CX-16 · Digital Self-Service & Chatbot Management

📊

Process Flow Diagram (BPMN)

📋

L4 Process Steps

Step	Step Name	Role / Swim Lane	System	Input	Output	KPI	Dec?	Exc?
Phase 1 1.1	Define digital channel scope & use-case catalogue	Digital Product Owner	Confluence	Contact centre volume reports, top-10 call driver analysis from Genesys Cloud CX	Approved use-case catalogue listing automatable intents (flight status, rebooking, refund, baggage, check-in)	Use-case coverage ≥70% of top contact drivers	N	N
1.2	Configure NLU intents, entities & training utterances	Conversational AI Engineer	Google Contact Center AI (CCAI)	Use-case catalogue, historical chat transcripts from Salesforce Service Cloud	NLU model with ≥50 utterances per intent, entity extraction rules for PNR, flight number, date, airport code	Intent recognition accuracy ≥88% in offline evaluation	N	N
1.3	Set containment thresholds & escalation routing rules	Contact Centre Operations Manager	Genesys Cloud CX	NLU confidence score ranges, queue capacity SLAs, skills-based routing matrix	Escalation policy: confidence <0.70 → clarification prompt; <0.55 → live agent; authenticated transaction failure → priority queue	Mis-escalation rate <5% (contacts escalated that chatbot could have resolved)	N	N
1.4	UAT regression test chatbot flows & approve release	QA Analyst	Google Contact Center AI (CCAI)	Configured NLU model, test script covering all use-case catalogue scenarios including exception paths	UAT sign-off report; failed flows returned to step 1.2 for remediation	UAT pass rate ≥95% across all intent test cases before prod deployment	Y	N
Phase 2 2.1	Receive & parse inbound digital interaction	Automated — Chatbot Runtime	Genesys Cloud CX	Customer message via web chat, mobile app, WhatsApp Business API, or Facebook Messenger	Normalised interaction record with channel source, session ID, timestamp, and raw utterance	Channel availability ≥99.9% uptime; median response latency <800 ms	N	N
2.2	Classify intent & extract booking / loyalty entities	Automated — NLU Engine	Google Contact Center AI (CCAI)	Normalised customer utterance	Top intent label with confidence score, extracted entities (PNR, flight number, date, loyalty tier)	Live intent accuracy ≥85%; confidence score logged for every interaction	Y	N
2.3	Authenticate customer for transactional self-service	Automated — Identity Verification	Amadeus Altéa PSS	Customer-supplied PNR + last name, or loyalty member ID + PIN via Salesforce Service Cloud identity service	Authenticated session token; customer profile (booking history, loyalty tier, contact preferences) retrieved	Authentication success rate ≥92% on first attempt; failed auth rate <3%	Y	Y
2.4	Execute self-service task via PSS & loyalty APIs	Automated — Transaction Engine	Amadeus Altéa PSS	Authenticated session, validated intent (rebooking, refund request, seat selection, check-in, baggage add-on, flight status)	Transaction confirmation (new PNR, refund reference, boarding pass, baggage receipt) or structured error response	Self-service transaction completion rate ≥80% for eligible intents; API error rate <1%	N	Y
Phase 3 3.1	Detect escalation trigger & package conversation context	Automated — Orchestration Layer	Genesys Cloud CX	CCAI confidence score below threshold, customer explicit escalation request ("speak to agent"), or transaction API failure flag	Escalation event with full conversation transcript, intent history, authentication status, and customer tier	Context transfer completeness ≥98% (no dropped transcripts); escalation routing latency <5 s	Y	N
3.2	Route escalation to live agent by skill & priority	Automated — ACD Router	Genesys Cloud CX	Escalation event, agent skill matrix (language, fare complexity, loyalty tier authorisation), real-time queue depth	Agent assignment with conversation context pre-loaded in Salesforce Service Cloud agent console	Average speed to answer (ASA) for escalated contacts <90 s; first-contact resolution ≥75%	Y	N
3.3	Agent reviews transcript & resumes resolution	Contact Centre Agent	Salesforce Service Cloud	Full chatbot transcript, customer profile (PNR, loyalty tier, prior cases), CCAI-suggested resolution from agent assist	Case updated in Salesforce; resolution actioned in Amadeus Altéa PSS (rebooking, waiver, compensation)	Agent handle time for escalated digital contacts <8 min; re-open rate <5%	N	N
Phase 4 4.1	Ingest low-confidence & failed interaction logs	Conversational AI Analyst	AWS S3 / Redshift	Daily CCAI interaction logs exported via Genesys Data Actions; confidence scores, unmatched utterances, fallback counts	Curated dataset of misclassified and unresolved intents ranked by volume and customer impact	Log ingestion SLA <24 h post-interaction; coverage 100% of chatbot sessions	N	N
4.2	Update training utterances & FAQ knowledge content	Knowledge Manager	Salesforce Knowledge	Prioritised intent gap list, agent feedback on frequently asked questions, product/policy change notifications	Updated intent training dataset uploaded to CCAI; revised FAQ articles published in Salesforce Knowledge	Knowledge article freshness: 100% reviewed within 30 days of policy change; new utterances ≥20 per reclassified intent	N	N
4.3	Retrain, validate & promote NLU model to production	ML Engineer	Google Contact Center AI (CCAI)	Updated training dataset, baseline model accuracy benchmarks from last release	Validated NLU model version; A/B test report comparing new vs. current model on held-out test set	New model accuracy ≥baseline +2%; no regression on existing intents; staging validation pass rate 100%	Y	N
Phase 5 5.1	Monitor containment rate, CSAT & escalation KPIs	Digital Channel Analyst	Tableau	Genesys Cloud CX interaction data, Medallia post-interaction CSAT survey results, AWS Redshift chatbot analytics	Weekly digital channel performance dashboard: containment rate, CSAT, escalation rate, top unresolved intents	Chatbot containment rate ≥65%; digital CSAT ≥4.0/5.0; escalation rate <35%	Y	N
5.2	Identify intent gaps & journey abandonment patterns	Contact Centre Analytics Manager	AWS S3 / Redshift	Tableau dashboard KPI alerts, session path analysis, drop-off rates by intent and step	Root-cause analysis report identifying top 5 intent gaps and highest-abandonment journey points	Journey abandonment rate <15% per intent flow; report produced within 5 business days of KPI breach	N	N
5.3	Prioritise & schedule chatbot enhancement backlog	Digital Product Owner	Confluence	Root-cause analysis report, capacity estimates from Conversational AI Engineering team, commercial priority from Revenue Management	Prioritised enhancement backlog in Confluence; sprint plan for next NLU model training cycle (step 1.2)	Backlog groomed and sprint-planned within 10 business days of KPI breach; enhancement cycle ≤6 weeks end-to-end	N	N

