INTERACTIVE DEMOS

AD/Marketing Tech

Real-time Bid Optimization Multi-Touch Attribution

A DSP (demand-side platform) processes 2M bid requests/second across 15 ad exchanges. Each bid decision must happen in under 100ms. The ML model's bid price needs continuous recalibration based on win rates, CPAs, and remaining campaign budgets.

⚠Campaign CAMP-8812's CPA drifted from target $4.20 to $7.80 over 3 hours because the bidding model wasn't recalibrated against real-time win/conversion data.

LIVEbid_events

bid_id	campaign_id	exchange	bid_price	won	impression	click	conversion	ts
bid_9a3f01	CAMP-8812	google_adx	2.4	true	true	false	false	2024-03-15T14:00:01.112Z
bid_9a3f02	CAMP-8812	google_adx	2.55	true	true	true	true	2024-03-15T14:00:01.445Z
bid_9a3f03	CAMP-8812	openx	2.2	false	false	false	false	2024-03-15T14:00:01.667Z
bid_9a3f04	CAMP-8812	pubmatic	2.8	true	true	true	false	2024-03-15T14:00:02.001Z
bid_7b2e01	CAMP-4455	google_adx	1.9	true	true	false	false	2024-03-15T14:00:01.230Z
bid_7b2e02	CAMP-4455	index_exchange	1.75	true	true	true	true	2024-03-15T14:00:01.890Z

Streaming SQLRunning

▶Compute rolling win rate, CTR, and CPA per campaign in 5-min windows✓

CREATE SOURCE bid_events WITH (
  connector = 'kafka',
  topic = 'dsp.bid_events',
  properties.bootstrap.server = 'broker:9092'
) FORMAT PLAIN ENCODE JSON;

CREATE MATERIALIZED VIEW campaign_metrics AS
SELECT
  campaign_id,
  window_start,
  COUNT(*) AS total_bids,
  COUNT(*) FILTER (WHERE won) AS wins,
  ROUND(
    COUNT(*) FILTER (WHERE won)::NUMERIC
    / NULLIF(COUNT(*), 0) * 100, 1
  ) AS win_rate_pct,
  COUNT(*) FILTER (WHERE click) AS clicks,
  ROUND(
    COUNT(*) FILTER (WHERE click)::NUMERIC
    / NULLIF(COUNT(*) FILTER (WHERE impression), 0) * 100, 2
  ) AS ctr_pct,
  COUNT(*) FILTER (WHERE conversion) AS conversions,
  CASE
    WHEN COUNT(*) FILTER (WHERE conversion) = 0 THEN NULL
    ELSE ROUND(
      SUM(bid_price) FILTER (WHERE won)::NUMERIC
      / COUNT(*) FILTER (WHERE conversion), 2
    )
  END AS effective_cpa,
  ROUND(AVG(bid_price) FILTER (WHERE won), 2) AS avg_win_price
FROM TUMBLE(bid_events, ts, INTERVAL '5 MINUTES')
GROUP BY campaign_id, window_start;

▶Generate bid adjustment signals when CPA deviates >15% from target✓

bid_adjustmentsauto-updating

campaign_id	current_cpa	target_cpa	deviation_pct	bid_modifier	win_rate	action
CAMP-8812	5.1	4.2	21.4	0.85	68.2	REDUCE_MODERATE
CAMP-4455	2.85	3.5	-18.6	1.15	72.1	INCREASE_MODERATE
CAMP-2190	6.35	4.8	32.3	0.7	55.4	REDUCE_AGGRESSIVE

✓RisingWave recalculates effective CPA per campaign every 60 seconds. When CAMP-8812's CPA hits $5.10 (21% above target), bid prices auto-reduce by 15% — CPA returns to $4.40 within 20 minutes.

Talk to Sales