# AI predictive and affinity routing in contact centers Canonical: https://warmtransfer.net/knowledge/predictive-routing-ai Last verified: 2026-10-02 # AI predictive and affinity routing in contact centers Genesys Cloud predictive routing uses machine learning to rank each agent in the target agent pool by how well that agent is predicted to handle a specific interaction against a chosen KPI[^39]. Afiniti describes Afiniti Pairing as an overlay on existing routing infrastructure that works within the existing routing framework rather than replacing the ACD[^2]. ## How Genesys Cloud predictive routing selects agents In Genesys Cloud predictive routing each queue optimises only 1 KPI at a time, and different queues can use different KPIs[^36]. Average handle time and next contact avoidance are generally available KPIs for Genesys Cloud predictive routing[^19]. Predictive routing supports voice, email and message interactions, where messages include third-party messaging platforms, inbound SMS, Genesys Cloud web messaging and open messaging[^21]. Genesys Cloud defines Standard ACD routing as routing interactions to the next available agent while considering skills according to the evaluation method, and this is the skills-based baseline predictive routing is compared against[^50]. Before scoring, predictive routing filters agents for required language skills and, when skill matching is enabled, non-language ACD skills[^41]. If 3 or fewer agents are available on a queue, Genesys Cloud does not score agents and routes with the standard routing method[^43]. When there is an agent surplus, predictive routing ranks agents by combining their time since last interaction with their predictive score[^18]. When there is an interaction surplus, it ranks interactions by combining waiting time with predictive score and assigns the available agent to the highest-ranked interaction[^28]. The predictive routing timeout must be an integer between 12 and 259,200 seconds[^46]. During the timeout Genesys Cloud first offers to the highest-scoring agents and gradually widens to lower-scoring agents, and on expiry routes with standard routing[^45]. Genesys predictive routing uses white-box models in which each input feature is given a percentage or score representing its importance[^49]. Genesys Cloud shows the top 10 features influencing predictive routing decisions per queue and per media type on the queue's Predictive Model tab[^47]. ## Data, training and retention in Genesys Cloud Genesys recommends at least 90 days and ideally 180 days of data for predictive routing[^26]. A data source is used only if its volume meets a minimum threshold; the thresholds are set at organisation level, are not editable by customers, are updated by Genesys, and their values are not published[^27]. Agent profile data is considered only if it is populated for more than 50 percent of the agents working on queues with predictive routing enabled[^37]. The agent profile fields Genesys lists for predictive routing are hire date, department, certifications, employee type, user skill and user language[^38]. Custom KPI outcomes are supplied through the External Attribution API, with uploads recommended daily or at minimum weekly, and each record must be unique per KPI ID, conversation ID and agent ID[^25]. Genesys Cloud updates predictive routing scoring features with daily data and retrains the models weekly[^48]. KPI processing to build the model can take several hours, and when predictive routing is activated all interactions use standard routing until processing completes[^29]. If a model retrain fails, standard routing is used for the affected media types while predictive routing continues for unaffected media types[^40]. Genesys states that no data is retained in predictive routing models for more than 90 days[^17]. Genesys states that predictive routing does not use personally identifiable information in agent scoring and generates features only from transactional conversation data without PII[^35]. Genesys also states that predictive routing does not use data models that require data such as gender and nationality[^34]. Genesys documentation acknowledges that predictive routing, by learning from the most recent historical conversations, may introduce temporal bias[^42]. ## Measuring benefit in Genesys Cloud Genesys describes 3 phases for adopting predictive routing: benefit assessment, comparison testing, and then either ongoing value monitoring or full activation[^44]. A comparison test compares predictive routing against either standard routing or bullseye routing[^22]. During the test, interactions are routed in an hour-on, hour-off pattern, with predictive routing 50 percent of the time and the comparison method the other 