Omnichannel care analytics · seed 20260929

32.0%of digital-first journeys later reach a call

32.0% (971/3035). 4000 synthetic journeys, 6376 events, one journey per customer. Journey resolution is 81.4% (3256/4000).

Journey resolution

81.4%3256/4000

Digital-first journeys that later reach a call

32.0%971/3035

Filter those journeys by channel and intent in the live demo. This page keeps the committed totals.

Explore the live demo Read the KPI report

Diagram labelled SYNTHETIC data, seed 20260929. A customer moves from USSD to the app to social to a call. The pipeline scores 4,000 journeys and 6,376 events. The outcome shown is 32.0 percent of digital-first journeys later reaching a call, 971 of 3,035. — Committed story chart. The 32.0% outcome is 971 of 3035 digital-first journeys.

Funnel

Resolution 81.4% (3256/4000). Resolved on the first contact 55.6% (2223/4000).

Time to first response

ussd median 1.0 minutes (n=1289). social median 30.0 minutes (n=325).

Resolution by channel and intent

call 65.6% (1556/2372). social 19.4% (102/526).

Scope

Three inputs, kept apart

Synthetic journeys

Seed 20260929. 4000 journeys. 6376 events. All journey rates on the dashboard come from this log.

Bitext training set

26000 examples, 26 intents, 7 categories, 1000 examples of each intent. Equal counts are how the file was built. They are not demand.

Twitter preview

93 data rows (49 inbound, 44 outbound). Not a telco journey study. The reply delay below is not the synthetic time-to-first-response.

Public · Bitext

Training examples by category

Bitext examples by category. Counts from reports/headlines.md.
Category	Examples	Intents
BILLING	2000	2
COMPLAINTS	3000	3
CONSUMPTION	3000	3
CONTACT	2000	2
PAYMENT	4000	4
SERVICES	7000	7
SUBSCRIPTION	5000	5

Language-tag shares

Shares of the 26000 training examples. Shortest median instruction length: 42.0 characters (pay). Longest: 62.0 characters (check_excess_data_charges).

colloquial (Q)49.5%

12862/26000

errors and typos (Z)49.8%

12942/26000

offensive language (W)35.7%

9271/26000

interrogative structure (I)72.7%

18899/26000

keyword mode (K)0.0%

0/26000

politeness (P)47.3%

12310/26000

abbreviations (E)4.7%

1228/26000

negation (N)0.0%

0/26000

Public · Twitter preview

Not a journey sample

42 inbound-to-company reply pairs sit inside the preview. Median reply delay on those pairs: 50.8 minutes. 90th percentile: 385.7 minutes. Preview tweets from the telecom and cable-ISP allowlist, combined: 4. Do not generalise these delays.

Read the study on GitHub