Introduction (00:16–01:25)

• Case overview: AT&T Long Distance faced a problem and Bell Labs engineers proposed a data-mining solution called the “bizocity” score.

Business context & data source (01:25–03:17)

• AT&T maintains a universe list of U.S. phone numbers (AT&T and non-AT&T). Traditional directory buying was costly/unreliable. Bell Labs noticed call-detail data (CDD) captured non-customer numbers when interacting with AT&T.

Problem statement (03:17–11:14)

• Business problem: inefficient customer acquisition/prospecting. Need to identify and target likely business customers (separate products for business vs. residential). Directory data lacked business/residence labels; calling to ask produced low response rates.

Detailed “as‑is” issues (11:14–17:05)

• Data-source unreliability and lack of segmentation (business vs. residential). Telemarketing response rates low; marketing budget constraints require better targeting.

Bell Labs’ approach — data insight (17:05–26:11)

• Use AT&T’s own Call Detail Data (CDD) instead of purchased directories: caller/receiver IDs, start/end times. CDD aggregates provide broad coverage of non-AT&T numbers.

Analytics formulation (26:11–33:03)

• Translate business problem to analytics: infer probability a number is business from behavioral indicators extracted from CDD.

Selected predictors and modeling (33:03–38:45)

• Key indicators: call time-of-day (day-hour activity), call duration (proxy for value), proportion of calls to businesses. Build a regression/classification model producing a daily bizocity score (weighted recent data more).

Scoring, profiling, and prioritization (38:45–44:00)

• Each number gets a bizocity score and a profile (recency, frequency, average duration, bizocity). Scores sorted to prioritize marketing targets within budget.

Business/marketing rationale (44:00–45:49)

• Combines likelihood (bizocity) with RFM-style value indicators to pick valuable, loyal prospects (maximize ROI on acquisition spend).

Implementation & operational considerations (45:49–48:13)

• Real-time/daily scoring requires infrastructure: ongoing data capture, model retraining, and integration into marketing workflows and BI architecture.

Key lessons and takeaways (throughout/end)

• Analytics must translate business needs into measurable models; leverage existing operational data; combine domain knowledge with statistical modeling; focus on business value, not just algorithmic sophistication.