Module 3 · Lecture 16

Dashboard Storytelling & the Recommendation

Four charts and a finding are not an answer. Today we turn them into the one page that goes to the board.

⏱ 90 minutes 🎯 CLO4 📝 Assignment 3

The Gap

You email last week's dashboard. The reply is one word.

“And?”

Level	Sounds like	Reader does
Description	“37.71% of bookings fail.”	Nods. Nothing changes.
Diagnosis	“Most failures are supply-side, peaking at 17:00.”	Finds it interesting. Nothing changes.
Recommendation	“Run an incentive at 16:30–17:30 in three hubs. Costs ₹X, recovers ₹Y, we know in three weeks.”	Decides.

Nobody ever acted on a chart. If your final slide has no verb, you built evidence for a case nobody made.

Deep-Dive · Step 1

Decomposition gives a number an owner

Outcome	Bookings	Share	Who can fix it
Cancelled by Driver	2,698	17.99%	Supply team
No Driver Found	1,036	6.91%	Supply team
Cancelled by Customer	1,022	6.81%	Product
Incomplete	901	6.01%	Operations

24.89%

fail on the supply side

3,734 bookings — nearly four times the customer-cancellation rate.

Why this matters

“37.71% fail” belongs to nobody. “A quarter fail on the supply side” names a department and a budget.

This is the single most valuable move in the whole analysis.

Deep-Dive · Step 2

We ask why, and get an unhelpful answer

Driver reason	Count	Share
Customer related issue	729	27.02%
More than permitted people in there	666	24.68%
Personal & Car related issues	652	24.17%
The customer was coughing/sick	651	24.13%

Interpretation 1

Causes really are evenly spread — so any single fix addresses at most a quarter. Say so before someone budgets for a silver bullet.

Interpretation 2 — more likely

Three of four blame the customer. Drivers pick from a dropdown under time pressure; the lowest-penalty option wins. This is a measurement artefact.

“Do not design an intervention around these codes until they are validated” is a real, defensible recommendation.

The Hidden Row

One category is absorbing another's failures

Customer reason	Count	Share
Driver is not moving towards pickup location	211	20.65%
AC is not working	210	20.55%
Driver asked to cancel	208	20.35%
Change of plans	203	19.86%
Wrong Address	190	18.59%

The driver did not want the trip, asked the passenger to cancel so the penalty fell on them, and the data now calls it a customer cancellation. Add the 211 “not moving towards pickup” and ~419 of 1,022 (41%) are plausibly driver-initiated.

Supply-side share moves from 24.89% toward 27.6%. Always hunt for the column hiding someone else's failures.

Stress Test · 1 & 2

Attack your own finding first

“Is this a small sample?”

Fair — it is exactly what catches 03:00 (43.48% on 46 bookings). But 17:00 carries 1,062. Stable to a fraction of a point.

Survives.

“Isn't 39.92% ≈ 37.14%?”

2.78 points — a 7.5% relative increase. About 30 extra failed trips in that hour, ~₹7,000 of fares. Material over a year, modest in one hour.

Partially survives — target a window, not a point.

Do this in private, before someone does it to you in a board meeting.

Stress Test · 3 & 4

The two attacks most analysts never run

“Composition, not time?”

If 17:00 carries more airport trips, and airport trips fail more, the hour is innocent. Segment 17:00 by vehicle type and pickup, compare to the daily mix. Same mix + different rate ⇒ time is the driver.

“What would make this wrong?”

If driver log-ins showed the same active supply at 16:00–17:00 as at 18:00, repositioning collapses. We do not have that data — and saying so names exactly which dataset resolves it.

This is the check that separates a finding from a coincidence — and the rubric rewards it.

The Second Data Point

16:00 is what turns a point into a trend

40.45%

16:00 failure rate

39.92%

17:00 failure rate

37.14%

18:00 — the demand peak

One anomalous hour is a curiosity. Two consecutive elevated hours immediately before the peak is a mechanism: drivers have not yet repositioned for the evening rush.

The intervention is now a window — 16:00 to 18:00 — not a single hour.

Geography

Time says when. Location says where.

Pickup location	Bookings	Cancelled	Rate
Noida Sector 62	1,209	405	33.50%
Indira Gandhi International Airport	1,183	388	32.80%
New Delhi Railway Station	1,179	391	33.16%
Gurgaon Cyber City	1,178	368	31.24%
Connaught Place	1,139	374	32.84%

Five places — an office cluster, an airport, a station, a business district, a city centre — carry 5,888 bookings: 39% of all demand.

