LUCID Workshop Example — Telecom Churn

Open in app ↗

LUCID Discovery WorkshopData Artisans - Roughwork & Insight Shortlist - Print double-sided A4

Page 1 / 2L - U - C ->

NAME

Alex Chen (Example)

DATASET / TOPIC

Telco Churn (Q2 2026)

DATE

21 Jul 2026

TEAM

Retention Analytics

Leakage"Name the chaos honestly"

20 min

Tension statement

e.g. "Revenue is growing - but 1 in 3 customers won't return next month."

We acquired 12,400 new customers in Q2 - but 3,847 existing customers churned, and 68% of them were on contracts with 6+ months remaining. We are growing the top line while silently bleeding the bottom.

The real question this data is asking

What does the dataset want to answer - even if nobody asked it directly?

Why do customers who have been with us for 2-4 years - our supposed "loyal base" - churn at a higher rate than new customers in their first 90 days? The data is asking: what changed after month 24?

Signal vs Symptom

Circle role for each variable: S = Signal Sy = Symptom ? = Unknown

VARIABLE / METRICROLENOTES

Monthly charges

High charges correlate but do not cause

Tenure (months)

U-shape: high churn at 0-3mo AND 24-36mo

Support tickets (last 90d)

3+ unresolved = 4.2x churn probability

Contract type

Month-to-month churn 42% vs annual 11%

Payment failures (last 60d)

2+ failures = strong predictor of churn

What confuses you most about this dataset?

First impression only - do not analyse yet. Just name the confusion.

1. Customers with low monthly charges churn more than those with high charges - opposite of what I assumed.
2. "Satisfaction score" is missing for 34% of churned customers - is the data missing because they already left, or did they leave because we never surveyed them?

What would you NOT say to the CEO right now?

"We need to lower prices to reduce churn." - because our lowest-price tier has the highest churn rate. The problem is not price; it is something that happens between month 18 and month 30.

One signal + one symptom from this dataset

Signal (causes the outcome):

Unresolved support tickets in the 90 days before churn. Customers with 3+ open tickets are 4.2x more likely to churn, and 61% of churned customers had at least 1 unresolved ticket in their final quarter.

Symptom (reflects the outcome):

Month-to-month contract status. 73% of churned customers were on month-to-month contracts, but this is a symptom of their already-low commitment, not the root cause of their departure.

Team opening line - 30 words max

The sentence that opens the entire LUCID story. Read it aloud.

"We are losing 3,847 customers per quarter, and 61% of them had unresolved support tickets in their final 90 days - our service recovery is broken, not our pricing."

Unification"Build the single canvas"

15 min

Outcome variable ( exactly one )

e.g. "Customer churn rate" or "Student pass rate" or "Revenue per store"

Customer churn within 90 days (binary: 1 = churned, 0 = retained)

Canvas name - [Subject] [Decision] Intelligence

e.g. "BrewCo Customer Retention Intelligence" - Name must imply a decision

"TelcoPro Service Recovery Intelligence" - because the decision is not "should we retain customers?" but "which at-risk customers need intervention, and what kind?"

What does this canvas allow a decision-maker to do?

Identify the 15-20% of customers most likely to churn in the next 90 days, ranked by actionable signals (not just probability), so the retention team can prioritise outreach and fix the specific service gap that is driving each customer's risk.

Variable audit

O = Outcome D = Driver C = Context N = Noise - There is exactly ONE Outcome

VARIABLE NAMEROLE ( circle )K / R

Churn_90d (binary)

Support_tickets_unresolved_90d

Payment_failures_60d

Tenure_months

Monthly_charges

Contract_type

Customer_age

Noise variables removed + reason

Variables with no causal path to the outcome - eliminate them

1. Customer_age - No correlation with churn (r=0.03). Age predicts plan choice, not departure.
2. Device_model - Indirect effect only (via tech support needs), already captured in ticket count.
3. Acquisition_channel - Historical artifact; current service experience overrides how they arrived.

Causality"Rank the signals, mute the noise"

25 min

Driver ranking

Rank 1 = strongest influence on outcome. Complete individually before team debrief.

DRIVER NAME%BAR

Unresolved tickets (90d)

38%

Payment failures (60d)

24%

Tenure (24-36mo window)

18%

Monthly charges (high tier)

12%

Contract type (monthly)

Other / residual

Why is rank 1 the primary driver?

