How is a journey different from an interaction?

An interaction is a single touchpoint on one channel. A journey is the timeline of all interactions for one customer across channels. Resolution can be reached on the journey even when individual interactions look incomplete.

How does FutureAGI evaluate an omnichannel journey?

FutureAGI joins traceAI spans on customer ID, aggregates ConversationResolution and Groundedness across all interactions in a window, and surfaces a journey-resolution score plus cross-channel divergence per intent.

What Is the Omnichannel Customer Journey? FutureAGI (2026)

Q: What is the omnichannel customer journey?

It is the full chronological sequence of touchpoints a single customer has with a brand across every channel — voice, chat, email, SMS, social, in-app — treated as one continuous narrative for routing, evaluation, and reporting.

What Is the Omnichannel Customer Journey?

The omnichannel customer journey is the full chronological sequence of touchpoints a single customer has with a brand across every channel — voice, chat, email, SMS, social, in-app — treated as one narrative thread. A typical journey for a return-and-refund case might be: chatbot triage on the website, then an email with a return label, then a voice call to confirm pickup, then an SMS for refund status. In a 2026 LLM-driven contact center, the AI must reason over prior journey steps to behave consistently in the current step, which makes the journey itself the unit of evaluation.

Why It Matters in Production LLM and Agent Systems

Single-interaction metrics flatter the contact center; journey metrics tell the truth. Each interaction can score a 0.85 on ConversationResolution, but a customer who needed five interactions across three channels to resolve a single case experienced a poor journey. Conversely, a customer who escalated once but then sailed through the rest of the journey had a good outcome — and that nuance is invisible if you only score interactions.

The pain falls across roles. A CX VP reports interaction CSAT trending up while NPS trends down because journey effort is the variable that moves NPS. A product manager shipping a self-service flow does not know whether the new chat path reduces voice-call follow-up — without journey joins, the feedback loop is missing. An ML engineer trains a planner to optimize the current turn and accidentally creates more downstream interactions because the planner does not see the journey ahead. A compliance officer is asked whether the customer’s PII was redacted on every step in the journey and has no joined view to check.

In 2026, journey-level evaluation is the next frontier in CX measurement. Vendors expose interaction analytics; few expose journey analytics. FutureAGI’s customer-ID-keyed trace joins plus journey-resolution aggregator make the journey itself a first-class object engineers can evaluate, alert on, and regress against.

How FutureAGI Handles the Omnichannel Customer Journey

FutureAGI’s approach is to join trace spans by customer_id over a configurable window (default seven days) and aggregate evaluator scores at the journey level. Voice spans come from traceAI-livekit and traceAI-pipecat; chat, email, and social spans come from traceAI-langchain, traceAI-openai, and traceAI-anthropic. Each span carries customer_id, case_id, channel, intent, and a sequence index. ConversationResolution scores each interaction; a derived journey_resolution score uses the last interaction’s resolution if the case is closed, or weighted average if still open. Tone and a CustomEvaluation track persona consistency across the journey.

A concrete example: a retail brand’s CX team wants to know whether their new chatbot reduces voice escalation. They define a journey window of 48 hours from first chatbot contact, join traces by customer_id, and compare two cohorts: customers who used the new chatbot first vs the old one. The new chatbot’s journey-resolution score is 0.74 vs 0.69 for the old one — but voice-escalation rate is identical, which means the journey wins come from chat-internal resolution, not voice avoidance. They run RegressionEval on the journey-level score weekly and treat it as the deploy gate. Interaction-only metrics never would have surfaced this.

How to Measure or Detect It

The journey is measured by aggregating per-interaction signals plus journey-specific derived metrics:

ConversationResolution per interaction: the foundation, joined by customer_id for journey rollups.
Journey-resolution score: a derived metric — last-interaction resolution if case closed, weighted average if open — over a 7-or-14-day window.
Touchpoint count per case: the number of interactions a customer needed to resolve one case; lower is better, controlled for case type.
Channel-switching index: how many channels a single journey spanned; flag journeys that bounced 3+ times.
Tone and persona-consistency: scores whether brand voice was consistent across the journey.
NPS-vs-journey-resolution correlation: ties business outcome to journey quality.

Minimal Python:

from fi.evals import ConversationResolution

resolver = ConversationResolution()
# evaluate each interaction in the journey, then aggregate by customer_id
result = resolver.evaluate(
    input="Customer wants refund on order 8721 (touchpoint 3 of 4)",
    output=conversation_transcript,
)
print(result.score, result.reason)

Common Mistakes

Optimizing turn-level instead of journey-level. A planner that hits its turn metric while creating more touchpoints is making the journey worse.
Hardcoded short journey windows. A 24-hour window misses cases that span a week; tune the window to your case-type distribution.
Ignoring channel-switching cost. Every channel switch adds friction; a 2-touchpoint single-channel journey beats a 2-touchpoint cross-channel journey.
No persona consistency check across the journey. The brand voice should be the same on every step; score it explicitly.
Reporting only closed-journey resolution. Open journeys carry signal too — long open journeys are detractors in the making.