Acquisition Trends

Acquisition trends — multi-month performance analysis

"分析下我过去 6 个月的拉新情况" / "show me the last 12 months"

Time-series view of the affiliate node: new invitees, activation rates, deposits, volume, commission, and active-trader counts. Surfaces inflection points and the events behind them.

When to use this skill

• User asks for trends, history, MoM, QoQ, YoY analysis
• Quarterly review / pitch deck preparation
• Investigating a sudden change ("why is commission down 30% this month?")
• Replicating a past success ("what did we do in August?")

What the agent does

Step 1 — Choose the window

Default to 12 months if user says "trend" / "history" without a number. For specific requests like "last 6 months" or "Q1 2026" honor that exactly.

Step 2 — One call per calendar month

Loop over each month in the window:

{
  "name": "okx-affiliate-performance-summary",
  "arguments": {
    "begin": "<month start ms>",
    "end":   "<month end ms>"
  }
}

Cap parallelism at ≤ 5 concurrent calls; the endpoint rate-limits with code: 50011. Insert a 200–500 ms delay between batches.

Step 3 — For finer-grained trends, pull daily data

If the user asks for daily-active-trader (DAT) history or wants to identify peak days, loop day-by-day over the window:

{
  "name": "okx-affiliate-performance-summary",
  "arguments": {
    "begin": "<day 00:00:00 UTC ms>",
    "end":   "<day 23:59:59 UTC ms>"
  }
}

90 days = 90 calls. Run serially with sleep 0.4 between requests to avoid 429s. For >180 days, suggest the user request only monthly granularity unless they truly need the daily resolution.

Step 4 — Cache aggregates

Save results as CSV (one row per period) for follow-up analysis without re-pulling:

date,invitees,firstTimeTrader,traderCnt,depAmt,vol,commission

Use the details[] array from the response to sum across SPOT/DERIVATIVE/BSC.

Insights to extract

• Direction of travel — first 30 days vs middle 30 vs last 30. Quote the deltas.
• Peak month and peak day — the highest value of each metric and the date
• Acquisition velocity vs commission lag — new invitees in month M often produce most of their commission in months M+1 to M+3; surface this lag if it's pronounced
• Activation rate trend — firstTimeTrader / invitees per month. Healthy nodes hold this ratio above 40%; below 30% signals quality erosion
• Volume-per-trader trend — total vol / traderCnt. Falling per-trader volume with steady trader count = users trading smaller (market conditions or product-mix shift)
• Spot vs derivatives mix drift — if derivatives share is climbing, the user base is getting more risk-tolerant; useful for product recommendations
• Concentration risk — % of monthly commission from the top 3 users. If > 50% the node is fragile

Sample output format

📊 Acquisition trend — <START> to <END> (<N> months)

📈 Monthly snapshot
| Month   | New inv | First-time | Activation | Deposit | Volume | Commission |
| 2025-06 |     338 |        135 |        40% |  $2.13M | $756M  | $61,970    |
| ...

📊 Quarterly view
• Q2 (M–M): X new / Y first / $Z commission — strongest acquisition quarter
• Q3 (M–M): ...
• ...

🏔️ Peaks
• Peak monthly new invitees: <N> in <YYYY-MM>
• Peak monthly commission: $X in <YYYY-MM>
• Peak DAT day: <N> traders on <YYYY-MM-DD>

🔍 Key insights
1. <inflection point with date and likely cause>
2. <activation rate trend>
3. <commission concentration>
4. <volume-per-trader drift>

📌 Suggested action
<one paragraph: what to replicate from the best period, what to watch in the current
period, where to focus the next 30 days>

📁 Raw data
CSV with daily/monthly rows saved at <path>. Useful for replay against BTC price,
campaign timestamps, etc.

Recommended follow-ups

• "Why did spike?" — pull the channel-level (okx-affiliate-link-list) data for that month to see if a specific link drove it
• "Cross-reference with BTC price" — pull a price chart from blockbeats and overlay
• "Who were the top users in ?" — re-run high-potential-invitees with the peak month's date range and see if they're still active

Gotchas

• Custom-range timestamps are Unix milliseconds in UTC. int(datetime(2026,4,1).timestamp()*1000) in Python. Off-by-year mistakes are common.
• Months have different lengths — when comparing MoM, normalize daily averages or be explicit about absolute vs. per-day numbers.
• The endpoint returns 429 above ~5 RPS, then succeeds again seconds later. Build retry logic with exponential backoff; do not panic on a single 429.
• okx-affiliate-sub-affiliate-list is lifetime-only in the new MCP — for time-windowed sub-affiliate analysis pull invitee-list with subAffiliateUid... but that filter is currently broken, so client-side filtering is needed.
• For long backfills, consider chunked monthly calls (12–24 calls) rather than daily (300+ calls). The user usually wants the shape, not every day.