Dataset Improvements

I scaled from 10,000 → 40,000 sessions, because 7 days don’t reveal retention patterns.

Main Goals of the day:

Scale simulation from 1 week → 4 weeks (28 days)
Keep the number of players (1,200)
Simulate 40,000 sessions → 33.3 avg sessions per player
Model progressive churn decay — not just “Day 7”, but Week 1 to Week 4

Step by Step

📍 Step 1: Changed start = datetime(2025, 9, 1) → end = datetime(2025, 9, 28)
📍 Step 2: Adjusted Poisson lambda from 8.33 → 33.3
📍 Step 3: Kept n_players = 1200 — because real operators don’t get 10K new players every week
📍 Step 4: Modified session duration distribution to reflect weekend spikes and mid-week drop-offs
📍 Step 5: Adjusted bonus claim probability to decay after Week 1 — real players stop claiming after 7–10 days

Challenges / Insights

✅ More time = better insight.

In 7 days, everyone looks like a loyal player.
In 28 days?
→ 76% drop after Week 1
→ 39% after Week 2
→ Only 15% still active at Week 4

That’s not noise.
That’s the real iGaming economy.

I used to think:

“I need more players.”

Now I know:

“I need more time.”

Because churn doesn’t happen on Day 1.
It happens after the bonus runs out.

Code Snippet Final

``` python
# New parameters
start = datetime(2025, 9, 1)
end = datetime(2025, 9, 28)  # 4 weeks
n_players = 1200
n_sessions = 40000
avg_session = n_sessions / n_players  # 33.3 — realistic for active players

´´´

Next Step

I have the data, the model, and the dashboard.

Next: Understanding the strengths and the weaknesses of my work.