50 percent[^23]. Genesys recommends running the comparison test for at least 2 weeks, noting that high-volume queues may produce results sooner[^24]. Genesys requires at least a 2-week 50-50 comparison test that showed a desirable benefit before ongoing value monitoring is set up on a queue[^32]. Ongoing value monitoring routes interactions with predictive routing 80 percent of the time interval and with the baseline routing method 20 percent of the time interval[^30]. It repeats a 6-day cycle that alternates blocks of 8 hours of predictive routing with 2 hours of the baseline routing method[^31]. Genesys recommends running ongoing value monitoring through 6 cycles of 6 days before treating the comparison as valid, and says accuracy can take up to 2 or 3 months[^33]. Genesys Cloud predictive routing is billed per interaction routed predictively and is consumed through AI Experience tokens[^20]. ## Afiniti behavioral pairing Afiniti describes Afiniti Pairing as working within the existing routing framework rather than replacing the ACD[^2]. Afiniti says Afiniti Pairing impact is measured with continuous benchmarking in which a live control group runs in parallel with AI pairing[^1]. Afiniti's behavioral pairing patent distinguishes behavioral pairing from performance-based routing: PBR tries to maximise each interaction's expected outcome, while behavioral pairing accounts for subsequent pairings to balance agent utilisation and improve overall performance[^8]. The US9300802B1 method orders contacts and agents, for example as percentiles, and selects pairs by comparing differences in ordering between candidate contact-agent pairs[^7]. Afiniti's benchmarking patent describes cycling among pairing strategies such as FIFO and behavioral pairing, alternating more often than once per day or even once per hour[^4]. It gives both 50/50 and unequal 80/20 behavioral-pairing-to-FIFO allocations as examples of on/off cycling[^5]. It corrects combined on/off results for the Yule-Simpson effect, because unequal contact distributions across periods can reverse an aggregated comparison[^6]. ## Amazon Connect routing criteria Amazon Connect routing criteria are a sequence of routing steps, each a set of requirements built from predefined attribute, value, comparison operator and proficiency level, which an agent must meet to be joined to the contact[^10]. Routing steps allow up to 8 AND attributes, up to 3 OR conditions, and proficiency levels compared with >= or as a range from 1 to 5[^13]. When every routing step has expired, the contact is offered to the longest available agent who has the queue in their routing profile[^15]. Amazon Connect documentation describes a routing step that targets up to 10 preferred agents by user ID, including an agent predicted as best fit by the customer's own custom machine learning model, followed by attribute-based steps[^9]. A contact restricted to a preferred agent stays restricted until that routing step expires, even if the agent is offline, busy, in a nonproductive status or deleted[^12]. Since 17 August 2026 Amazon Connect dashboards can filter agents by assigned proficiencies, group metrics by routing step and report contacts queued in a routing step[^14]. ## Webex Contact Center Personalized AI Routing Cisco announced Personalized AI Routing as available for Webex Contact Center in the what's new entry dated 1 October 2026[^53]. Using AI routing for live traffic requires the organisation to hold a specific entitlement, in addition to AI routing being enabled for the organisation[^55]. The AI routing optimisation check returns High potential, Low potential or Failed for each queue analysed[^58]. When evaluation mode starts, Webex Contact Center trains the AI routing model on up to 90 days of historical interaction data for the selected business outcome[^52]. In evaluation mode the model predicts the recommended agent and expected outcome while regular routing continues to handle live interactions[^56]. With AI routing active, existing queue, skill, priority, flow, availability and capacity rules still determine which agents are eligible[^54]. AI routing monitoring compares the selected KPI average under AI routing and under regular routing and reports percentage and absolute improvement[^57]. Reports for evaluation mode and AI routing mode include only interactions longer than 30 seconds[^51]. Webex Contact Center allows at most 100 queues per organisation to use AI routing, at most 100 queues in evaluation mode at once, and 50 queue IDs per optimisation check request[^59]. Since 25 June 2026 Webex Contact Center Weighted Agent Insights routing lets administrators assign Dynamic Skills directly to agents outside skill profiles and apply weighted requirements