Contextualized Growth Intelligence Loop
Combine internal patterns with external signals to explain growth differences
- Difficulty
- Advanced
- Time to result
- ~weeks to results
- Steps
- 7
- Confidence
- 96%
Frame a concrete growth objective, provide the model with secure internal data, and ask it to identify the small set of metrics or patterns most relevant to that goal. Analyze differences across geographies, cohorts, sessions, activation behavior, lifetime value, and acquisition economics while controlling for dominant market variables. Add customer journeys, prior experiment results, and operational constraints so recommendations reflect what the company has already tried. Then pair anonymized internal trends with external research, competitive traffic data, and local market signals to explain why patterns may differ. The output is a ranked set of hypotheses and experiments, not a causal verdict. Teams test the strongest explanations and return results to the corpus, making subsequent analysis progressively better informed.
Origin
Extracted from Marketing Against The Grain as Mayur Gupta and the hosts mapped a repeatable AI workflow for answering Kraken's global growth questions.
Core principles
- 01Begin with a specific growth question rather than a general data dump
- 02Use AI to search for multivariable patterns in internal data
- 03Control for dominant market effects before interpreting differences
- 04Add external evidence to contextualize internal trends
- 05Feed historical experiments and customer journeys into recommendations
How to run it
- 1
State the decision
Define the target, time horizon, and business decision the analysis must inform.
Pro tip Ask which few metrics matter most before requesting a full growth plan.
Watch out Broad requests encourage generic best-practice answers.
- 2
Prepare governed internal data
Supply relevant metrics, cohort data, journey flows, and unit economics through an approved secure environment.
Pro tip Anonymize trend data when external research tools cannot accept sensitive records.
Watch out Do not move confidential raw data into unauthorized cloud systems.
- 3
Control dominant variables
Identify market-wide forces that explain broad movement and account for them before comparing local outcomes.
Pro tip Use trusted price or market indexes as control variables when available.
Watch out Uncontrolled global effects can be mistaken for regional execution differences.
- 4
Find internal patterns
Compare geographic growth, activation, high- and low-value cohorts, session behavior, lifetime value, CAC, and payback metrics.
Pro tip Ask the model to rank candidate drivers and show the supporting variables.
Watch out Pattern recognition alone does not establish causation.
- 5
Add operational context
Provide previous experiments, customer journey maps, constraints, and known failures so the plan does not repeat old work.
Pro tip Include negative experiment results as first-class evidence.
Watch out A plan without historical context may recommend already-disproven ideas.
- 6
Correlate external signals
Use deep research, competitive data, and geography-specific evidence to contextualize the timing and direction of internal trends.
Pro tip Look for independent signals that agree across multiple sources.
Watch out External correlation can suggest an explanation but still requires testing.
- 7
Test and recycle learning
Turn the ranked explanations into experiments, measure outcomes, and return the results to the repository.
Pro tip Record why each experiment succeeded or failed.
Watch out Do not let AI-generated confidence replace controlled experimentation.
In the wild
A global exchange observes stronger activation in one geography than another. The team securely provides funnel events, customer journeys, cohort value, and historical experiments, while controlling for global Bitcoin-price movements. It then adds local search trends, competitive traffic, and market research. The model ranks plausible causes, including a journey difference and a geography-specific demand shift, which the team tests separately.
→ The growth team moves from a slow sequence of manual queries to a prioritized and externally contextualized experiment plan.
Common mistakes
Confusing correlation with cause
AI can surface strong patterns, but experiments or further evidence are required before claiming a causal relationship.
Ignoring external context
Internal trends may remain misleading when market, competitive, or geographic conditions are omitted.
Omitting experiment history
Without prior results, the model can recommend ideas the team already tested or rejected.
Is it for you?
Best for
Global growth teams analyzing geographic performance, cohort value, activation, retention, payback periods, or customer journeys.
Not ideal for
Decisions based on sparse, unreliable, improperly governed, or causally ambiguous data that cannot support meaningful comparison.
From the transcript
“One is in a global company, I'm always looking at growth rates, difference between geographies.”
“Pair it with external trends in those geographies.”
“So that can eliminate that variable and really understand what else is happening, you know, in a particular geography different from another one.”
From the episode
How a $1B+ Crypto Company Really Uses AI in Marketing