Genesys Cloud routing explained: bullseye, conditional, preferred and predictive

Genesys Cloud ACD uses each queue's routing method to match interactions with agents. For skills-based routing, an evaluation method decides how the skills an interaction requests are compared with agents' skills. A scoring method decides which waiting interaction goes first when an agent becomes free, and utilisation decides whether an agent has room for another interaction at all. An agent must also be a member of the queue and on queue. Most routing surprises come from one of these settings interacting with another, so it helps to treat them as one design rather than four separate options.

Genesys documents five routing methods, set on the queue's Routing tab. Standard routing is the default. The others either widen or narrow the pool of agents over time, or use a model to pick the best match. Several can be combined with conditional group activation, which runs alongside the routing method rather than replacing it. The table summarises each in Genesys's words.

Agents hold ACD skills and language skills, each rated from 0 to 5 stars. Genesys documents that skill changes take effect immediately, whether or not the agent is on queue, and that an agent who does not speak an interaction's language does not receive it, whatever their ACD skills. Interactions request skills in Architect: the Transfer to ACD action can add ACD skills, a language skill, a priority, preferred agents and a direct agent, and in-queue flows can change skills with the Set Skills action. If a flow sets no language skill, the flow's default language applies.

Genesys has added ACD skill expression filters, which combine ACD skills, language skills and proficiency into a single rule per interaction. They work with standard or predictive routing using the All skills matching evaluation method, and Genesys states that they do not work with preferred agent, last agent, direct or bullseye routing. Where a flow sets both a filter and individual skills, the filter is used. Workforce management does not use skill expression filters, so forecasts and schedules still plan by individual skills. Keep the skill catalogue small and named for the work, not for the team, so that routing, reporting and planning stay aligned.

With standard routing, the evaluation method decides how requested skills count. The table sets out the three methods. Two details from Genesys matter in practice. Best available skills looks at the 100 agents with the longest time since their last interaction, picks the best average proficiency among them and uses longest idle as the tie-break, so on a large queue it is not a search of every agent. And time since last interaction is counted across all of an agent's queues, looking back seven days, so an agent who works several queues is not ranked as idle on one while busy on another.

When interactions are waiting and an agent becomes free, the scoring method decides the order. With conversation score, Genesys combines arrival time and priority: one priority point counts as one minute, so a call with a higher priority is treated as if it arrived earlier. With priority score, the highest priority goes first, and time in queue breaks ties. Genesys notes that where queues use different scoring methods, interactions in priority-score queues are routed first, and that changing the method while interactions are waiting can assign them in an unexpected order. Set the scoring method when you create the queue, or when it is empty, and use priority deliberately rather than as a quick fix for one caller.

Bullseye routing offers an interaction to a small, targeted group of agents first and widens the pool if nobody suitable answers in time. Genesys describes the expansion as concentric rings, each a fallback for the one before. Each ring has a timeout from 2 seconds to 72 hours. Rings can be built in two ways: by assigning users or groups, including skill expression groups, to each ring on the Members tab, or by removing requested skills as the interaction moves out of a ring, up to 50 skills per ring. Removed skills stay on the interaction for reporting but no longer count in the next ring, and agents in earlier rings keep priority as the pool grows.

Skill removal works best with the All skills matching or Best available skills evaluation methods; with Disregard skills, next agent, skills do not matter in any ring. Genesys's Terraform provider allows up to six rings per queue. Keep ring timeouts short enough that customers do not wait for a perfect match that is not available, and check service levels by ring rather than for the queue as a whole.

A bullseye queue in CX as Code (Genesys Terraform provider): specialists first, then skills relaxed after 20 and 40 seconds

resource "genesyscloud_routing_queue" "billing" {
  name                    = "Billing"
  # NONE | BEST | ALL
  skill_evaluation_method = "BEST"
  # the default scoring method
  scoring_method          = "TimestampAndPriority"
  acw_wrapup_prompt       = "MANDATORY_TIMEOUT"
  acw_timeout_ms          = 120000

  media_settings_call {
    alerting_timeout_sec      = 15
    service_level_percentage  = 0.8
    service_level_duration_ms = 20000
  }

  # ring 1: wait 20 s, then drop the specialist skill
  bullseye_rings {
    expansion_timeout_seconds = 20
    skills_to_remove = [genesyscloud_routing_skill.disputes.id]
  }
  # ring 2: wait 20 s more, then drop the language skill
  bullseye_rings {
    expansion_timeout_seconds = 20
    skills_to_remove = [genesyscloud_routing_skill.welsh.id]
  }
}

Conditional group routing lets a queue under pressure borrow agents from groups whose own queues are healthy. Its rules combine a KPI, currently estimated wait time, with agent availability, up to five rules each with a wait period, and rule 1 checks whether the original queue is overloaded. Genesys documents that after the last rule there is no fallback routing method, that it does not support last agent routing or workitems, and that it suits single-media queues best. Genesys has also stated that it plans no further enhancement to conditional group routing and recommends migrating its rules to conditional group activation.