at queue or contact-type level[^60]. ## See also - [Genesys Cloud CX routing and Architect flows](https://warmtransfer.net/knowledge/genesys-cloud-routing) - [Setting up queues and routing profiles in Amazon Connect](https://warmtransfer.net/knowledge/amazon-connect-routing-profile-setup) - [Webex Contact Center queues teams and routing strategies](https://warmtransfer.net/knowledge/webex-cc-routing) - [Contact center KPI definitions and calculation](https://warmtransfer.net/knowledge/contact-center-kpis) - [Contact centre AI data residency retention and training boundaries](https://warmtransfer.net/knowledge/ai-data-residency-governance) ## Applicability Applies to: Genesys Cloud, Afiniti Pairing, Afiniti, Amazon Web Services Amazon Connect, and Cisco Webex Contact Center. Deployments: multi-tenant and overlay. Sources checked 2026-10-02. Afiniti describes Afiniti Pairing as an overlay on existing routing infrastructure[^2]. Apart from Afiniti's overlay description, the deployment model of Genesys Cloud, Amazon Connect and Webex Contact Center is not covered by the sources below. The behavioral pairing method described here is from patent US9300802B1[^7]. Amazon Connect routing-step and proficiency reporting applies from 17 August 2026[^14]. Webex Contact Center Personalized AI Routing applies from the what's new entry dated 1 October 2026[^53]. Webex Contact Center Weighted Agent Insights routing applies from 25 June 2026[^60]. ## What remains uncertain Webex Contact Center AI routing for live traffic requires a specific entitlement[^55]; the SKU and price of that entitlement are not covered by the sources below. Genesys Cloud predictive routing is consumed through AI Experience tokens[^20]; the number of tokens consumed per predictively routed interaction is not covered by the sources below. Webex AI routing regional availability and the locality of its model training data are not covered by the sources below. Webex AI routing trains on a selected business outcome[^52]; the enumerated list of selectable business outcomes is not covered by the sources below. Amazon Connect documentation describes targeting an agent predicted by the customer's own custom machine learning model[^9]; whether Amazon Connect has any native ML agent-matching feature is not covered by the sources below. AI, behavioural routing and AI agent-matching capabilities of contact centre platforms other than those named in this article are not covered by the sources below. The current application dates for EU AI Act high-risk obligations following the digital omnibus are not covered by the sources below. ## Sources [^1]: Afiniti says Afiniti Pairing impact is measured with continuous benchmarking in which a live control group runs in parallel with AI pairing. Source: [Afiniti Pairing](https://www.afiniti.com/products/afiniti-pairing/), product page measurement section. Checked 2026-10-02. [^2]: Afiniti describes Afiniti Pairing as an overlay on existing routing infrastructure that works within the existing routing framework rather than replacing the ACD. Source: [Afiniti Pairing](https://www.afiniti.com/products/afiniti-pairing/), product page integration section. Checked 2026-10-02. [^3]: Routing that scores agents on their individual historical performance to allocate interactions to them plausibly falls within the task-allocation wording of EU AI Act Annex III point 4(b), so an EU deployer should assess it as potentially high-risk (inferred). Source: [Regulation (EU) 2024/1689 of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act)](https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202401689), Annex III, point 4(b); read with Article 6(3) derogation. Checked 2026-10-02. [^4]: Afiniti's benchmarking patent describes cycling among pairing strategies such as FIFO and behavioral pairing, alternating more often than once per day or even once per hour. Source: [US10827073B2 - Techniques for benchmarking pairing strategies in a contact center system](https://patents.google.com/patent/US10827073), detailed description, benchmarking module 140. Checked 2026-10-02. [^5]: The benchmarking patent gives both 50/50 and unequal 80/20 behavioral-pairing-to-FIFO allocations as examples of on/off cycling. Source: [US10827073B2 - Techniques for benchmarking pairing strategies in a contact center system](https://patents.google.com/patent/US10827073), detailed description, cycle allocation examples. Checked 2026-10-02. [^6]: The benchmarking patent corrects combined on/off results for the Yule-Simpson effect, because unequal contact distributions across periods can reverse an aggregated comparison. Source: [US10827073B2 - Techniques for benchmarking pairing strategies in a contact center