QVCCS innovation · How-to guides
Borrowing agents safely: conditional group routing vs Conditional Group Activation
When one queue is drowning and another is coasting, Genesys Cloud CX can lend agents across queues automatically, but only as safely as the rules allow. We explain how conditional group routing really evaluates its rules, why Genesys now recommends Conditional Group Activation instead, and how we design, test and migrate both.
Did you know that if the first conditional group routing rule is not met, Genesys Cloud treats the queue as healthy, deactivates the groups of every later rule and evaluates no further rules, rather than moving on to rule 2?
Did you know that Conditional Group Activation counts agents who are working on workitems as idle in its idle agents KPI, while agents deactivated by CGA or CGR are not counted?
Did you know that conditional group routing cannot route workitems, and that CGA and CGR can never both be configured on the same queue?
Did you know that CGA evaluates queue health on a schedule, for example every 30 seconds, so a group it opens is available to every waiting conversation, not only to the one that has just arrived?
01
09:40 on a Monday: one queue drowning, another coasting
It is 09:40 on the Monday after a billing run. The Billing queue's estimated wait time is climbing past two minutes. Across the floor, the Sales queue has agents sitting idle between calls, several of whom were trained on billing enquiries last quarter. A supervisor can see both numbers on the queue dashboards and could move people by hand, editing queue memberships and hoping to remember to put them back at lunchtime. Genesys Cloud CX offers two native ways to do this automatically, and the difference between them matters far more than their similar names suggest.
Conditional group routing (CGR) is a routing method that lends groups of agents from one queue to another only while the KPI rules you set are met. Conditional Group Activation (CGA) is a separate gatekeeper that opens and closes agent groups on a schedule, alongside whichever routing method the queue already uses. Genesys now recommends migrating CGR rules to CGA and plans no further enhancements to CGR. When we first wrote about CGR in July 2023, we described its rule logic the way many early write-ups did, and part of that description was wrong. This guide sets the record straight, explains both features, and shows how we design, test and migrate them.
02
What conditional group routing does, and what it shares
CGR shares groups of agents between queues based on rules. Each rule combines a KPI with agent availability, and the only KPI CGR offers is estimated wait time (EWT). You choose conditional group routing as the queue's routing method and add up to five rules on the queue. Each rule tests the EWT of a queue against a threshold, names the groups to make available when the condition is met, and carries a wait period. CGR supports general groups, skill expression groups and work teams, and a rule can make several groups available at once.
Two membership points catch people out. First, a group can only appear in a rule if it is a member of the original queue, and only queues in the same division as the original queue are offered when you build rules. Second, individual agents, and any member groups that are not placed in a rule, are always eligible: they are the queue's home team. Genesys recommends deliberately keeping some groups out of the rules for exactly this reason, so CGR looks to your own people first and only reaches further when they are not available. It also recommends grouping agents with similar skills, because CGR works best where several queues share common groups of agents.
03
How CGR evaluates its rules, step by step
When an interaction enters a CGR queue, the agents who are not in any rule are available straight away. Genesys Cloud then evaluates rule 1, which tests the original queue's own EWT to decide whether the pool needs expanding at all. If the condition is met, the groups in rule 1 become available alongside the home team, and the interaction waits for rule 1's wait period. If it is still unassigned when that period expires, Genesys Cloud evaluates rule 2.
This is the point our 2023 article got wrong. If rule 1 is not met, Genesys Cloud does not move on to rule 2. It treats the queue as healthy, deactivates rule 1's groups, the subsequent rules and any other groups previously activated for CGR, and evaluates no further rules. Rules 2 to 5 behave differently, because they test the queue that is lending agents. A lender's groups are made available only if that queue's KPI meets its threshold, so its own customers are not short-changed. If a later rule is not met, its groups are deactivated and the next rule is evaluated. If any request involved in a rule's evaluation times out, the evaluation fails and Genesys Cloud evaluates the next rule.
