Erlang C for Customer Support: When You Need It, When It Fails, and What Small Teams Should Do Instead

Quick Summary: Most first-time support leads who research staffing math try plugging their Intercom or Zendesk chat volumes into an Erlang C spreadsheet macro, only to get absurd headcount numbers. Erlang C was engineered for 1917 telephone switchboards where one human spoke to one caller at a time. It breaks down completely for multi-chat concurrency, customer idle lag, and small teams under 20 agents where minimum floor coverage matters far more than Poisson arrival formulas. This guide explains where Erlang C works, why it misleads small teams, and how to use a practical floor-and-workload model instead.
The Over-Engineering Trap: When a 10-Person Team Tries Call Center Math
Every support lead who starts looking for staffing answers eventually lands on the same acronym: Erlang C.
You read an operations article or an enterprise call center textbook. You download an Excel macro packed with factorial formulas. You paste in your hourly ticket arrivals and Average Handle Time (AHT). And then you stare at the results in confusion:
- It tells you that to maintain an 80% SLA on live chat at 2:00 PM, you need 14.2 agents online.
- Meanwhile, you look at your queue right now, see 4 agents handling the conversations with zero backlog, and wonder what universe this formula came from.
If you manage a small or mid-sized support team (5 to 25 people), here is the blunt truth: You almost certainly do not need Erlang C.
In fact, applying traditional Erlang C to live chat or small remote teams doesn’t just give you inaccurate numbers—it actively misleads you into overstaffing or believing your team is underperforming.
Here is why 1917 telephony math fails for modern channels, why small teams break the formula's assumptions, and what practical framework you should use instead.
1. What Erlang C Was Actually Built For
To understand why Erlang C misbehaves on modern support queues, you have to look at what Danish mathematician A.K. Erlang was studying in 1917: telephone switchboard lines.
The Erlang C formula calculates the probability that an incoming customer will have to wait in a queue, given:
- Arrival rate (λ): Number of contacts arriving per interval (usually 30 or 60 minutes)
- Average Handle Time (Ts): Call duration plus wrap-up time
- Number of available agents (m)
For traditional telephone lines, Erlang C works brilliantly because call queues satisfy three rigid conditions:
- Strict 1:1 concurrency: One phone agent can speak with exactly one human being at a time. An agent on a call cannot pick up another handset.
- Synchronous, continuous engagement: From the moment the call connects until the receiver hangs up, both parties are speaking. There are no 4-minute pauses while someone checks their email.
- Infinite queue tolerance (assumed): The basic Erlang C model assumes that customers will wait indefinitely until an agent answers (it ignores abandonment).
If you run a 150-seat voice call center taking uninterrupted phone calls, Erlang C remains the industry benchmark. But modern customer support looks nothing like a 1917 switchboard.
2. Why Erlang C Breaks Down on Live Chat
When team leads plug live chat metrics into an Erlang C calculator, the math falls apart for three structural reasons:
The Concurrency Lie (1:1 vs. 1:3)
Erlang C assumes a "server" is binary: either 100% busy or 100% idle.
On live chat, support specialists rarely take one conversation at a time. Depending on product complexity, agents handle 2 to 3 concurrent chats simultaneously.
If you feed your raw chat volume into an Erlang C calculator without adjusting for concurrency, the formula assumes every simultaneous incoming chat requires its own dedicated human. It instantly inflates your headcount requirement by 200% to 300%.
The Customer "Ghosting" Gap (Active Touch vs. Elapsed Time)
On a phone call, a 12-minute call represents 12 minutes of active conversation.
On live chat, an 18-minute session typically consists of:
- 4 minutes of the agent typing answers and troubleshooting.
- 14 minutes of waiting for the customer to reply, find their order number, or click a verification link.
If you define Average Handle Time (AHT) as total session duration (18 minutes), Erlang C treats the agent as fully occupied for all 18 minutes. In reality, the agent is actively working on other chats or handling an email ticket during those waiting intervals.
Queue Abandonment & Channel Switching
Erlang C assumes callers sit on hold until someone picks up.
In chat, customers behave very differently:
- If wait time exceeds 90 seconds, customers frequently close the browser tab, send an email, or submit a duplicate ticket.
- Erlang C cannot factor in customer patience without complex extensions (like Erlang A), which add another layer of guesswork to your spreadsheet.
3. The Small Team Reality: Why Erlang C Fails Under 20 Agents
Even if you adjust for chat concurrency, Erlang C has a second, fatal limitation: it relies on the Law of Large Numbers.
Queue theory works when you have large, pooled volumes of requests and dozens of interchangeable agents. When you apply it to a small team (under 15–20 people), the statistical assumptions collapse.
| Factor | Large Enterprise (100+ Agents) | Small Team (5–15 Agents) |
|---|---|---|
| Statistical Smoothing | Fluctuations even out across 40+ people on shift. | One agent going on a bio break reduces queue capacity by 25% to 50%. |
| Rounding Impact | A calculation of 42.3 agents rounds to 43 (+1.6% variance). | A calculation of 1.4 agents rounds to 2 (+42.8% variance). |
| Primary Driver | Driven by statistical arrival distributions. | Driven by minimum floor coverage across timezones. |
| Agent Flexibility | Agents are dedicated to single queues. | Agents multi-task across chat, email, and internal Slack pings. |
The "Fractional Agent" Problem
If Erlang C tells a 100-person contact center they need 84.4 agents at 10 AM, rounding up to 85 is an unnoticeable budget difference.