system](https://patents.google.com/patent/US10827073), detailed description, Yule-Simpson worked example. Checked 2026-10-02. [^7]: The US9300802B1 behavioral pairing method orders contacts and agents, for example as percentiles, and selects pairs by comparing differences in ordering between candidate contact-agent pairs. Source: [US9300802B1 - Techniques for behavioral pairing in a contact center system](https://patents.google.com/patent/US9300802B1/en), abstract. Checked 2026-10-02. [^8]: Afiniti's behavioral pairing patent distinguishes behavioral pairing from performance-based routing: PBR tries to maximise each interaction's expected outcome, while behavioral pairing accounts for subsequent pairings to balance agent utilisation and improve overall performance. Source: [US9300802B1 - Techniques for behavioral pairing in a contact center system](https://patents.google.com/patent/US9300802B1/en), background and summary describing FIFO and PBR. Checked 2026-10-02. [^9]: Amazon Connect documentation describes a routing step that targets up to 10 preferred agents by user ID, including an agent predicted as best fit by the customer's own custom machine learning model, followed by attribute-based steps. Source: [Flow block in Connect Customer: Set routing criteria](https://docs.aws.amazon.com/connect/latest/adminguide/set-routing-criteria.html), Use routing criteria to target a specific preferred agent, FAQ entries. Checked 2026-10-02. [^10]: Amazon Connect routing criteria are a sequence of routing steps, each a set of requirements built from predefined attribute, value, comparison operator and proficiency level, which an agent must meet to be joined to the contact. Source: [Flow block in Connect Customer: Set routing criteria](https://docs.aws.amazon.com/connect/latest/adminguide/set-routing-criteria.html), Description and How routing criteria works sections. Checked 2026-10-02. [^11]: Amazon Connect's documented route to ML-driven agent matching is a customer-built model feeding routing criteria, and no native predictive routing model comparable to Genesys or Webex was found in the pages reviewed (inferred). Source: [Flow block in Connect Customer: Set routing criteria](https://docs.aws.amazon.com/connect/latest/adminguide/set-routing-criteria.html), preferred agent FAQ, custom machine learning model example. Checked 2026-10-02. [^12]: An Amazon Connect contact restricted to a preferred agent stays restricted until that routing step expires, even if the agent is offline, busy, in a nonproductive status or deleted. Source: [Flow block in Connect Customer: Set routing criteria](https://docs.aws.amazon.com/connect/latest/adminguide/set-routing-criteria.html), FAQ, If the preferred agent is not available, what happens. Checked 2026-10-02. [^13]: Amazon Connect routing steps allow up to eight AND attributes, up to three OR conditions, and proficiency levels compared with >= or as a range from 1 to 5. Source: [Flow block in Connect Customer: Set routing criteria](https://docs.aws.amazon.com/connect/latest/adminguide/set-routing-criteria.html), How routing criteria works, items you can use. Checked 2026-10-02. [^14]: Since 17 August 2026 Amazon Connect dashboards can filter agents by assigned proficiencies, group metrics by routing step and report contacts queued in a routing step. Source: [Amazon Connect Customer dashboards now support reporting on routing steps and agent proficiencies](https://aws.amazon.com/about-aws/whats-new/2026/08/amazon-connect-routing-steps/), announcement body. Checked 2026-10-02. [^15]: When every Amazon Connect routing step has expired, the contact is offered to the longest available agent who has the queue in their routing profile. Source: [Flow block in Connect Customer: Set routing criteria](https://docs.aws.amazon.com/connect/latest/adminguide/set-routing-criteria.html), How routing criteria works, step 3. Checked 2026-10-02. [^16]: EU AI Act Annex III point 4(b) lists as high-risk AI systems intended to allocate tasks based on individual behaviour or personal traits or characteristics, or to monitor and evaluate the performance and behaviour of persons in work-related relationships. Source: [Regulation (EU) 2024/1689 of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act)](https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202401689), Annex III, point 4(b). Checked 2026-10-02. [^17]: Genesys states that no data is retained in predictive routing models for more than 90 days. Source: [Use of AI in predictive routing](https://help.genesys.cloud/articles/use-of-ai-in-predictive-routing/), section on model training and data retention. Checked 2026-10-02. [^18]: When Genesys Cloud predictive routing has an agent surplus, it ranks agents by combining their time since last