Build the Rule Matrix: Turning Five Customer Segments Into a Personalization Plan

Marketing Academy · Field Work●AI in Marketing

All projects·From: AI Personalization at Scale

MiniBuild the Asset· 35 minutes

Go Digit General Insurance

Objective: Given synthetic behavioral data for five customer segments at an insurance company, build a personalization-rule matrix that maps each segment to the correct lever (dynamic content, product recommendation, or triggered message), the trigger event, and a one-line message variant.

You're a lifecycle marketing associate at Go Digit General Insurance, the Bengaluru-founded, Nasdaq-listed general insurer. Digit's app team just shipped an event stream, and you have five weeks of behavioral exports for five customer segments but no personalization plan yet.

Sort each segment by its dominant behavioral signal, assign the correct lever from the lesson's three levers, and write the trigger and message variant that fits that lever, not a generic email blast.

Before you start

What you'll need

Free path (everything below is enough to finish)

Google Sheets

FreeBuild and sort the segment-to-lever matrix

No account friction, sorting and column formulas are all this task needs

The process

1 step

Step 01 of 01

Matching personalization levers to behavioral signals, not demographics

The lesson's playbook names three levers, dynamic content, product recommendations, and triggered 1:1 messaging, and Stage 1 says the lever choice should follow the behavioral signal, not a static demographic label.

Segment C is 'quote started, never completed, price-sensitive' with a 41% cart-abandon rate on the premium calculator. Segment E is 'policy renewal due in 14 days, no app open in 30 days.' Which lever and trigger fits each, and why is a generic 'we miss you' email the wrong call for both?

Google Sheets— Import the five-segment export, one row per segment with columns for signal, volume, and current treatment (if any).

Procedure

Import the export and freeze the header row
Add three columns: Lever, Trigger Event, Message Variant
For each segment, read its dominant signal column first, ignore the demographic columns entirely
Assign Segment C (quote-abandon, price-sensitive) to triggered 1:1 messaging fired on cart-abandon, offering a rate-lock reminder rather than a discount, since Stage 4's optimize loop shows discounting price-sensitive users first erodes margin before it's tested
Assign Segment E (renewal due, app-dormant) to triggered 1:1 messaging fired 14 days before lapse, since a dynamic content change on a page nobody is visiting reaches zero people
Assign Segment A (browsing multiple policy types, no purchase) to a product recommendation lever surfacing the policy type they've viewed most
Leave a Notes column stating which lever you rejected for each segment and why

Sample output

Segment | Signal | Lever | Trigger | Message Variant
C | Quote abandoned, price-sensitive (41% abandon) | Triggered 1:1 messaging | Cart-abandon, 2hr delay | "Your quote is saved. Complete it before your rate-lock window closes."
E | Renewal due in 14 days, app-dormant 30+ days | Triggered 1:1 messaging | T-minus-14-days | "Your policy renews on [date]. Review your coverage in 2 minutes."
A | Viewed 3 policy types, no purchase | Product recommendation | Next app session | Surface the policy type with the longest dwell time first
...2 more rows

Healthy

Every segment's lever is justified by its own signal column, and at least one lever choice explicitly overrides what a demographic-only segmentation would have picked.

Unhealthy

All five segments get the same lever (usually 'send an email'), or the Notes column is empty because no lever was ever rejected.

What this means

If every row picked the same lever, the matrix is really just segmentation with extra steps, exactly the Common Mistakes trap the lesson calls out.

So what do I do about it?

Symptom	Action	Effort
Price-sensitive abandoners get the same treatment as dormant renewal segments	Split the matrix by trigger event first, lever second	30 min

YouYou can do this yourself, no engineering access required.

Final deliverable

A five-row personalization-rule matrix (segment, signal, lever, trigger, message variant) with a Notes column justifying each lever choice.

See a reference example

Sample output

Sephora Beauty Insider, Q3 personalization matrix (excerpt)

Segment | Signal | Lever | Trigger
Lapsed high-spender | No purchase in 60 days, prior AOV $180+ | Triggered 1:1 messaging | Day-60 dormancy
Virtual Artist browser, no cart add | Used AR try-on 3x, no purchase | Product recommendation | Session end
New visitor, no account | First site visit, no email captured | Dynamic content | Homepage load

Notes: rejected a blanket 20%-off email for the lapsed high-spender segment, past purchase history shows this cohort responds to new-arrival curation, not discounting.

Success criteria

You're done when you can:

Each of the 5 segments has a lever, trigger, and message variant tied to its own signal column
At least one lever choice explicitly rejects a demographic-only alternative in the Notes column