If Erlang C tells a small startup team that they need 1.3 agents on shift during the European afternoon:
- If you schedule 1 agent, your SLA collapses the moment two chats arrive at once.
- If you schedule 2 agents, you are paying for 50% idle capacity.
In small teams, operational reality always trumps arrival math.
4. What Small Teams Should Use Instead: The Floor-and-Workload Model
If you shouldn't use Erlang C for your small support team, how do you determine how many people to schedule?
For teams of 5 to 25 people handling chat and tickets, a Workload + Floor Staffing Model is faster, easier to maintain, and far more accurate.
Step 1: Calculate Minimum Floor Coverage First
Before looking at volume, calculate your non-negotiable operational floor.
- To provide 24/7 coverage with at least 1 agent online at all times across an 8-hour shift structure, you mathematically require 4.2 to 5 full-time agents just to keep the lights on—regardless of whether you receive 5 chats or 50 chats a day.
- If your baseline requirement is "at least 1 person covering Chat and 1 person covering Escalated Email," your floor is 2 agents per shift. Your floor is your true baseline, not Erlang C.
Step 2: Use Effective Concurrency for Busy Intervals
For peak hours where volume exceeds your floor, calculate required heads using concurrent workload rather than queue probability:
Workload (Hours) = (Hourly Chat Volume × Active Agent Touch Time in minutes) ÷ 60 minutes
Then divide by target occupancy (typically 75–80% for sustainable pacing):
Required Online Heads = Workload (Hours) ÷ (Target Occupancy × Concurrent Chat Capacity)
Example Calculation:
- You receive 30 chats between 14:00 and 15:00 UTC.
- Agents spend 6 minutes of active touch time per chat (ignoring customer wait lag).
- Total active workload = (30 × 6) ÷ 60 = 3.0 hours of active work.
- If agents sustainably handle 2 concurrent chats at 75% target utilization (2 × 0.75 = 1.5 effective capacity):
- Required agents = 3.0 ÷ 1.5 = 2 agents.
You don't need complex Poisson factorials to see that 2 focused agents can handle that hour cleanly.
Step 3: Add Realistic Shrinkage
Never schedule people for 100% of their rostered shift. Factor in breaks, handovers, 1:1s, team meetings, and sudden sick leave:
- Add 20% to 25% shrinkage to your gross headcount calculations.
- If your workload demands 2 people online at all times, you need roughly 2.5 to 2.7 people rostered across that window to ensure breaks don't drop you below your minimum floor.
5. Erlang C vs. Workforce Management: Formulas Don't Work Shifts
Here is the biggest distinction most team leads miss when comparing mathematical models to workforce management:
Erlang C tells you a number. Workforce management tells you who is working, when they sleep, and whether their schedule is fair.
Even if an Erlang C calculator produced the exact right number (e.g., "You need 3 agents at 14:00 UTC"), it does not solve any of your actual day-to-day roster problems:
- Timezone Distribution: Which of your agents in Manila, Lisbon, or Austin is awake and qualified to take that shift?
- Channel Burnout: If an agent spends 8 consecutive hours on high-velocity live chat, their response quality tanks by hour 5. Who rotates onto asynchronous email tickets to give them cognitive rest?
- Leave & Roster Disconnect: When someone takes an approved HR vacation day, does your scheduling sheet automatically flag the coverage gap, or do you only notice when the chat queue turns red on Monday morning?
- Weekend & Shift Equity: Are the same two junior agents getting stuck with Sunday night shifts while senior staff take weekday mornings?
Erlang C is a capacity estimation tool. It is not a scheduling tool.
6. The Support Lead’s Sanity Checklist
Before you spend another afternoon trying to debug an Erlang C spreadsheet macro, run through this 4-step sanity check:
- Are you purely phone-based? If you are running an inbound phone queue with more than 30 agents, keep Erlang C. If you are running live chat, email, or a small remote team, park the formula.
- Have you established your floor coverage? Identify the absolute minimum number of agents needed to maintain your 24/7 or operating-hours presence before doing any volume math.
- Are you measuring active touch time? Base your chat capacity on active typing/troubleshooting minutes, not total session duration inflated by customer silence.
- Are you solving the roster, or just the math? Shift your energy from calculating decimal headcount to balancing task intensity, protecting sleep buffers, and automating schedule generation.
How OneRosta Eliminates Staffing Guesswork
Calculating Erlang formulas and shrinkage percentages by hand in spreadsheets is slow and fragile. If someone edits a cell formula or timezone column, the entire week’s coverage collapses.
OneRosta simplifies staffing management for remote teams:
- Set Channel Thresholds per Hour: Define minimum required agents for Live Chat, Phone, and Email independently across each hour of the week.
- Visual Hourly Deficit Strips: If an approved leave or sudden absence drops your staffing below minimums, OneRosta immediately flags the deficit in red (
-1 Chat). - Auto-Balanced Shift Distribution: Click one button to populate schedules while respecting agent rest buffers, timezone preferences, and channel qualifications.
- Burnout Protection: Automatically balances high-intensity chat shifts with lower-stress email and offline project slots.
Stop relying on telephone-era formulas for modern support teams. When you budget for real arrival patterns and build shrinkage into your baseline, your team stays rested, customer response times stay fast, and your 24/7 coverage runs smoothly.
Want to see how your team's coverage looks in a live 24/7 scheduler?
Explore OneRosta's free interactive sandbox demo (no sign-up required) →
Staffing & shrinkage calculator
Seat-hours per week
336
Agents needed (minimum)
12
A planning estimate only — it does not model queue arrival patterns.
See it on a real rota
Try the interactive sandbox — no account or sign-up needed.