interaction with their predictive score. Source: [Agent selection process](https://help.genesys.cloud/articles/agent-selection-process/), agent surplus section. Checked 2026-10-02. [^19]: Average handle time and next contact avoidance are generally available KPIs for Genesys Cloud predictive routing. Source: [Predictive routing overview](https://help.genesys.cloud/articles/predictive-routing-overview/), supported KPIs list. Checked 2026-10-02. [^20]: Genesys Cloud predictive routing is billed per interaction routed predictively and is consumed through AI Experience tokens. Source: [Predictive routing pricing and billing](https://help.genesys.cloud/articles/predictive-routing-pricing-and-billing/), opening paragraph. Checked 2026-10-02. [^21]: Genesys Cloud predictive routing supports voice, email and message interactions, where messages include third-party messaging platforms, inbound SMS, Genesys Cloud web messaging and open messaging. Source: [About predictive routing](https://help.genesys.cloud/articles/about-predictive-routing/), overview paragraph on supported interaction types. Checked 2026-10-02. [^22]: A Genesys Cloud predictive routing comparison test compares predictive routing against either standard routing or bullseye routing. Source: [Predictive routing comparison test](https://help.genesys.cloud/articles/test-predictive-routing-for-your-queues/), steps to start a comparison test, comparison routing method selection. Checked 2026-10-02. [^23]: During a Genesys Cloud predictive routing comparison test, interactions are routed in an hour-on, hour-off pattern: predictive routing 50 percent of the time and the comparison method the other 50 percent. Source: [Predictive routing comparison test](https://help.genesys.cloud/articles/test-predictive-routing-for-your-queues/), opening section of comparison test article. Checked 2026-10-02. [^24]: Genesys recommends running the predictive routing comparison test for at least two weeks, noting that high-volume queues may produce results sooner. Source: [Predictive routing comparison test](https://help.genesys.cloud/articles/test-predictive-routing-for-your-queues/), test duration guidance. Checked 2026-10-02. [^25]: Custom KPI outcomes for Genesys predictive routing are supplied through the External Attribution API, with uploads recommended daily or at minimum weekly, and each record must be unique per KPI ID, conversation ID and agent ID. Source: [Custom KPI for predictive routing](https://help.genesys.cloud/articles/custom-kpi-for-predictive-routing/), outcome data upload section. Checked 2026-10-02. [^26]: Genesys recommends at least 90 days and ideally 180 days of data for predictive routing. Source: [Sources of data for predictive routing decisions](https://help.genesys.cloud/articles/sources-of-data-for-predictive-routing-decisions/), Considerations heading. Checked 2026-10-02. [^27]: Genesys Cloud predictive routing uses a data source only if its volume meets a minimum threshold; the thresholds are set at organisation level, are not editable by customers and are updated by Genesys, and their values are not published. Source: [Data requirements for predictive routing](https://help.mypurecloud.com/articles/data-requirements-for-predictive-routing/), note on threshold values. Checked 2026-10-02. [^28]: When Genesys Cloud predictive routing has an interaction surplus, it ranks interactions by combining waiting time with predictive score and assigns the available agent to the highest-ranked interaction. Source: [Agent selection process](https://help.genesys.cloud/articles/agent-selection-process/), interaction surplus section. Checked 2026-10-02. [^29]: Genesys predictive routing KPI processing to build the model can take several hours, and when activating predictive routing all interactions use standard routing until processing completes. Source: [Predictive routing during KPI processing phase](https://help.genesys.cloud/articles/predictive-routing-during-kpi-processing-phase/), routing during processing, activation scenario. Checked 2026-10-02. [^30]: Genesys Cloud ongoing value monitoring routes interactions with predictive routing 80 percent of the time interval and with the baseline routing method 20 percent of the time interval. Source: [Predictive routing ongoing value monitoring](https://help.genesys.cloud/articles/monitor-ongoing-value-of-predictive-routing/), opening section. Checked 2026-10-02. [^31]: Genesys ongoing value monitoring repeats a six-day cycle that alternates blocks of 8 hours of predictive routing with 2 hours of the baseline routing method. Source: [Predictive routing ongoing value monitoring](https://help.genesys.cloud/articles/monitor-ongoing-value-of-predictive-routing/), section describing the monitoring