Two further behaviours shape every design. Once an interaction reaches the last rule, it simply waits to be assigned: the last rule's wait period moves it nowhere, and there is no fallback routing method. And although CGR evaluates each interaction individually, it applies its conclusions to interactions that are already waiting. When a group is deactivated for a new arrival, it becomes unavailable to earlier waiting interactions too, even if they had been offered it; when a group is activated for a new arrival, earlier interactions can use it as well. Under load, the pool for the whole queue breathes in and out with each new interaction.
Read this diagram as text
A queue box on the left sits beside three numbered rule rows, each running left to right from a condition to a group to a wait, with three outcome boxes underneath.
- The queue box, "Queue · Support", shows "Always eligible – Queue members not in any rule" with three agent icons, then "Pool grows as rules are met: + Team Amber + Team Birch + Team Cedar".
- Rule 1: "If EWT of Support ≥ 40 s" ("Is the original queue overloaded?"), then add group "Team Amber", then "Wait if unassigned: 20 s, then rule 2". An arrow leads down from this wait to rule 2's wait.
- Rule 2: "If EWT of Sales ≤ 25 s" ("Is the lending queue healthy?"), then add group "Team Birch", then "Wait if unassigned: 15 s, then rule 3". An arrow leads down to rule 3's wait.
- Rule 3: "If EWT of Billing ≤ 25 s" ("Is the lending queue healthy?"), then add group "Team Cedar", then "Wait if unassigned: 10 s, then next rule".
- Outcome "Rule 1 not met": "Queue is healthy: later rules and borrowed groups are deactivated".
- Outcome "Later rule not met": "Its group is deactivated and the next rule is evaluated".
- Outcome "Evaluation fails": "For example the EWT request times out: the next rule is evaluated".
- A footnote reads "Up to five rules · EWT KPI · queues in the same division · names and values fictitious", and a strip at the bottom reads: "Borrow agents only while your queue is struggling and the queue lending them is healthy."
04
Where CGR needs care
CGR compares one KPI value with one threshold, so it is best suited to queues that carry a single media type. On a mixed-media queue where each media type has a different EWT, it cannot compare the individual media-level values with the threshold. CGR does not support last agent routing, and it does not support workitem routing at all, so queues serving Work Automation need a different approach. If you forecast and schedule with Genesys Cloud workforce management, Genesys suggests managing expansion with skill expression groups that share the same skills at different proficiency levels, which keeps the groups meaningful to planners as well as to routing.
The rule arithmetic deserves respect too. A rule 1 threshold set too low makes borrowing the norm rather than the exception. A lender threshold set too generously lets a busy queue give away its last available agents. Wait periods add up: with five rules and a 30-second wait on each, an interaction does not see the full pool until two minutes after it arrives. And because each new arrival can change group availability for everyone already queued, behaviour under real load is far easier to understand in a test organisation with controlled traffic than on a whiteboard.
05
Conditional Group Activation: a gatekeeper, not a routing method
CGA approaches the same problem from the other side. Rather than evaluating each interaction, it runs as a parallel background process that evaluates queue health periodically, for example every 30 seconds, and marks groups as available for all new and waiting conversations while the activation conditions are met. It is not a routing method. The queue keeps its own routing method, whether standard, bullseye, preferred agent or predictive routing, and that method chooses an agent from the pool CGA has opened: agents in groups CGA has made available, individual members, and member groups that are not linked to any CGA rule. Direct routing and last agent routing ignore CGA group availability altogether.
You configure CGA on the queue's Members tab, under Conditional Group Activation. An optional pilot rule checks the current queue first; if its conditions are not met, every group linked to the numbered rules is deactivated and only the home team is considered. Then come up to five rules, each with up to ten conditions, so one rule can test several queues and KPIs together, such as the lending queue's EWT alongside its idle agents, combined in a condition expression. The KPIs are estimated wait time and idle agents, meaning agents whose presence is On Queue and Idle in the selected queue. As with CGR, shared groups must be members of the queue, and only queues in the same division are offered.
