Skills-based routing: sending calls by agent skill

The short answer

Skills-based routing sends a call not to the first free agent but to an agent who has the skill that call needs. A skill can be a language, a product, a region, a technical level or an authority. The aim is to cut transfers and get the caller's question to the right person in the first conversation.

Not every team needs this model. In a team of five or six generalist agents, skill rules split the queue artificially and increase waiting. Skills-based routing makes sense when there are real differences between agents and a call that reaches the wrong person always has to be transferred.

What a skill is and is not

A skill is an agent's ability to finish a certain type of call well: speaking Russian, knowing loan products, having the authority to handle B2B customers. A job title or length of service is not a skill. Writing "senior agent" says nothing; "closes mortgage questions without transferring" is a skill.

Every skill must be linked to a customer need. If no call type requires a particular skill, do not put it into a routing rule — it only complicates reporting.

How to identify what a call needs

To route by skill, the system first has to know what the call wants. There are four sources: the number dialled (a separate line per product), a menu choice, a conversational agent's question, and data stored in the CRM (language, segment, account manager). The more reliable the source, the more accurate the routing.

The skills matrix

Put agents in rows and skills in columns, and give each cell a level. This table is the basis for both the routing rules and the training plan.

1. 1Level 0Does not take this call.
2. 2Level 1Takes it but needs help in complex cases — fallback group.
3. 3Level 2Finishes the call independently — primary group.
4. 4Level 3Solves complex cases and supports others — expert.

Primary and fallback groups

The most forgotten part of a skill rule is the waiting threshold. The call goes to the primary group first; if nobody becomes free within a set time — for example 30–60 seconds — it opens up to the fallback group. Without a fallback group, callers hang up while waiting for the ideal agent. A level-1 agent after a short wait is often better than a long wait.

Illustrative example: insurance support

Not a real customer case. The team has 14 agents. Calls arrive in three languages and relate to three products: car, property and health insurance. Previously calls were distributed from one general queue, and roughly one in five was transferred to a second agent — mostly because of language and health-insurance questions.

The new rule uses two skills: language and health insurance. Car and property questions stay generalist, because every agent knows them. Russian-language calls go first to the five people at level 2 in Russian, and after 45 seconds to the four at level 1. The result is two skills rather than six — much easier to manage.

The conversational agent's role

The weakest link in skills-based routing is identifying what the call needs. In a menu the caller guesses the option; a conversational agent hears the question in the caller's own words and can ask one or two clarifying questions: "Is this about your existing policy, or do you want a new one?" That way the call reaches the operator with the right label.

The agent can also recognise skills such as language from the first sentence. But it does not replace the skill rule: the agent identifies the need, and the rule decides which group the call goes to. Keeping these two layers separate lets you test each one on its own.

How to roll out a new rule

Do not apply a skill rule to every call on the same day. Start with one skill — the one causing the most transfers — and for two weeks track the transfer rate, waiting time and the calls reaching the fallback group. If the results are good, add a second skill. Before every change, explain to agents which calls will now come to them — otherwise they treat the new call type as "misrouting".

When a small team does not need it

• Every agent can finish the same call types
• The transfer rate is already low
• There are only one or two calls in the queue at a time — splitting increases waiting
• Nobody is responsible for keeping the skills matrix up to date

In such a team, a better approach is a simple rule for one or two rare skills (English, for example) and a general queue for everything else.

Common mistakes

• Too many skills: every agent has ten, and nobody understands the rules
• No fallback group or waiting threshold
• The matrix was filled in once and is not updated when agents change
• All difficult calls go to experts, who burn out, and quality drops
• Trying to solve a skills gap with routing alone rather than training

Keeping the skills matrix alive

The matrix needs an owner — usually a team lead or the workforce management (WFM) lead. In a monthly review, ask three questions: which skill always has a long queue, which agent has moved up a level, and which skill is no longer required by any call. When someone joins or leaves, the matrix is updated the same day.

What to measure

• Transfer rate — before and after the skill rule
• Waiting time and abandonment per skill
• The share of calls reaching the fallback group — if high, the primary group is too small
• Experts' workload compared with other agents

Limits

Skills-based routing only works when the call's need is identified correctly; a wrong menu choice or outdated CRM data sends the call to the wrong group. Skill levels can be assessed subjectively — check them against quality results. This article describes industry practice (for example, the queue and skill models in Amazon Connect and Twilio documentation); capabilities on a specific platform may differ.

A call's need in Vexvon

In Vexvon AI Call Center priority rules decide where a call goes: business hours, the dialled number and the caller's number (for example, a VIP list going straight to the sales department). Before a rule goes live, the simulator answers "where would a call from this number at this time land?". The AI agent asks the reason for the call in conversation, extracts the fields you define in the scenario and passes the call to a live operator, who sees the summary in advance. The conversation language is set in the scenario.

Handoff by language is covered in the multilingual voice agent, and the menu itself in IVR menu best practices.

First step

Look at last month's transferred calls and write down the reasons. If most reasons fall into one or two skills, add only those to the matrix. More articles are in the IVR & call routing section; to set up your routing together, get in touch.