Nobody decides that support has become a problem. The person who answers everything goes on vacation and the queue backs up for a week. A customer with a contract asks who handles their account and nobody can say. Your best engineer loses a Tuesday to tickets. Somewhere between 100 and 500 employees, support stops being one person’s job and becomes an operation, usually before anyone has planned one.
Here is how that stretch usually goes, stage by stage, including when you should not outsource yet.
The headcounts below are our judgment, not a measurement. We know of no public dataset that cuts support operations at these lines. Support load follows your customers and your product more than your payroll, so a 120-person software company with enterprise contracts can be further along than a 450-person retailer with simple orders. Trust the signs over the numbers.
Under about 100 employees, keep support close and start writing things down. Past about 100, your first hire is holding four jobs. Past about 200, a shared pool strains once tickets need real product knowledge. Toward 500, informal arrangements have to become written ones. Read your escalation rate next to first contact resolution, never alone.
Size explains less than you would expect
The largest public dataset we know of is Freshworks’ Customer Service Benchmark Report 2025: 32,000+ companies and 1.2 billion tickets from 2024. It is one vendor’s customer base, so read it as that. Its size bands split at 250 employees, which puts a 180-person company and a 400-person company in different rows. We have not blended them.
| Tickets | Top tier | Median | Bottom tier |
|---|---|---|---|
| First response, 0 to 250 employees | 3m 32s | 1h 3m | 7h 21m |
| First response, 250 to 5,000 | 2m 31s | 53m 16s | 6h 27m |
| First contact resolution, 250 to 5,000 | 35% | 15% | 6% |
| CSAT, 250 to 5,000 | 97.50% | 88.00% | 61.10% |
Freshworks, ticketing, January to December 2024. Top tier is its top 20th percentile. The last two rows cover one band, the one we could confirm.
The median first response differs by about ten minutes between the two bands. Inside the larger band it runs from 2m 31s at the top tier to 6h 27m at the bottom. Company size does not explain a spread like that, and the data does not say what does. Our view is that it comes down to how the operation is set up. The stages below are where that tends to fail.
Read channels separately too. In the same band, conversations (chat, messaging, social) post a median first contact resolution of 81.88% and CSAT of 95.15%, against 15% and 88.00% on tickets. One blended number can look healthy while the email queue is not.
Under about 100 employees: support runs on memory
A person or two know the customers, the product and which engineer to ask. It works better than most processes would, and the tickets double as product research. What breaks is that person’s vacation, and everything they never had to write down.
Should you outsource? Usually not. Spend the time on ticket tags, a list of top contact reasons and readable timestamps. You will want all three next.
Past about 100: the first support hire stops scaling
You usually see this in a calendar before you see it in a report. Your first hire is answering customers, training whoever joined last month, taking every hard ticket and being the link to product. That is four jobs. Response times move with one person’s schedule. Engineers get pulled in by direct message. Nobody can picture this person being out for a week.
The natural fix is to hire two more people and make your first hire the lead. That is often right, and it has a cost: the queue loses its best agent, the new lead may never have managed anyone, and new hires learn from a knowledge base nobody wrote.
Hiring also turns into permanent work. The Bureau of Labor Statistics projects customer service representative jobs to decline 5% from 2025 to 2035 while still producing about 289,500 openings a year, all of them to replace people who transfer out or leave the workforce. Plan to replace people as well as add them.
Past about 200: a shared pool stops being credible
In a shared pool, a vendor’s agents split their time across several clients. At low volume with simple, repetitive tickets you avoid paying for an idle seat, and it is a sensible way to cover nights or overflow. A very small dedicated team has its own weakness: with two people, one sick day takes out half your coverage.
It stops being credible once the work needs your product in someone’s head. Four signs:
You keep explaining the same thing to different people. What one agent learned last week did not reach whoever picked up your ticket today.
You cannot name who answered. A customer with a contract asks, or QA finds a bad reply, and there is nobody specific to coach.
The hard tickets come back. Anything unfamiliar gets sent upward, so your team still does the difficult work and now manages a vendor as well.
The same errors repeat. Findings go to a vendor supervisor and the next sample looks the same.
There are two honest ways out: a bigger in-house team, or a dedicated team from outside the company. In-house tends to win when hiring in your market is easy and support sits close to product. A dedicated team tends to win when recruiting is the constraint, or you need hours or languages you cannot hire for. A pool can also keep simple queues like password resets while the hard ones move.
A week has 168 hours and a full-time person works 40, so one seat covered around the clock takes 4.2 people before vacations, sick days or training. That headcount, more than any one salary, is what drives the cost of night coverage.
Before paying for nights, check your own timestamps. In Fixify’s IT help desk data, 82% of tickets arrive during business hours. That is internal IT, so your pattern may differ. You can model the cost against your current setup in the outsourcing savings calculator.
