How to build a contact center from scratch

Companies usually build a contact center from scratch when sales reps can no longer keep up with calls and messages, or when they open a new sales channel. Mistakes at the start cost money: a missed call is a lost sale, an extra agent is wasted payroll, a rebuilt integration is weeks of delay. The five steps below help you avoid them.

1. Define goals and channels

First, answer why the company needs a contact center and what it will do every day. Typical jobs:

• inbound support: customer questions by phone, web chat, messaging apps, and email;
• outbound sales: calling back web leads, calling your customer list, payment reminders;
• technical support: tier 1 and tier 2, escalating complex cases to engineers;
• booking appointments and confirming visits.

For each job, write down the channels and expected volume: how many calls and messages a day and when the peak hits. For example, an online store expects 1,200 calls and 500 messages a day, with up to 120 calls an hour at the 10:00 AM-1:00 PM peak. You will need these numbers at every step that follows.

While you are at it, calculate the cost of waiting. If the store currently misses 15% of calls, that is 180 calls a day. Half of them were potential purchases, and at a 10% conversion rate, 9 sales walk out the door every day. Plug your own numbers into the missed calls calculator.

2. Choose a platform and deployment

The platform should bring all your chosen channels into one agent workspace, work with your CRM – such as Bitrix24 or Kommo (amoCRM) – and show real-time reports. Decide separately where it will run:

• In the cloud – the fastest start. The CONTACT.WORK cloud goes live in one day, and we maintain the servers. A good fit if you have no in-house server or telephony specialists.
• On your own servers – recordings and customer data stay inside your network. Security teams in banking, healthcare, and government often require this. Installation takes 2-7 days.
• Hybrid – for example, telephony and recordings on your servers, with AI agents and analytics in the cloud.

See the detailed comparison in our article on cloud vs. on-premises. Check the pricing model up front, too: with CONTACT.WORK the platform is free and there are no licenses – you pay for technical support under your support plan, plus hosting in the cloud.

3. Document processes and metrics

A platform without processes does not work. Before launch, document:

• routing: which calls and messages go to which agent group, and what callers hear in the IVR (phone menu);
• workflows: call scripts, canned responses, when and to whom to escalate a complex question;
• queues: priority for VIP customers, a callback option for people who do not want to wait, and automatic callbacks for missed calls;
• metrics: service level – the share of calls answered within a set time, most often 80% in 20 seconds; average handle time (AHT) – how long a conversation lasts, including after-call work; first contact resolution (FCR) – the share of issues resolved without a repeat contact, with a target of 70-80%.

Why the benchmark is 80/20 and when to change it is covered in our article on service level.

4. Size and train your team

Calculate headcount from workload, not by gut feel. At the store's peak hour – 120 calls, 4-minute average handle time, 80/20 target – you need 11 agents on the line. Some time goes to breaks, training, and sick days: with 30% shrinkage, the schedule for that hour needs 16 people. If the line runs 12 hours a day, 7 days a week, with 8-hour shifts and 5 shifts per person per week, you need 45 people on staff.

This is your biggest expense. 45 people on staff at 160 paid hours a month comes to about 7,200 paid agent hours a month if you keep peak staffing all day. Off-peak, calls drop, and a schedule built around hourly workload costs less. So calculate each hour in the agent calculator and your headcount in the shift calculator.

Why you cannot save "just one" agent. A queue does not grow gradually – it jumps. Per the Erlang C formula, at 120 calls an hour, 11 agents answer 81% of calls within 20 seconds, and the average caller waits about 20 seconds. With one agent fewer, 65% are answered within 20 seconds and the average wait is 49 seconds. With two fewer, it is 40% and more than two and a half minutes: callers hang up, and the payroll savings turn into lost sales.

Besides agents, you need supervisors – they watch queues in real time and help with difficult calls – a manager who owns the budget and the metrics, and an analyst who digs into the reports; in a small team the supervisor takes on that role. Train agents on more than the software: they need product knowledge, the skill to handle an unhappy customer, and fluency in the CRM. A knowledge base with ready-made answers helps new hires get up to speed.

5. Run a pilot and improve

Do not try to make everything perfect for day one. Launch in stages:

1. Pilot: 2-4 weeks with part of your agents and one or two channels.
2. Review: check service level, abandonment rate, and average handle time, listen to call recordings, and ask agents what gets in their way.
3. Adjust: fix routing, scripts, and settings.
4. Scale: add the remaining channels and the whole team.

After launch, review reports once a week, listen to a sample of calls, and remove whatever gets in the way of agents and customers.

What to do this week

1. Export at least a month of hourly call and message volumes from your phone system and CRM.
2. Calculate how much revenue walks away with missed calls – that is your budget argument.
3. Calculate agents for the peak hour and headcount by shift.
4. Ask your security team whether customer data can be stored in the cloud.
5. Pick a group of agents and one or two channels for the pilot.