The Boring Chart

A flat distribution is a finding when it rules something out

Cancellation rates across the top five hubs span 31.24% to 33.50% — barely two points. There is no catastrophic location.

What that proves

The problem is systemic, not local. A fix aimed at “the bad hub” cannot work, because there is no bad hub.

What it buys you

The intervention must be time-based and system-wide, concentrated where volume already is. The chart that looked boring did real work.

Combined: target 16:00–18:00 at the five hubs carrying 39% of demand. Defined, costable, measurable.

Structure

Five findings in discovery order are a lab notebook

#	Beat	Content
1	The stake	37.71% fail; up to ₹13.3 lakh at risk
2	The narrowing	~25% fail on the supply side — it has an owner
3	The mechanism	Peaks at 16:00–17:00, before the demand peak
4	The target	Five hubs, 39% of demand, uniform rates
5	The action	One costed recommendation with a date
6	The limits	What this cannot establish, and what would

In the right order they are an argument.

Layout

Readers scan in a Z

Top-left — read first

The stake. The status donut frames everything that follows.

Top-right — read second

The mechanism. The 16:00–18:00 failure curve.

Bottom-left

Supporting detail — revenue by product, cancellation reasons.

Bottom-right — read last, ignored first

Never put your key finding here.

Three rules do most of the work: title states question + population · every axis carries its unit · every number traces to a runnable pipeline.

The Deliverable · 1 of 2

Ride Completion — Recommendation

Finding

5,657 of 15,000 bookings (37.71%) carried no passenger, against ₹22.0 lakh realised. At ₹235.73 average fare that is up to ₹13.3 lakh forgone — an upper bound, since some customers rebooked.

Diagnosis

At least 24.89% of all bookings (3,734) fail on the supply side. A further 419 customer cancellations cite “driver asked to cancel” or “driver not moving”, pushing the share toward 27.6%.

Mechanism

Failure peaks at 16:00 (40.45%) and 17:00 (39.92%), above the 18:00 peak's 37.14% despite 18:00 carrying more bookings. Supply lags demand by one to two hours.

The memo is the deliverable. The dashboard is the appendix.

The Deliverable · 2 of 2

…and what we will do about it

Recommendation

Pilot a time-boxed driver incentive, 16:00–18:00, at the five highest-volume hubs (39% of demand). Size it below ₹235.73 so the economics hold at any recovery rate above break-even.

Expected impact

Closing the 16:00–17:00 gap to the 18:00 baseline recovers ~60 trips/day across the two hours — roughly ₹14,000 daily, ₹51 lakh annualised, before incentive cost.

How we will know

Compare pilot vs non-pilot hubs over three weeks. If the gap does not close by at least one percentage point, the explanation is wrong and we stop.

Naming the condition under which you abandon your own recommendation is the strongest credibility signal available.

Why It Works

Five properties of a page that gets acted on

Every number has a denominator. Not “cancellations are high” but “3,734 of 15,000”.
The recommendation has a verb and a boundary. Pilot — these five hubs — this window — below this price.
The impact is arithmetic the reader can check. Show the multiplication; never assert the product.
There is a stopping rule. A date and a threshold at which you walk away.
The limitations are specific. “We lack driver log-in data” is useful. “Further research is needed” is filler.

A reader should be able to act, or refuse to act, in ninety seconds.

Limitations

Under-reporting limits is what destroys trust

Gap	Weak	Strong
No driver supply data	“More data would help.”	“Driver log-in timestamps would confirm or kill this in a day.”
Unreliable reason codes	“Reasons are unclear.”	“Three of four blame the customer in near-equal shares — consistent with penalty-avoidance.”
₹13.3 lakh estimate	“Approximate.”	“Upper bound assuming no rebooking. The conclusion holds even if it is half.”
Correlation only	“Causation unproven.”	“The pilot is the experiment. That is why we test on a subset.”

Pattern: name the gap, state what closes it, say whether the conclusion survives. Handled that way, a limitation adds authority.

Your Turn

Finish the dashboard, then write the page

In the lab

Add the pickup chart and the customer-reason chart. Verify the 208 yourself. Run the composition check. Re-sequence into the six beats. Retitle every chart as a question.

Assignment 3

Write your own one-page memo. You may well disagree with the incentive pilot — a well-argued alternative scores higher than a copy.

Then hand it to a classmate and ask for the weakest number on the page. Whatever they find is what the marker will find.

Next — Module 4: this dashboard recomputes on every open. Fine weekly, hopeless for a wall screen refreshing every second. Enter Redis.