KPI dashboard · seed 20260929

Funnel, first response, and resolution

4000 synthetic journeys and 6376 events. A journey is one customer, one intent, and one or more contacts. Digital channels are ussd, app, web, and social. The assisted channel is call.

Journey resolution

81.4%3256/4000

Digital-first journeys that later reach a call

32.0%971/3035

Funnel

Each bar is a committed count divided by its own denominator. The two digital-to-call rows share a numerator and use different denominators.

Journey resolution81.4%

3256/4000 · Journeys whose end reason is resolved.

Resolved on the first contact55.6%

2223/4000

Repeat contact36.5%

1462/4000 · Two or more contacts in the same journey.

Channel switch30.9%

1235/4000 · More than one distinct channel. A second contact on the same channel is not a switch.

Digital-first75.9%

3035/4000 · First contact on ussd, app, web, or social.

Digital-to-call, all journeys24.3%

971/4000 · Numerator is the 971 journeys that start digital and later reach a call.

Digital-to-call, digital-first journeys32.0%

971/3035 · Same 971 journeys, divided by the 3035 digital-first journeys.

Abandoned18.6%

744/4000 · Of the abandoned journeys, chose-abandon ends: 542, and max-contact ends: 202.

First channel to journey outcome

The Sankey splits the same 4000 journeys into resolved on the first contact, resolved after another contact, or abandoned.

Resolved on first contact Resolved after another contact Abandoned

ussd

n=1289

app

n=873

web

n=548

social

n=325

call

n=965

Bar length follows the committed Sankey counts. Exact counts are in the table.

Committed Sankey counts by first channel. Source: reports/kpis.json.
First channel	Resolved on first contact	Resolved after another contact	Abandoned	Journeys
ussd	616	388	285	1289
app	502	225	146	873
web	276	151	121	548
social	75	152	98	325
call	754	117	94	965

SYNTHETIC Sankey chart. First channel, ussd, app, web, social, or call, flows to resolved on first contact, resolved after another contact, or abandoned. — Static copy of the committed Sankey. SYNTHETIC data.Interactive HTML on GitHub

Time to first response

Delay, in minutes, between a journey's first inbound timestamp and the response on that same first contact. Every synthetic contact has a response. Median and the 90th percentile use DuckDB quantile_cont (linear), shown to one decimal as in reports/headlines.md.

Median 90th percentile

ussdn=1289

median 1.0 min · p90 1.7 min

appn=873

median 2.0 min · p90 4.0 min

webn=548

median 4.8 min · p90 10.5 min

socialn=325

median 30.0 min · p90 69.9 min

calln=965

median 5.9 min · p90 11.2 min

Scale ends at the longest 90th percentile in this table (social, 69.9 minutes).

Synthetic time-to-first-response by first channel.
First channel	Journeys	Median minutes	90th percentile minutes
ussd	1289	1.0	1.7
app	873	2.0	4.0
web	548	4.8	10.5
social	325	30.0	69.9
call	965	5.9	11.2

Contact resolution by channel

Resolved contacts divided by contacts on that channel. A journey has at most one resolved contact, the one that closes it.

ussd43.4%

699/1611

app50.8%

582/1145

web43.9%

317/722

social19.4%

102/526

call65.6%

1556/2372

SYNTHETIC bar chart of contact resolution by channel: ussd 43.4 percent (699 of 1611), app 50.8 percent (582 of 1145), web 43.9 percent (317 of 722), social 19.4 percent (102 of 526), call 65.6 percent (1556 of 2372). — Static copy of the committed channel chart. SYNTHETIC data.

Resolution by intent

Complaint journeys move on to the call channel more often than lookup journeys. The rates below are the group totals printed in reports/headlines.md.

self-service candidate

Journey resolution94.7%

1401/1480

Digital-to-call among digital-first12.9%

144/1112

assisted

Journey resolution83.0%

1276/1538

Digital-to-call among digital-first36.6%

434/1185

complaint

Journey resolution59.0%

579/982

Digital-to-call among digital-first53.3%

393/738

SYNTHETIC grouped bars. Self-service candidate: journey resolution 94.7 percent (1401 of 1480) and digital-to-call 12.9 percent (144 of 1112). Assisted: 83.0 percent (1276 of 1538) and 36.6 percent (434 of 1185). Complaint: 59.0 percent (579 of 982) and 53.3 percent (393 of 738). — Static copy of the committed intent-group chart. SYNTHETIC data.