📋

Process Attributes

Identification

Process IDCX-16

L1 DomainCustomer Experience & Loyalty

L2 ProcessCustomer Service Contact Centre

L3 NameDigital Self-Service & Chatbot Management

L4 Steps17 across 5 phases

Decision Gates7 (all with iteration loops)

Exceptions2 documented

Swim Lanes (Roles)

Digital Product Owner

Conversational AI Engineer

Contact Centre Operations Manager

QA Analyst

Automated — Chatbot Runtime

Automated — NLU Engine

Automated — Identity Verification

Automated — Transaction Engine

Automated — Orchestration Layer

Automated — ACD Router

Contact Centre Agent

Conversational AI Analyst

Knowledge Manager

ML Engineer

Digital Channel Analyst

Contact Centre Analytics Manager

Systems & Tools

ConfluenceGoogle Contact Center AI (CCAI)Genesys Cloud CXAmadeus Altéa PSSSalesforce Service CloudAWS S3 / RedshiftSalesforce KnowledgeTableau

Key Performance Indicators

Define digital channel scope & use-case catalogueUse-case coverage ≥70% of top contact drivers

Configure NLU intents, entities & training utterancesIntent recognition accuracy ≥88% in offline evaluation

Set containment thresholds & escalation routing rulesMis-escalation rate <5% (contacts escalated that chatbot could have resolved)

UAT regression test chatbot flows & approve releaseUAT pass rate ≥95% across all intent test cases before prod deployment

Receive & parse inbound digital interactionChannel availability ≥99.9% uptime; median response latency <800 ms

Classify intent & extract booking / loyalty entitiesLive intent accuracy ≥85%; confidence score logged for every interaction

Authenticate customer for transactional self-serviceAuthentication success rate ≥92% on first attempt; failed auth rate <3%

Execute self-service task via PSS & loyalty APIsSelf-service transaction completion rate ≥80% for eligible intents; API error rate <1%

Airline-Specific Risks & Pain Points

Airlines frequently undercount automatable intents by excluding irregular operations (IRROP) scenarios, leaving high-volume disruption contacts undeflected

Airline-specific entities (IATA airport codes, fare basis codes, PNR formats) require custom entity extractors not available in generic NLU libraries; training data is sparse for low-frequency intents

Overly aggressive confidence thresholds during IRROP events flood live agent queues; Genesys routing rules require manual override during declared irregular operations

Regression testing is manual and time-consuming; no automated test harness exists for end-to-end chatbot + PSS API integration paths, increasing release cycle to 3–4 weeks

WhatsApp Business API throughput is capped by Meta rate limits during high-demand periods (e.g., weather events); fallback to web chat is not seamlessly communicated to passengers

Code-switching (multilingual passengers mixing English with Spanish or Portuguese) degrades CCAI intent accuracy by 10–15%; no per-language model version management currently in place

Inputs / Outputs

Primary InputContact centre volume reports, top-10 call driver analysis from Genesys Cloud CX

Primary OutputPrioritised enhancement backlog in Confluence; sprint plan for next NLU model training cycle (step 1.2)

←PreviousCX-15 · Customer Service Contact Centre Operations NextFO-01 · Flight Dispatch & Release→