Toward 500: the informal arrangements run out
By now you have managers, more than one tier and some form of QA. What runs out is everything that worked on favors. Escalating to engineering by direct message needs an owner, a written definition of what qualifies and a response time. Product feedback in the hallway stops, because nobody in product reads tickets unless a route exists. One CSAT on the board slide needs splitting by channel and queue. Workforce planning and QA need owners of their own, separate from the lead who hires and coaches.
The AI question gets loud here too. Keep two numbers apart. Freshworks reports AI agent deflection of 55% at companies of 251 to 5,000 employees, measured on its own product. Gartner, surveying 5,728 customers, found 14% of issues fully resolved in self-service. Deflection counts a contact that did not reach an agent. Resolution counts a problem solved.
What the tooling looks like at each stage
| Stage | Tooling | What breaks first |
|---|---|---|
| Under about 100 | Shared inbox or one help desk on default settings | Coverage when one person is away |
| Past about 100 | Tags, routing, SLA clocks, macros, a first knowledge base, reports by channel | The first hire’s four jobs |
| Past about 200 | QA scorecards, a maintained knowledge base, a link to the engineering tracker, real scheduling | Product knowledge |
| Toward 500 | Workforce planning, QA by queue, CRM integration, AI where the knowledge is ready | Routing rules and automations nobody owns |
This table is judgment too. Buy at the stage where you feel the problem. A QA tool does little for a team with no written standard to score against.
What your escalation rate tends to tell you
We found no public escalation benchmark by company size that we trust. The closest yardstick is SQM Group’s North American contact center research, where agents need help from another agent, a supervisor or a help desk on 20% of the calls they handle. That is phone work, so treat it as rough.
SQM’s more useful number is the gap. Agent-assisted calls show 16% lower first call resolution, and SQM says the gap is commonly under 2% at high performers. So track two things: how often tickets escalate, and whether escalated tickets end as well as the others. Then compare yourself with yourself month to month, because contact mix moves everything. SQM’s first call resolution runs 73% on general inquiries, 60% on technical support and 48% on complaints. It comes from post-call surveys, so do not set it against the Freshworks ticket data above.
- Rising in one topic. Usually a product or documentation problem. More agents will not fix it.
- Rising everywhere after a hiring round. Training and the knowledge base have fallen behind.
- Falling while reopens and complaints climb. Escalating is being discouraged, often by a metric.
- High and flat, with tier two doing tier one’s work. Nobody has written down what qualifies. The escalation matrix generator is a quick way to draft that rule.
When you should not outsource yet
Plenty of companies in this band should wait. Hold off if any of these is true:
- You are under about 100 employees and a person or two can carry the volume. Tickets are still your best product research.
- You cannot state your volume by channel and hour, or your CSAT by channel. Any vendor will be measured against a baseline you do not have. Collect a few months first.
- Your top contact reasons are bugs. More agents means paying to answer the same question. Fix the product.
- Your process lives in someone’s head. A handoff exports the confusion. Write down your top contact reasons and how to handle them.
- Your customer contracts limit who can touch their data or require specific certifications. Read them before any proposal. They may narrow your list.
- Cost is your only reason and the team is very small. Managing a vendor takes time too, and it can eat the saving.
The case for moving gets stronger when hiring cannot keep up, when you want 24/7 coverage without staffing three shifts, when quality varies by agent, or when you need languages you cannot hire for. Even then, start with one queue.
If you are past that point
Touch Support builds dedicated, embedded teams. Our agents work inside your Zendesk, Intercom or Salesforce, using your macros, playbooks and standards, and most clients go live within a week. Many pilot with 1 to 3 agents and grow from there. A $300M hosting company started with recruiting assistance and a handful of agents, which grew into a fully integrated global operation. Our published CX figures are 92.7% average CSAT across channels and 90% resolved on first contact. Those are our own numbers, so do not compare them directly with the benchmarks above.
“One of our consistent pieces of negative feedback was, ‘you have a really knowledgeable team, but you guys are so slow’… We don’t have that anymore. The tickets get picked up really fast.”
Greg Harvey, Managing Director, Code Enigma
Not sure which model fits? The Support Model Selector takes about ninety seconds and needs no form. The free CX audit is 15 questions on your team, operations, tools, channels and metrics. You get a score, and it asks for your contact details before the full improvement plan.
Stage boundaries and the signs at each stage are our judgment, not measured findings. The 4.2 people is our arithmetic. Sources: Freshworks, SQM metrics, SQM FCR, BLS, Fixify (50,000+ tickets, 30+ companies), Gartner. More benchmarks with sample sizes are in our State of Customer Support Operations 2026 report.