This becomes the annotation on your Causality slide

Customers with 3+ unresolved tickets in 90 days have a 67% churn rate vs 12% baseline. The effect is directional and actionable: we can intervene before churn, unlike tenure (which we cannot change) or contract type (which is a symptom).

Noise variables ( muted from the story )

Variables ranked 5th+ or removed - name them and cross them out

Customer_age - no causal path
Device_model - captured by tickets
Acquisition_channel - historical noise
Contract_type - ranked 5th, only 5% explanatory power, symptom not signal

Disputed signal - where did the team disagree?

What data evidence would resolve the disagreement?

Dispute: Is "monthly charges" a driver or just a symptom of plan dissatisfaction?
Resolution: Run a controlled experiment - offer 15% discount to high-charge at-risk customers. If churn drops, it is a driver. If not, it is a symptom of service quality.

Team top 3 ( must agree )

Unresolved support tickets (90d)

Payment failures (60d)

Tenure (24-36mo window)

80% of variance explained by:

Unresolved tickets + Payment failures + Tenure window = 80% of churn variance. These three variables alone separate the "at-risk" population with 84% precision.

Individual rankings

Complete before team debrief

Student A:

Tickets -> Payment -> Tenure

Student B:

Payment -> Tickets -> Charges

Student C:

Tickets -> Tenure -> Payment

LUCID Discovery WorkshopData Artisans - Page 2 - Illumination - Decision - Blog Shortlist

Page 2 / 2<- I - D - Blog

Illumination"One insight. No distractions."

20 min

Candidate insights ( generate at least 3 before choosing )

Format: "[Variable X] means [Outcome Y] is [Zx higher/lower]. This matters because [impact]."

Candidate 1

Customers with 3+ unresolved tickets in 90 days are 4.2x more likely to churn than those with 0-1 tickets. This matters because 61% of all churned customers had at least 1 open ticket in their final quarter - our service recovery is the leak.

Quantified (has a number)?

CEO would act on this alone?

Non-obvious / surprising?

Candidate 2

Customers in the 24-36 month tenure window churn at 31% - higher than new customers (22%) and much higher than 4+ year customers (9%). This matters because we assumed loyalty increases with tenure; the data shows a dangerous mid-life crisis.

Quantified (has a number)?

CEO would act on this alone?

Non-obvious / surprising?

Candidate 3

Customers with 2+ payment failures in 60 days have a 58% churn rate - but 73% of them never contacted support. This matters because we are losing customers silently; the payment failure is a distress signal we are ignoring.

Quantified (has a number)?

CEO would act on this alone?

Non-obvious / surprising?

* Chosen illumination insight ( must pass all three filters above )

Insight headline ( one sentence with a number ):

61% of churned customers had unresolved support tickets in their final 90 days - our service recovery is the leak, not our pricing.

Key number / ratio:

4.2x churn rate with 3+ tickets vs baseline

Chart type: Bar / Line / Scatter / Heat map / Table / Other: ___________

Why does this insight matter? ( business / human impact )

Quantify the impact if possible - revenue, people, risk, time, scale

At $72 ARPU and 3,847 churned customers/quarter, this leak costs $277K/month in lost revenue. If we resolve tickets within 48h for at-risk customers, we project retaining 40% of would-be churners = $111K/month recovered. The intervention (proactive support outreach) costs ~$8K/month.

Dual cover test ( BOTH must be YES before moving to stage D )

Cover the chart - does the headline alone tell the story?YES

Cover the headline - does the chart alone show the insight?YES

Slide annotation ( one sentence pointing at the key data point )

e.g. "Low-frequency customers generate 71% of all churn events - 38% of the base"

"3+ unresolved tickets = 67% churn probability. The cluster in the top-right is your at-risk population - 847 customers, $61K/month at stake."

Chart sketch ( draw or label your illumination slide )

One chart only - Mark the key data point - Add one annotation

Unresolved Tickets (90d) ->

Churn Probability % ->

* 67% churn rate
847 customers
$61K/mo at risk

Baseline 12%

What have you removed from this slide?

Every removed element strengthens the one that remains

Removed: Contract type breakdown, device model legend, acquisition channel color coding, 3D effects, trend lines, secondary Y-axis.

Kept: One scatter plot, one red cluster, one annotation, one baseline. The chart shows where the risk lives; the headline says what to do about it.

Decision"Insight becomes action"

20 min

ACT NOW

Specific enough to action tomorrow - no follow-up question needed

Launch "48-Hour Recovery Protocol" for all customers with 2+ unresolved tickets. Assign a dedicated support agent to each case, escalate to team lead at 24h, and offer a service credit at 36h if unresolved. Target: 847 at-risk customers identified in the scatter plot cluster.

