Buy vs Build: Scoring Nubank's Fraud-Alert Copy Workflow
Objective: Given two candidate AI content workflows, apply the lesson's three-question buy-vs-build framework in its stated order and reach a defensible Build or Buy call for each.
You're on Nubank's growth marketing team evaluating two AI content workflows this quarter: personalized fraud-alert push notification copy across Brazil, Mexico, and Colombia, and the standard monthly product newsletter.
Score both use cases against the three-question framework, in order, and recommend Build or Buy for each with reasoning tied to the framework, not intuition.
Before you start
What you'll need
Free path (everything below is enough to finish)
Free, shareable, no account friction for a quick scoring exercise
The process
1 step
Step 01 of 01
The lesson's three questions, asked in order: how commoditized is the use case, does the team have technical resource to maintain a build, and how core is the workflow to differentiation.
Nubank's growth team has two candidate AI content workflows this quarter: personalized fraud-alert push copy across 3 countries, and the standard monthly product newsletter. Score both against the three questions and recommend build or buy for each.
Procedure
- List both use cases as rows: fraud-alert push copy, monthly newsletter.
- Score column 1 (commoditized?) Yes/No for each: newsletter = Yes (every fintech sends one), fraud-alert copy = No (it depends on Nubank's own fraud model and risk segments).
- Score column 2 (technical resource available?) Yes/No: Nubank already has a data science team scoring fraud risk, so Yes for the alert workflow.
- Score column 3 (core to differentiation?) Yes/No: personalized, real-time fraud alerts are part of Nubank's trust-driven brand; a generic newsletter is not.
- Where columns 2 and 3 are both Yes and column 1 is No, mark 'Build'. Otherwise mark 'Buy'.
Nubank AI workflow scoring (excerpt) Use case: Monthly product newsletter Commoditized? Yes Technical resource? Yes Core to differentiation? No Recommendation: BUY (all-in-one platform) Use case: Personalized fraud-alert push copy (BR/MX/CO) Commoditized? No Technical resource? Yes (existing fraud data science team) Core to differentiation? Yes (trust is Nubank's brand) Recommendation: BUILD (custom on LLM API, layered on existing fraud model)
Healthy
Newsletter scored Buy, fraud-alert copy scored Build, each for a different one of the three reasons.
Unhealthy
Recommending build for the newsletter because 'AI is strategic', ignoring that it's a commodity task with no differentiation upside.
What this means
A use case only earns a Build recommendation when all three questions point the same direction. A single No is enough to default back to Buy.
So what do I do about it?
| Symptom | Action | Effort |
|---|---|---|
| A team defaults to build because they have the technical skill, without checking commoditization first | Score commoditization before scoring technical resource, per the framework's own stated order | 5 min |
Final deliverable
A one-page build-vs-buy recommendation memo scoring both use cases against the three-question framework, with a clear Build or Buy call for each.
See a reference example
Casper Sleep, Q3 AI workflow scoring (excerpt) Use case: Return/exchange policy chatbot answers Commoditized? Yes (every mattress retailer needs this) Technical resource? No dedicated engineer on marketing Core to differentiation? No Recommendation: BUY (all-in-one platform's support module) Use case: 'Which mattress firmness fits your sleep position' quiz logic Commoditized? No (Casper's own comfort-matching IP) Technical resource? Yes (in-house data team) Core to differentiation? Yes (this quiz is Casper's signature conversion tool) Recommendation: BUILD
Success criteria
You're done when you can:
- Scores both use cases against all three framework questions, in the framework's stated order
- Reaches a Build or Buy call consistent with the scoring, not an intuition-based guess