Semantic keywords: offered load in Erlangs, seats on the phone, service level 95% in 90 seconds, busiest half-hour, abandonment with seconds waited, attempted versus connected calls, containment and handle time, fallback route to existing queue