Expected impact:

Reduce churn in this segment from 67% to ~35% = retain ~270 customers = $19.4K/month recurring revenue protected

MONITOR

What metric to track, how often, what threshold triggers the next action

Metric: "Ticket-to-Resolution Rate within 48h" for at-risk segment.
Frequency: Weekly dashboard, reviewed every Monday.
Threshold: If resolution rate drops below 70% for 2 consecutive weeks, escalate to VP Customer Success and trigger hiring review.

Expected impact:

Early warning system prevents service degradation before it hits churn numbers

TEST

If [action], then [outcome], because [signal] - testable hypothesis

Hypothesis: If we proactively call customers with 1+ payment failure within 24h (instead of waiting for them to call us), then churn in that segment will drop from 58% to 40%, because payment failure is a distress signal of disengagement, not just a billing issue.

Expected impact:

A/B test over 6 weeks with 400 customers. If validated, scale to all payment-failure segments.

Confidence score ( 0 - 100% )

"Be honest: 74% = act with one hypothesis to validate. 45% = monitor first."

78%

Justification:

Strong correlation (4.2x) with clear mechanism (unresolved frustration -> departure). Historical data supports the pattern across 4 quarters. Uncertainty: we have not yet run the intervention, so the 35% retention estimate is modeled, not proven.

Confidence statement for the final slide

"Based on [X drivers], we recommend acting with [score]% confidence because..."

"Based on 3+ unresolved support tickets and 2+ payment failures as primary drivers, we recommend launching the 48-Hour Recovery Protocol with 78% confidence because the correlation is consistent across 4 quarters, the intervention is low-cost and reversible, and the at-risk population is precisely identifiable."

What would raise confidence to 90%?

Additional data, a test result, or a validation step

1. A 4-week pilot with 200 at-risk customers showing retention lift.
2. Qualitative interviews with 20 churned customers who had tickets - confirm "unresolved" was the primary reason.
3. Competitor benchmark - do other telcos with faster resolution have lower churn in equivalent segments?

BLOG SHORTLIST - From roughwork to publish-ready Circle top findings - Star the headline insight - Cross out the noise

Key finding (with numbers)

Top insight worth publishing

Blog headline angle / story hook

* 61% of churned customers had unresolved tickets in final 90d

Service recovery is the leak, not pricing. 4.2x churn with 3+ tickets.

"We thought price was killing us. It was our support queue."

24-36 month tenure window: 31% churn vs 9% for 4+ year customers

The "mid-life crisis" of customer loyalty - tenure is not linear protection

"Your loyal customers are not who you think they are."

73% of payment-failure customers never contacted support before churning

Silent churners: payment failure is a distress signal we ignore

"The customers who leave without complaining."

Low-price tier has highest churn rate

Price is not the driver - symptom of service quality

Noise - do not publish

48-Hour Recovery Protocol: $8K cost vs $111K/month revenue recovery

ROI of proactive support: 14:1 return on intervention cost

"The $8K fix that saves $111K a month."

Month-to-month contracts: 42% churn vs annual 11%

Contract type is a symptom, not a cause - already-disengaged customers choose monthly

Too obvious - skip

Satisfaction score missing for 34% of churned customers

Data gap itself is a signal: we are not surveying the customers who need us most

"The customers we never asked are the ones who left."

Reflection ( 5 min )

Which stage was hardest today?

Causality - separating correlation from mechanism. The team kept wanting to say "monthly charges cause churn" without proving the mechanism.

Which stage do you skip in real projects?

Decision. We often stop at "here is the insight" without naming who does what by when. This template forces the action.

Which insight surprised you most?

That low-price customers churn MORE than high-price. Completely counter-intuitive. Changed our entire retention strategy.

What will you do differently next week?

Run the 48-Hour Recovery Protocol pilot with 200 customers. Set up the weekly resolution-rate dashboard. Schedule the payment-failure A/B test.

One person you will share this story with:

VP Customer Success - she owns the support team and needs to see the revenue impact of ticket resolution speed.

Commitment card - Keep this

I will apply LUCID to:

All quarterly churn analysis reports

Dataset / project name:

TelcoPro Q3 Churn Prevention Initiative

Deadline ( within 14 days ):

04 Aug 2026 - pilot launch

Stage I will focus on most:

Decision (D) - I usually skip it

My name:

Alex Chen