What is routing in a contact center?

Routing starts before anyone hears hold music. The first rules read what the system already knows about a call, which is usually the number that was dialed and the time of day.

Xima, the platform behind Atlantech's contact center service, gives one documented example. Every call enters one Primary IVR, the top-level call flow. Conditions inside it check the dialed number and can send the call to a different menu, skill group, agent or outside number "before the caller ever hears a prompt." Schedules are another condition a flow can check. Xima's Time of Day profiles cover business hours, lunch periods and after-hours coverage, and separate Holiday profiles handle closures. A call that arrives after closing can go to voicemail or an on-call team without ever joining the daytime queue.

Then come the questions only the caller can answer. A menu asks them to press 1 for sales or 2 for billing, and the answer picks the queue.

A lookup can replace some of those questions. Xima calls this a data dip. During the call, the IVR sends a value such as the caller's phone number to a configured external web service. That service searches a connected data source, such as a CRM, and returns fields about the customer. Xima's examples include an assigned representative, a language preference, VIP status and a region. The call flow can then branch on what came back, so a known customer skips part of the menu. A lookup rule inherits the mistakes in the record, though. If the language field is wrong, that customer lands in the wrong queue on every call whose flow branches on that field.

Written channels get the same treatment with different inputs. Xima's email routing rules match keywords in the subject line or body and send the message to a skill group. The rules are checked in ranked order.

Entry rules also need a default. Xima's guide to its older Call Routing menu warns that no default skill is set out of the box, and that a call matching no rule "is disconnected immediately unless a default skill is configured." Whatever platform you use, decide where an unmatched contact goes and place a test call to confirm it.

Once a contact is in a queue, routing has to pick a person. Genesys Cloud's ACD documentation gives one example, which filters first and chooses second. It lists the agents who are on the queue, removes anyone who lacks the language or skill the contact requires, and picks from those who remain. If no qualified agent is left, the contact waits.

Amazon Connect's default is simpler: the contact goes to the agent who has been available longest. Xima lets each skill group choose its own method. The methods themselves, such as longest idle, rotation and highest skill, belong to the automatic call distributor, and we describe them there.

Genesys Cloud also offers predictive routing, which scores each available agent for the contact with a machine learning model and falls back to standard routing when three or fewer agents are available. Genesys recommends running its comparison test against your current method for at least two weeks.

Where the rules are stored differs by platform, which matters when you move from one to another. Genesys Cloud attaches a routing method to each queue. Amazon Connect attaches a routing profile to each agent, and the profile names the channels and queues that agent serves. A routing plan has to be rebuilt when you switch, because the settings don't map one to one.

The opposite case comes up at busy times, when one agent comes free and several customers are waiting. In a 2002 survey of call center research, Ger Koole and Avishai Mandelbaum treat these as two separate rules. Agent selection covers how an arriving call picks among idle agents. Call selection covers how a free agent picks among waiting calls.

Genesys Cloud and Amazon Connect both answer waiting contacts in arrival order by default. Priority is how a business overrides that order, and the two platforms attach it in different places.

Genesys Cloud puts priority on the contact. Under its default scoring, each priority point counts as one minute of waiting, so the system treats a prioritized call as if it had arrived earlier. Suppose two calls are waiting for the same group of agents. Call A arrived at 10:00:00 with no priority, and call B arrived at 10:01:30 with a priority of 3. B is treated as though it arrived at 9:58:30, so B is answered first. Give B a priority of 1 and it is treated as a 10:00:30 arrival, which leaves A ahead.

One point is worth one minute, so it moves a caller ahead only of unprioritized callers who arrived less than a minute earlier. And raising everyone's priority changes nothing within a queue, because every call shifts by the same amount.

Amazon Connect puts priority on the queue. In its example, a group of sales agents has the Sales queue at priority 1 and the Support queue at priority 2, so they take a Support contact only when no Sales contact is waiting. A second configuration adds a delay. With Support at priority 1 and a 30-second delay, a Support contact is treated as priority 1 once it has waited 30 seconds. Xima takes the queue approach too. Each skill group gets a Skill Priority from 0 to 10, which ranks the groups for an agent who is eligible for more than one.

Strict matching has a cost. If only your strongest billing agents may take billing calls, callers wait whenever those agents are busy, even while other agents sit idle. Xima and Genesys Cloud both let the match loosen as the wait grows.

Xima does this with Availability Tiers. The first tier applies the moment a call arrives and sets a minimum skill level. Later tiers lower the minimum after a set time in queue. Suppose a billing skill group starts at level 8 and adds a second tier at level 5 after two minutes. A call is offered only to agents at level 8 or higher for the first two minutes, and after that to anyone at level 5 or above. The call never leaves the skill group, so the caller's position in line and estimated wait time remain consistent.

Genesys Cloud calls its version bullseye routing. Agents are arranged in rings. If nobody in the inner ring is free within a set time, the pool widens to the next ring, either by admitting agents with lower ratings or by dropping one of the required skills. Agents in the inner rings keep first claim if they come free during the wait. Genesys gives the reason plainly: strict skill enforcement hurts a queue's service level and abandon rate when the agents who are free don't hold every required skill.

How long to wait before loosening is a judgment the business has to make. Koole and Mandelbaum observed in 2002 that skill-based routing comes "as a list of options that managers can choose from, but without any guidelines to accompany them." Vendor documentation still explains what each setting does and leaves the values to you. We'd start with a simple rule, watch wait times and abandoned calls for a few weeks, and add tiers only where the numbers show a problem.

The principles carry over to written channels, and each channel has its own rules, as Xima's email keyword rules show. Capacity is the other difference. An agent can hold one phone call but several chats, so routing needs a limit for each channel. Genesys Cloud's default limits allow one call, one email and up to four chats at a time. Xima uses a single number per agent called Max Simultaneous Sessions. Its guide illustrates a value of 2 as one call plus one chat, and 3 as one call plus two chats.

Those limits decide how one group of agents is shared between channels, which is the subject of what a unified queue is. Whether a customer can switch channels without starting over is a separate matter, covered in omnichannel versus multichannel customer service.

Routing can only distribute the agents who are logged in and free. Genesys's bullseye documentation says agents who are busy or at capacity are not candidates, however well they match the contact. If waits stay long after the rules are tuned, look at how many agents are scheduled for that hour.