The interplay with routing methods is precise. With bullseye routing, a group in a ring is considered only if CGA has also made it available; if not, the ring's skill requirement is relaxed and the next ring is evaluated. With predictive routing, only agents who are both ranked by the model and in CGA-available groups are considered. With preferred agent routing, preferred agents must also be in available groups, and CGA cannot be configured when CGR is the timeout routing method. CGA and CGR are mutually exclusive on any one queue. We manage CGA rules as code, and CX as Code provides a dedicated resource for them, outlined below.
resource "genesyscloud_routing_queue_conditional_group_activation" "billing_cga" {
queue_id = genesyscloud_routing_queue.billing.id
# Pilot rule: is Billing itself under pressure?
pilot_rule {
condition_expression = "C1"
conditions {
simple_metric {
metric = "EstimatedWaitTime"
}
operator = "GreaterThan"
value = 60
}
}
# Rule 1: lend Sales agents while Sales answers quickly or has idle agents
rules {
condition_expression = "C1 or C2"
conditions {
simple_metric {
metric = "EstimatedWaitTime"
queue_id = genesyscloud_routing_queue.sales.id
}
operator = "LessThan"
value = 20
}
conditions {
simple_metric {
metric = "IdleAgentCount"
queue_id = genesyscloud_routing_queue.sales.id
}
operator = "GreaterThanOrEqualTo"
value = 2
}
groups {
member_group_id = genesyscloud_group.sales_billing_trained.id
member_group_type = "GROUP"
}
}
}06
Migrating from CGR to CGA without surprises
Genesys's note on both overview pages is clear: it does not plan any future enhancement to CGR and recommends that you migrate your CGR rules to CGA. CGR continues to work, so there is no cliff edge, but every new capability we are asked for, such as a second condition, an idle-agent check, or compatibility with bullseye and predictive routing, sits on the CGA side. A migration is a redesign rather than a copy, because the two features answer different questions: CGR asks what this interaction may use now, while CGA asks which groups the queue may use at this moment.
Our mapping starts simply. CGR rule 1, which tests the original queue, usually becomes the CGA pilot rule. Each lender rule becomes a numbered CGA rule, often with an extra condition CGR could not express, such as a minimum number of idle agents on the lending queue. What has no direct equivalent is the per-interaction wait period, because CGA availability applies to the whole queue at a point in time. Where the business wants the pool to widen the longer a contact waits, we pair CGA with bullseye routing: the ring delays provide the timing, and CGA provides the health gate on each borrowed group.
Cut-over happens queue by queue. Because CGA cannot be configured on a queue that has CGR enabled, each queue moves off conditional group routing to its new routing method and gains its CGA rules in one controlled change. We capture the existing configuration as code first, so the rollback is a known, tested state rather than a memory. Then we watch EWT, service level and abandons on both the borrowing and the lending queues through the first busy periods, and tune thresholds from evidence rather than instinct.
07
How QVCCS designs, tests and delivers group sharing
Group sharing is a routing decision with consequences for two sets of customers, so we treat it as a design exercise from the start. In Discover, our Senior Business Consultant analyses interval data for every queue involved: volumes, EWT, service level and which agents are trained for what. In Define, each rule becomes a testable acceptance criterion, for example: given Billing EWT above 60 seconds and Sales with at least two idle agents, when the next evaluation runs, then the Sales billing-trained group is available to Billing. The Solution Architect records queues, divisions, member groups, pilot and numbered rules, conditions and the routing method alongside in the low-level design, with each threshold justified in the design decision log.
Senior Developers build the configuration with CX as Code in version control, under the Senior Platform Practice Lead's engineering standards and four-eyes review. Our Systems Integration Tester then proves the behaviour in a test organisation with controlled traffic: pilot rule not met, lender unhealthy, home team busy, a group in a bullseye ring that CGA has not opened, and the effect of each new arrival on contacts already waiting. Cut-over follows a rehearsed runbook with a rollback, delivered remotely, and supervisors are briefed on what to expect when groups are borrowed. Every delivery team member holds, or is completing, Genesys Cloud certification, and in Run and evolve we review each Genesys release for routing changes and revisit thresholds in optimisation reviews.