Synthetic results by intent group. Rates match reports/headlines.md.
Intent group	Journeys	Resolved	Journey resolution	Resolved on first contact	Digital-first	Digital-to-call	Digital-to-call among digital-first
self-service candidate	1480	1401	94.7% (1401/1480)	1183	1112	144	12.9% (144/1112)
assisted	1538	1276	83.0% (1276/1538)	778	1185	434	36.6% (434/1185)
complaint	982	579	59.0% (579/982)	262	738	393	53.3% (393/738)

Contact resolution by intent group and channel

The base probability is the config input. The realised rate is the output. Realised rates sit near the base inputs because of sampling and the two adjustments for later attempts and slow responses.

Base probability and realised contact resolution. Source: reports/headlines.md.
Intent group	Channel	Config base	Contacts	Resolved contacts	Realised rate
self-service candidate	ussd	0.82	540	420	77.8% (420/540)
self-service candidate	app	0.86	351	300	85.5% (300/351)
self-service candidate	web	0.74	212	160	75.5% (160/212)
self-service candidate	social	0.30	139	48	34.5% (48/139)
self-service candidate	call	0.90	554	473	85.4% (473/554)
assisted	ussd	0.38	607	226	37.2% (226/607)
assisted	app	0.52	464	227	48.9% (227/464)
assisted	web	0.45	291	128	44.0% (128/291)
assisted	social	0.20	203	34	16.7% (34/203)
assisted	call	0.78	959	661	68.9% (661/959)
complaint	ussd	0.12	464	53	11.4% (53/464)
complaint	app	0.22	330	55	16.7% (55/330)
complaint	web	0.18	219	29	13.2% (29/219)
complaint	social	0.15	184	20	10.9% (20/184)
complaint	call	0.60	859	422	49.1% (422/859)

Counts by intent

reports/headlines.md prints rates for the three groups, not a rate for every intent. This table stays in counts from reports/kpis.json.

Synthetic journey counts by intent. No rate is added here.
Category	Intent	Group	Journeys	Resolved journeys	Digital-first	Digital-to-call
BILLING	dispute_invoice	complaint	212	125	158	78
BILLING	invoices	self-service candidate	317	294	230	31
COMPLAINTS	get_compensation	complaint	119	64	88	41
COMPLAINTS	report_poor_signal_coverage	complaint	220	134	163	89
COMPLAINTS	report_problem	complaint	289	173	218	115
CONSUMPTION	check_excess_data_charges	self-service candidate	186	179	142	24
CONSUMPTION	check_usage	self-service candidate	467	448	347	42
CONSUMPTION	set_usage_limits	assisted	111	95	90	38
CONTACT	customer_service	assisted	150	117	115	36
CONTACT	human_agent	complaint	142	83	111	70
PAYMENT	check_mobile_payments	self-service candidate	209	194	167	19
PAYMENT	pay	assisted	217	185	166	63
PAYMENT	payment_methods	self-service candidate	91	87	68	8
PAYMENT	schedule_payments	assisted	131	109	108	42
SERVICES	activate_call_management_services	assisted	84	72	65	22
SERVICES	activate_phone	assisted	115	90	91	37
SERVICES	activate_roaming	assisted	123	101	86	37
SERVICES	check_signal_coverage	self-service candidate	141	132	106	17
SERVICES	deactivate_call_management_services	assisted	39	34	31	13
SERVICES	deactivate_phone	assisted	74	63	60	17
SERVICES	install_internet	assisted	82	57	60	21
SUBSCRIPTION	cancel_plan	assisted	123	113	93	30
SUBSCRIPTION	change_plan	assisted	164	130	127	44
SUBSCRIPTION	change_provider	assisted	40	34	32	8
SUBSCRIPTION	check_cancellation_fee	self-service candidate	69	67	52	3
SUBSCRIPTION	sign_up_for_plan	assisted	85	76	61	26

