I kept running pricing experiments.
Raise price → churn increases.
Lower price → revenue drops.
Try again → same mistakes.
The problem wasn’t strategy.
It was memory.
What We Built
We built ExpTrack AI — a system that remembers every business experiment and uses that memory to guide future decisions.
Not dashboards.
Not analytics.
A system that learns from past mistakes before you repeat them.
At its core, the system does three things:
Stores every pricing experiment as structured memory
Runs a pre-flight check before new experiments
Uses past outcomes to generate real-time insights
The Core Problem: Stateless Decision-Making
Most businesses run experiments like this:
Try something
Observe results
Move on
But the system forgets.
There is no accumulated intelligence.
[
]
Even when data exists, it’s:
Scattered across tools
Not connected to decisions
Not used proactively
So teams repeat:
Failed price increases
Ineffective discounts
Wrong assumptions
Introducing Hindsight-Based Memory
Instead of treating each experiment independently, we treat them as linked memory objects.
Each experiment stores:
Price change (delta %)
Hypothesis
Outcome (success/failure)
Reason (what actually happened)
Before running a new experiment, the system performs a memory recall.
How the System Works
- Pre-Flight Check (Before Decision)
When a user inputs:
Current price
Proposed price
Hypothesis
The system calls:
POST /check-experiment
It retrieves similar past experiments and returns:
Number of related experiments
Pattern-based insight
Risk signal (warning or approval)
Instead of guessing, the system says:
“This failed before — here’s why.”
- Memory Update (After Decision)
Once the experiment is completed:
PATCH /update-result
The system logs:
Outcome (Success / Failure)
Actual reason
This closes the feedback loop.
Now the agent doesn’t just store data — it learns causality.
- Pattern Extraction
Over time, the system surfaces insights like:
“Price increases >15% fail frequently”
“Discounting improves conversion but hurts retention”
“Small incremental changes outperform large jumps”
This is not static analytics.
This is behavioral pattern learning from memory.
What Makes This Different
Most AI tools:
Respond to prompts
Generate outputs
Forget everything
This system:
Remembers
Compares
Adapts
The difference is statefulness.
Without memory → AI is reactive
With memory → AI becomes strategic
UI as a Reflection of Memory
The interface is intentionally simple but structured around memory:
Experiment input panel → decision entry
Pre-flight check → memory recall
Experiment cards → stored history
AI insights → learned patterns
Even locally, experiments are persisted and reused
This design ensures:
No loss of context
Continuous learning
Immediate feedback loop
Real Behavior Change (Before vs After)
Before
Decisions based on intuition
Repeated mistakes
No structured learning
After
Decisions validated against past outcomes
Early warnings before failure
Compounding intelligence over time
Why Hindsight Matters
Hindsight is not just history.
It’s compressed experience.
A single failed experiment is noise.
Ten similar failures become a pattern.
The system transforms:
“We tried this before.”
into
“This fails under these conditions — avoid it.”
What We Learned Building This
Memory is more valuable than raw data
Learning happens only when outcomes are tracked
AI without context is just autocomplete
Business decisions need historical grounding
Small feedback loops outperform big predictions
Where This Goes Next
This system can extend beyond pricing:
Marketing campaigns
Product feature experiments
Sales strategies
User onboarding flows
Anywhere decisions repeat → memory compounds value.
Final Thought
AI doesn’t fail because it’s not smart enough.
It fails because it doesn’t remember enough.
Once you give it memory,
it stops being a tool…
and starts becoming a system that learns with you.
Github: Exptracker.AI
Top comments (0)