How it compares
Conditional group routing and Conditional Group Activation side by side
Native behaviour as documented in the Genesys Cloud Resource Center, checked in October 2026.
| Aspect | Conditional group routing (CGR) | Conditional Group Activation (CGA) |
|---|---|---|
| What it is | A routing method selected on the queue. | A gatekeeper that works alongside the queue's routing method. |
| When it evaluates | When each interaction arrives, then rule by rule as wait periods expire; conclusions also apply to interactions already waiting. | Periodically, for example every 30 seconds, for the queue as a whole. |
| First check | Rule 1 always tests the current queue; if it is not met, no further rules are evaluated. | An optional pilot rule tests the current queue; if it is not met, groups in the numbered rules are deactivated. |
| Rules and conditions | Up to five rules, each with one EWT condition and a wait period. | Up to five rules, each with up to ten conditions; no per-interaction wait period. |
| KPIs | Estimated wait time. | Estimated wait time and idle agents. |
| Routing methods | Is the routing method, and can be the timeout method for preferred agent routing; no last agent routing. | Works with standard, bullseye, preferred agent and predictive routing; ignored by direct and last agent routing. |
| Groups | General groups, skill expression groups and work teams that are members of the queue; queues in the same division. | Groups that are members of the queue; queues in the same division. |
| Outlook | Still supported, but Genesys plans no further enhancements and recommends migrating to CGA. | The target Genesys recommends for CGR rules. Never configured on the same queue as CGR. |
Workitems: Genesys documents that CGR does not support workitem routing.
Questions
Common questions
What is the difference between conditional group routing and conditional group activation?
Conditional group routing is a routing method: up to five rules, evaluated as each interaction waits, with estimated wait time as the only KPI. Conditional group activation is not a routing method. It runs in the background, checks queue health periodically, for example every 30 seconds, and opens extra groups for all waiting conversations while the queue keeps its own routing method. Genesys recommends moving from CGR to CGA.
Can conditional group activation be combined with bullseye routing?
Yes. CGA works alongside standard, bullseye, preferred agent and predictive routing. With bullseye, a group in a ring is considered only if CGA has also made it available; if not, the ring's skill requirement is relaxed and the next ring is evaluated. Direct routing and last agent routing ignore CGA group availability.
Can conditional group routing use a condition other than estimated wait time?
No. Each CGR rule tests a queue's estimated wait time against a threshold. Conditional group activation adds idle agents, meaning agents who are On Queue and Idle in the selected queue, and allows up to five rules with up to ten conditions each, so one rule can test several queues and KPIs together.
What happens to an interaction after the last conditional group routing rule?
It waits to be assigned. Once an interaction reaches the last rule, the wait period moves it nowhere and there is no fallback routing method. Plan the last rule as the widest pool you are prepared to use, and handle long waits with in-queue flow options such as callbacks.
Which agents are always eligible in a CGR or CGA queue?
Individual members of the queue and member groups that are not placed in any rule are always eligible: they are the queue's home team. Genesys recommends deliberately keeping some groups out of the rules, so the queue looks to its own people first and reaches further only when they are not available.
The takeaways
- If CGR rule 1 is not met, the queue is treated as healthy and no further rules run.
- Lender rules protect the lending queue: its groups are shared only while its KPI meets the threshold.
- CGA opens groups for the whole queue on a timer and lets your routing method choose the agent.
- Genesys recommends migrating CGR rules to CGA; a migration is a redesign, not a copy.
- We prove every rule under controlled load before it touches live customers.
Designed, built and supported by the same certified team that delivers Genesys Cloud CX solutions for enterprises and integrators.
Sources
- Conditional group routing overviewhelp.genesys.cloud
- Conditional Group Activation overviewhelp.genesys.cloud
- Create and configure queueshelp.genesys.cloud
- Enhanced conditional routing with conditional group activation (CGA)help.genesys.cloud
- Why am I not able to change the queue name in the first CGR rule?help.genesys.cloud
- CX as Code: genesyscloud_routing_queue_conditional_group_activationregistry.terraform.io