Directly-follows

Among steps that change channel, the largest flow is ussd followed by call (385). A second contact on the same channel counts as a repeat contact and does not count as a channel switch.

call → resolved1556
ussd → resolved699
app → resolved582
ussd → call385
call → call382

All committed directly-follows counts. Rows are the preceding activity.
From	ussd	app	web	social	call	resolved	abandoned
ussd	180	78	40	44	385	699	185
app	42	72	31	62	257	582	99
web	39	55	42	15	160	317	94
social	18	28	20	43	223	102	92
call	43	39	41	37	382	1556	274

Download the committed CSV

Friction heatmap

Realised friction is one minus the contact resolution rate. Cells sit near the base probabilities in the config, moved by the attempt penalty and the slow-response penalty.

Download the committed friction matrix

How to read the rates

Time to first response

Time-to-first-response is the delay, in minutes, between a journey's first inbound timestamp and the response timestamp on that same first contact. Median and the 90th percentile use DuckDB quantile_cont (linear). The synthetic generator draws each delay from the channel lognormal in config/synthetic.yaml and then clips it. Every synthetic contact has a response, so there are no censored non-responses.

Repeat contact

Repeat-contact rate is the share of synthetic journeys with two or more contacts. A repeat contact is a later contact in the same journey after an unresolved contact. The synthetic log has one journey per customer, so this is not a later, separate issue from the same person.

Channel switch

Channel-switch rate is the share of synthetic journeys whose contacts use more than one distinct channel. A second contact on the same channel is a repeat contact and is not a channel switch.

Digital-to-call

A synthetic digital-to-call journey starts on a digital channel (ussd, app, web, or social) and has a later contact on the call channel. That is the operational definition of a failed digital attempt that ends in a call. The rate is reported for all journeys and again among digital-first journeys only.

Resolution rate

On the synthetic log, contact resolution rate for a channel, or for an intent group on a channel, is resolved contacts divided by contacts on that slice. A journey has at most one resolved contact, the one that closes it. Journey resolution rate is journeys whose end reason is resolved, divided by journeys. Realised rates are outputs. They are not the base probabilities in the config: each contact multiplies that base by attempt decay and, when the response is slower than the configured threshold, by the slow-response factor.

Design notes from the recommendations file

These are hypotheses for a care journey. Each one names whether it rests on the public Bitext training set or on the synthetic log. docs/recommendations.md

Lookup intents on USSD and in the app

Self-service contacts resolve at base 0.82, realised 77.8% (420/540) on ussd, and base 0.86, realised 85.5% (300/351) on the app. USSD first-contact time-to-first-response has median 1.0 minutes (n=1289). Digital-to-call in this group: 12.9% (144/1112) of its digital-first journeys.

Keep a person available for complaints

Complaint contacts resolve at base 0.12, realised 11.4% (53/464) on ussd, and base 0.60, realised 49.1% (422/859) on call. Journey resolution is 59.0% (579/982). Digital-to-call among its digital-first journeys is 53.3% (393/738).

Treat social as acknowledgement

The social clock is an input (lognormal mu 3.40). On simulated first contacts the median is 30.0 minutes and the 90th percentile is 69.9 minutes (n=325). The model multiplies resolution probability by 0.75 when a response is slower than 30.0 minutes. Lowest realised contact resolution: social, 19.4% (102/526).

Lookup intents, digital-to-call over journeys

The fraction in docs/recommendations.md divides digital-to-call by journeys for that intent. It does not use the digital-first denominator behind the 32.0% headline.

Self-service intents ordered as in docs/recommendations.md. The rate is digital-to-call over journeys.
Intent	Journeys	Digital-first	Digital-to-call	Digital-to-call / journeys
check_usage	467	347	42	9.0% (42/467)
invoices	317	230	31	9.8% (31/317)
check_excess_data_charges	186	142	24	12.9% (24/186)
check_mobile_payments	209	167	19	9.1% (19/209)
check_signal_coverage	141	106	17	12.1% (17/141)
payment_methods	91	68	8	8.8% (8/91)
check_cancellation_fee	69	52	3	4.3% (3/69)

Channels in the generator, in report order: ussd, app, web, social, call.