How to start a call center in Kenya

Two different projects go by this name, and they need almost nothing in common.

The first is a company that wants to answer its own customers properly — a SACCO, an insurer, a hospital, a retailer — and has decided that one mobile phone and a goodwill rota is no longer it. The second is a business whose product is answering calls: an outsourcer, a BPO, taking other people’s call volume for a fee.

Almost everything written about starting a call center is about the second. Most people asking the question are the first. This covers both, and says plainly which parts only apply to one.

1. Decide which of the two you are building

If you are answering your own customers, the hard part is organisational: who answers, during what hours, and what they are allowed to decide without asking. The technology is a weekend. Skip the licensing section below — you do not need one.

If you are selling call handling as a service, the hard part is commercial: clients, margins per seat, and proving quality to somebody who is handing you their reputation. Your costs are dominated by people and premises, and your risk is a single client leaving.

Confusing the two is the most expensive mistake available here. A company that builds an outsourcer’s floor to answer its own fifty calls a day has bought a building it did not need.

2. The licence question

Answering your own customers’ calls needs no telecoms licence. You are a business with a phone number, like every other business with a phone number.

Providing call center services to other organisations, or reselling voice capacity, is where the Communications Authority of Kenya becomes relevant, and where you should get an hour with somebody who does this for a living rather than taking advice from an article. The categories and the fee schedules change. Ask before you sign a client contract, not after.

Two things apply to everyone either way. You are a data controller under the Data Protection Act 2019 the moment you keep recordings or customer records, which means registering with the ODPC where the thresholds apply and being able to say what you keep and for how long. And if you are calling people who did not ask you to, consent and suppression lists are not optional — more on the recording side here.

3. Get the number right before anything else

The number is the part that is hard to change later, because it ends up printed on things. Three practical rules:

• One advertised number, not one per department. A menu does the routing now. Every extra number is another thing to print, remember and maintain.
• A landline if you want to look established, a mobile if your customers already have it saved. Toll-free only where the cost of the call is genuinely what is stopping people ringing.
• Ask whether you can keep the number you have. Eight years of printed material is an asset. Where a number cannot be ported, forward it into the new system and retire it on your own schedule.

There is a fuller comparison of the three types if you have not decided.

4. Work out how many people you need

This is arithmetic, not instinct, and getting it wrong in either direction is costly. The crude version that is good enough to start:

Take the calls you expect in your busiest hour. Multiply by the average length of a call in minutes. Add the time an agent spends after the call finishing the note — real, and usually a third of the call again. Divide by 60. That is the number of people who must be on the phone simultaneously to answer everything with no queue at all.

Then add roughly 30%. The spare capacity is not waste; it is what stops a ten-minute spike turning into a forty-minute queue, and the mathematics of queueing is unkind to teams running flat out. Finally, divide by how much of a shift anybody is actually available — allowing for breaks, training and absence, assume around 80%.

So: 60 calls in the peak hour, 4 minutes each plus 90 seconds of wrap, is 330 minutes, is 5.5 simultaneous agents. Add 30% for the spikes: 7. Divide by 0.8 for availability: 9 people rostered to cover that hour.

Two things fall out of that. Nine is a lot more than most people guess from “sixty calls an hour”. And three is the floor below which none of this helps, because the thing that fixes missed calls is a queue ringing more than one phone.

5. The floor, the kit and the connection

For an in-house desk, this is smaller than people expect: a laptop and a decent USB headset per agent, and a connection that will carry the calls. Agents answer in a browser, so there are no handsets to buy and nothing to rack.

Budget roughly 100 kbps per concurrent call, both directions, and treat that as sacred — a single fibre line shared with everybody streaming video is how call quality dies. Where it matters, get a second line from a different provider; a call center with one internet connection has one bad morning away from being a call center with no phones.

For an outsourcing floor, add premises, power backup that genuinely covers a full shift, a supervisor position per ten or so agents, and a quiet room for training. This is where the money goes, and it is why the two projects are not comparable.

6. The software

What you need from call center software, in the order it matters:

1. A queue that rings a team. The single feature that fixes missed calls.
2. A menu. So the account query stops arriving at the sales desk. Two or three options, in English and Kiswahili.
3. A call log. Every call, answered or not, with who took it and what happened. The first week of this is usually a shock.
4. Reports. Calls offered, answered, abandoned, and how long people waited — by hour, so you can see where you are short.
5. Recording. For training and for the argument about what was said. With retention you control.
6. Outbound. Click to call and campaign lists, once inbound is working.

Everything beyond that is useful but not load-bearing at the start. Be wary of buying a predictive dialler, workforce management or a full CRM in month one; you cannot configure any of them sensibly before you know your own numbers.

The one thing worth checking carefully is pricing shape. Call center software billed per seat per month with incoming calls free suits an inbound operation, because answering calls is almost all of what you do. Software that charges per inbound minute does not — the cost grows with your success at getting people to ring you.

7. What it costs to run

For an in-house desk of ten in Kenya, the recurring monthly costs are roughly:

• People. By far the largest line, and the one that does not compress.
• Software. Seats, from a few thousand shillings a month for a small team. Ours starts at KSh 1,780 for three seats with a Kenyan number included; the rate card is public.
• Calls. Nothing on the ones coming in, if your platform does not charge for inbound. Outgoing at a few shillings a minute, billed per second.
• Numbers. A monthly rental each. One is usually included; extras here are KSh 1,000 a month.
• Connectivity and power. Modest for an in-house desk, significant for a floor.

There is a longer breakdown of the costs, including the setup nobody quotes for.

8. The first month

Open in a small way and let the reports tell you what you got wrong. Specifically:

• Start with one menu, two options and one team. Add structure when a report proves you need it, not in advance.
• Publish your opening hours and record something sensible for outside them. A call that goes to silence is worse than a call that goes to a recording.
• Read the abandoned calls every morning for the first fortnight. They are the people you are losing, and they are the only metric with revenue attached.
• Ring your own number from a mobile, every week, as a customer would. This finds more problems than any dashboard.

The part that is actually hard

None of the above is the difficult bit. The difficult bit is that a call center exposes how your organisation makes decisions. The moment calls are logged, queued and measured, you can see exactly how long a customer waits for an answer that someone has to go and ask a manager for.

Teams that succeed at this give the people on the phone the authority to settle the common cases themselves. Teams that do not end up with an immaculately instrumented queue of people waiting for permission. Buy the software in an afternoon; spend the month on that.