cycle pattern. Checked 2026-10-02. [^32]: Genesys requires at least a two-week 50-50 comparison test that showed a desirable benefit before ongoing value monitoring is set up on a queue. Source: [Predictive routing ongoing value monitoring](https://help.genesys.cloud/articles/monitor-ongoing-value-of-predictive-routing/), prerequisites paragraph. Checked 2026-10-02. [^33]: Genesys recommends running ongoing value monitoring through six cycles of six days before treating the comparison as valid, and says accuracy can take up to two or three months. Source: [Predictive routing ongoing value monitoring](https://help.genesys.cloud/articles/monitor-ongoing-value-of-predictive-routing/), guidance on valid comparison duration. Checked 2026-10-02. [^34]: Genesys states that predictive routing does not use data models that require data such as gender and nationality. Source: [Use of AI in predictive routing](https://help.genesys.cloud/articles/use-of-ai-in-predictive-routing/), section on bias. Checked 2026-10-02. [^35]: Genesys states that predictive routing does not use personally identifiable information in agent scoring and generates features only from transactional conversation data without PII. Source: [Use of AI in predictive routing](https://help.genesys.cloud/articles/use-of-ai-in-predictive-routing/), section on data used. Checked 2026-10-02. [^36]: In Genesys Cloud predictive routing each queue optimises only one KPI at a time, and different queues can use different KPIs. Source: [Predictive routing overview](https://help.genesys.cloud/articles/predictive-routing-overview/), section on supported KPIs. Checked 2026-10-02. [^37]: Genesys predictive routing considers agent profile data only if it is populated for more than 50 percent of the agents working on queues with predictive routing enabled. Source: [Data requirements for predictive routing](https://help.mypurecloud.com/articles/data-requirements-for-predictive-routing/), agent profile data section. Checked 2026-10-02. [^38]: The agent profile fields Genesys lists for predictive routing are hire date, department, certifications, employee type, user skill and user language. Source: [Data requirements for predictive routing](https://help.mypurecloud.com/articles/data-requirements-for-predictive-routing/), agent profile data field list. Checked 2026-10-02. [^39]: Genesys Cloud predictive routing uses machine learning to rank each agent in the target agent pool by how well that agent is predicted to handle a specific interaction against a chosen KPI. Source: [Predictive routing overview](https://help.genesys.cloud/articles/predictive-routing-overview/), opening section, What predictive routing does. Checked 2026-10-02. [^40]: If a Genesys predictive routing model retrain fails, standard routing is used for the affected media types while predictive routing continues for unaffected media types. Source: [Predictive routing during KPI processing phase](https://help.genesys.cloud/articles/predictive-routing-during-kpi-processing-phase/), section on completion and retraining. Checked 2026-10-02. [^41]: Before scoring, Genesys Cloud predictive routing filters agents for required language skills and, when skill matching is enabled, non-language ACD skills. Source: [Agent selection process](https://help.genesys.cloud/articles/agent-selection-process/), agent selection procedure, filtering step. Checked 2026-10-02. [^42]: Genesys documentation acknowledges that predictive routing, by learning from the most recent historical conversations, may introduce temporal bias. Source: [Use of AI in predictive routing](https://help.genesys.cloud/articles/use-of-ai-in-predictive-routing/), section on bias. Checked 2026-10-02. [^43]: If three or fewer agents are available on a queue, Genesys Cloud does not score agents and routes with the standard routing method. Source: [Agent selection process](https://help.genesys.cloud/articles/agent-selection-process/), agent selection procedure. Checked 2026-10-02. [^44]: Genesys describes three phases for adopting predictive routing: benefit assessment, comparison testing, and then either ongoing value monitoring or full activation. Source: [Predictive routing overview](https://help.genesys.cloud/articles/predictive-routing-overview/), section describing the phases of predictive routing. Checked 2026-10-02. [^45]: During the predictive routing timeout Genesys Cloud first offers to the highest-scoring agents and gradually widens to lower-scoring agents, and on expiry routes with standard routing. Source: [Configure timeout for predictive routing](https://help.genesys.cloud/articles/configure-timeout-for-predictive-routing/), how the timeout works paragraph. Checked 2026-10-02. [^46]: The Genesys Cloud predictive