Conditional group activation works differently. It runs as a background process alongside the queue's routing method, evaluating the health of the queue periodically, for example every 30 seconds, and activating or deactivating extra member groups for all waiting conversations at once. Genesys documents that it works with standard, bullseye, preferred agent and predictive routing, but not with last agent or direct routing, and not alongside conditional group routing. With predictive routing, activation decides when more capacity is needed and the model decides who is best. Our Innovation article on borrowing agents safely compares the two in more detail.

Preferred agent routing lets a flow name up to 20 preferred agents with scores from 0 to 100; Genesys can raise the limit to 100 through Customer Care. Interactions go to preferred agents first, regardless of their ACD and language skills, choosing by time since last interaction among agents whose score meets the current threshold, in up to six rings with timeouts from 2 seconds to 72 hours. Preferred agents must be members of the queue and on queue; if none are, the ring timeout is skipped. When the rings run out, the queue's timeout routing method takes over: standard, bullseye or conditional group routing. Genesys also notes that preferred agents are dropped on transfer and that preferred agent routing is not compatible with workforce management scheduling and forecasting.

Last agent routing applies only to threaded email and messages, not voice. It is set on the queue's General tab, to the last agent on any queue (the default), the last agent on the conversation's queue, or disabled, and it needs the agent to be on queue. Genesys documents that it only works when email is sent directly to a queue rather than through an Architect email flow. Direct routing, for voice, SMS and email, sends interactions to a named agent through a queue as ACD work, with a backup agent or queue set in an in-queue flow; a direct agent in Transfer to ACD takes precedence over the queue's routing method.

Predictive routing uses machine learning to match each interaction with the available agent best able to handle it, optimising one KPI per queue: average handle time, next contact avoidance or a custom KPI. It supports inbound voice, email and asynchronous messages. Genesys documents a benefit assessment that needs at least 900 inbound interactions on 45 days within a 90-day period, incurs no cost and needs no subscription, followed by a comparison test and then ongoing operation, with standard routing as the fallback when the predictive timeout expires. Licensing, token use and when it pays back are covered on our predictive routing page, which the AI section keeps up to date.

Utilisation is the maximum number of concurrent interactions Genesys Cloud ACD can assign to an agent, set per media type at organisation level and optionally overridden for individual agents. Genesys's defaults are one call, one callback, one email, and up to four chats, four messages and four workitems. By default email can be interrupted by other interactions, calls and callbacks cannot interrupt each other, and non-ACD calls do not count unless you choose to count them. An agent who is fully utilised is skipped by every routing method, which is why a queue can show agents on queue and interactions waiting at the same time.

Utilisation labels refine this: up to 100 labels per organisation, one per interaction, set in Architect, with media-level limits checked before label limits. Use them to stop, for example, two complex complaint messages landing on one agent at once. Design utilisation with workforce management, because concurrency changes how many agents a forecast needs.

Start with the simplest design that meets the requirement and add complexity only where data shows a need. Most queues work well with standard routing and All skills matching or Best available skills. Reach for bullseye when specialist skills are scarce, conditional group activation when queues should share agents under pressure, preferred agent routing when relationships matter more than skills, and predictive routing when a queue has the volume and a clear KPI. The table gives starting points rather than rules.

Several queue settings sit around the routing method and cause most reported faults. The alerting timeout, from 7 to 59 seconds, decides how long an interaction rings before Genesys puts the agent into Not Responding and returns the interaction to the queue; a Not Responding agent receives nothing until made eligible again, and auto answer removes the alerting timeout altogether. After call work can be optional, mandatory, time-boxed or agent-requested, with a timeout of up to an hour, and it holds agents out of routing while it runs. An organisation setting, Routing Includes Agent Presence, decides whether a presence change resets an agent's idle time, which changes who is next.

Test routing before go-live with scripted scenarios on each queue: matching and non-matching skills, every bullseye ring, a fully utilised agent, a Not Responding agent, priority interactions and transfers. Then watch interaction detail and queue performance views by ring and by agent for the first weeks. QVCCS designs, builds and tests routing like this remotely unless agreed otherwise, and configures it as code so the design can be reviewed and repeated. Every delivery team member holds, or is completing, Genesys Cloud certification.