routing timeout must be an integer between 12 and 259,200 seconds. Source: [Configure timeout for predictive routing](https://help.genesys.cloud/articles/configure-timeout-for-predictive-routing/), timeout value paragraph. Checked 2026-10-02. [^47]: Genesys Cloud shows the top 10 features influencing predictive routing decisions per queue and per media type on the queue's Predictive Model tab. Source: [View features that influenced predictive routing decisions](https://help.genesys.cloud/articles/view-features-that-influenced-predictive-routing-decisions/), procedure section, Predictive Model tab. Checked 2026-10-02. [^48]: Genesys Cloud updates predictive routing scoring features with daily data and retrains the models weekly. Source: [Use of AI in predictive routing](https://help.genesys.cloud/articles/use-of-ai-in-predictive-routing/), section on model training. Checked 2026-10-02. [^49]: Genesys predictive routing uses white-box models in which each input feature is given a percentage or score representing its importance. Source: [Use of AI in predictive routing](https://help.genesys.cloud/articles/use-of-ai-in-predictive-routing/), section on model type. Checked 2026-10-02. [^50]: Genesys Cloud defines Standard ACD routing as routing interactions to the next available agent while considering skills according to the evaluation method, which is the skills-based baseline predictive routing is compared against. Source: [Routing and evaluation methods](https://help.genesys.cloud/articles/acd-evaluation-routing-methods/), Routing methods list, Standard. Checked 2026-10-02. [^51]: Webex Contact Center AI routing reports for evaluation mode and AI routing mode include only interactions longer than 30 seconds. Source: [Set up personalized AI routing](https://help.webex.com/en-us/article/nw4k40w), Limits and considerations. Checked 2026-10-02. [^52]: When evaluation mode starts, Webex Contact Center trains the AI routing model on up to 90 days of historical interaction data for the selected business outcome. Source: [Set up personalized AI routing](https://help.webex.com/en-us/article/nw4k40w), Evaluate AI routing. Checked 2026-10-02. [^53]: Cisco announced Personalized AI Routing as available for Webex Contact Center in the what's new entry dated 1 October 2026. Source: [What's new for administrators in Webex Contact Center](https://help.webex.com/en-us/article/a1gx3h/What's-New-in-Webex-Contact-Center), entry Personalized AI Routing, October 01, 2026. Checked 2026-10-02. [^54]: With Webex Contact Center AI routing active, existing queue, skill, priority, flow, availability and capacity rules still determine which agents are eligible. Source: [Set up personalized AI routing](https://help.webex.com/en-us/article/nw4k40w), Activate AI routing. Checked 2026-10-02. [^55]: Using Webex Contact Center AI routing for live traffic requires the organisation to hold a specific entitlement, in addition to AI routing being enabled for the organisation. Source: [Set up personalized AI routing](https://help.webex.com/en-us/article/nw4k40w), Before you begin. Checked 2026-10-02. [^56]: In Webex Contact Center AI routing evaluation mode the model predicts the recommended agent and expected outcome while regular routing continues to handle live interactions. Source: [Set up personalized AI routing](https://help.webex.com/en-us/article/nw4k40w), Evaluate AI routing. Checked 2026-10-02. [^57]: Webex Contact Center AI routing monitoring compares the selected KPI average under AI routing and under regular routing and reports percentage and absolute improvement. Source: [Set up personalized AI routing](https://help.webex.com/en-us/article/nw4k40w), Monitor AI routing. Checked 2026-10-02. [^58]: The Webex Contact Center AI routing optimisation check returns High potential, Low potential or Failed for each queue analysed. Source: [Set up personalized AI routing](https://help.webex.com/en-us/article/nw4k40w), Analyze queues for AI routing. Checked 2026-10-02. [^59]: Webex Contact Center allows at most 100 queues per organisation to use AI routing, at most 100 queues in evaluation mode at once, and 50 queue IDs per optimisation check request. Source: [Set up personalized AI routing](https://help.webex.com/en-us/article/nw4k40w), Limits and considerations. Checked 2026-10-02. [^60]: Since 25 June 2026 Webex Contact Center Weighted Agent Insights routing lets administrators assign Dynamic Skills directly to agents outside skill profiles and apply weighted requirements at queue or contact-type level. Source: [What's new for administrators in Webex Contact Center](https://help.webex.com/en-us/article/a1gx3h/What's-New-in-Webex-Contact-Center), entry Weighted Agent Insights routing, June 25, 2026. Checked 2026-10-02.