Executive Summary
Delivered marketing analytics and data governance solutions for AT&T’s advertising division (YP) during a critical period of digital transformation from print-first to omni-channel advertising. Built campaign performance tracking systems and advertiser intelligence dashboards that gave sales and marketing teams real-time visibility into campaign ROI. Established a data governance framework—standardised definitions, data dictionaries, and stewardship processes—that reduced data quality incidents by 34% and improved billing accuracy. Designed a multi-channel attribution model reconciling digital (search, display, social) and print channels previously tracked in separate silos. Performed advertiser segmentation and lifetime value modeling driving a 12% improvement in targeted renewal campaigns.
Where I Learned That Dirty Data Kills Good Marketing
YP Advertising was where I learned that bad data doesn’t just produce wrong reports—it loses you customers. My very first week, I was asked to pull a list of advertisers whose campaigns had lapsed so the renewal team could reach out. I ran the query, handed over 1,400 names, and the team made their calls. Two days later, the sales director was in my manager’s office: 340 of those advertisers were already active—their renewals had closed but the status hadn’t propagated from billing to the marketing database. We had called active customers offering them “win-back” deals. That was my introduction to data governance. Before that week was over, I was mapping every source system that fed our advertiser records and documenting where they disagreed with each other. Data governance wasn’t a project assigned to me—it was a fire I had accidentally uncovered, and I turned it into a framework.
Marketing Campaign Analytics
End-to-end campaign performance reporting; multi-channel attribution model reconciling digital (search, display) and print; automated dashboards reducing manual reporting from 3 days to same-day; 95%+ campaign coverage.
Data Governance
Enterprise data dictionary across 6 source systems; standardised advertiser master record; data stewardship processes reducing quality incidents 34%; governance framework adopted division-wide.
Advertiser Intelligence
Segmentation models and lifetime value analytics across 8,000+ advertiser accounts; churn prediction enabling targeted renewal outreach; 12% improvement in renewal campaign conversion.
Business Challenges
- Fragmented Data Governance: Six source systems (Salesforce CRM, billing, fulfillment, web analytics, campaign management, print operations) each held conflicting advertiser records with no master data layer and no agreed-upon definitions for key metrics like “active” or “churned.”
- Multi-Channel Attribution Gap: Digital campaign performance lived in separate tools from print; there was no unified view of total advertiser spend, reach, or ROI, making it impossible to show clients holistic campaign value.
- Manual Reporting Bottleneck: Weekly campaign performance reports were assembled by hand in Excel across three teams, taking 2–3 business days and containing frequent errors that damaged advertiser trust.
- Advertiser Churn Blind Spot: The renewal team operated on gut feel and tenure relationships with no predictive model to surface at-risk advertisers before cancellation decisions were made.
- Billing Data Quality: Inconsistent advertiser IDs across billing and CRM caused revenue mismatches in monthly reconciliations, delaying close and creating disputes.
The Data Governance Lesson That Changed How I Define Everything
Six months into the role, the VP of Marketing asked me to pull “total active advertisers.” I returned a number. The finance team returned a different number. The operations team had a third number. Three different answers to the same question—because “active” meant something different in every system. In billing, active meant “has an open invoice.” In CRM, it meant “account is not marked cancelled.” In fulfillment, it meant “at least one ad is currently running.” All three were defensible. None were the same. That meeting was the catalyst for the data dictionary project. I spent the next three months facilitating cross-functional workshops with finance, sales ops, and fulfillment to agree on canonical definitions for 47 business metrics. Every metric now had a definition, an owner, a source system, and a calculation rule. I didn’t just fix a reporting discrepancy—I established the governance infrastructure that made future discrepancies impossible.
Marketing & Campaign Performance
Multi-channel campaign tracking, advertiser ROI dashboards, and data quality scorecards across 8,000+ active advertiser accounts.
Multi-Channel Campaign Attribution
The Attribution Discovery That Saved Print
When I first built the multi-channel attribution model, digital looked overwhelmingly superior to print on a last-touch basis. The sales team was gearing up to aggressively push advertisers off print and onto digital-only packages. But the data was lying through omission: print ads didn’t generate clicks—they generated calls. Phone call volume wasn’t being fed back into the attribution model at all. When I integrated call tracking data and built a phone-call attribution layer, the picture changed completely. Print was driving 31% of verifiable customer contacts for local service businesses (plumbers, electricians, restaurants) who didn’t track online conversions. The “print is dying” narrative in the data was a measurement problem, not a market reality. Attribution only works when you measure all the channels that matter.
Solution Architecture
- Advertiser Master Data Layer: Built a unified advertiser master record by matching records across Salesforce CRM, billing, and fulfillment using fuzzy name matching and address standardisation; reduced duplicate accounts by 28% and established a single source of truth for advertiser status
- Data Governance Framework: Facilitated cross-functional workshops to define 47 canonical business metrics; maintained a living data dictionary documenting each metric’s definition, owner, source system, and calculation logic; implemented data stewardship assignments across finance, sales ops, and fulfillment
- Multi-Channel Attribution Model: Designed an attribution model integrating digital campaign data (search, display, social), print directory performance, and phone call tracking; weighted attribution by channel interaction sequence to give advertisers a holistic view of campaign ROI
- Automated Campaign Reporting: Built SQL-driven reporting pipelines in SSRS and Power BI replacing 2–3 day manual Excel processes; reports refreshed nightly and were available same-day for account managers and advertiser-facing teams
- Advertiser Segmentation & LTV Modelling: Developed RFM (Recency, Frequency, Monetary) segmentation across 8,200+ accounts; built a churn propensity regression model identifying at-risk advertisers 60 days before renewal; enabled the renewal team to prioritise outreach with data-backed talking points
- Data Quality Scorecard: Implemented automated data quality checks across all six source systems; weekly scorecards surfaced completeness, accuracy, and consistency metrics to data stewards; established SLAs for data quality remediation reducing incident volume 34% within six months
Why Data Governance Is Really a People Problem
I learned the hard way that you can build the best data dictionary in the world and it means nothing if people don’t use it. My first version was a beautifully formatted Excel file with tabs for every domain. Nobody looked at it. I’d send it out before meetings and people would nod politely, then still use whatever definition came from their own system. The turning point was when I stopped “documenting” and started “facilitating.” Instead of defining “active advertiser” for everyone in a document, I brought finance, sales ops, and fulfillment into a room and made them argue it out. When they reached consensus, I wrote it down. Suddenly it was “their” definition, not mine. Adoption happened organically because the people who would use the data had ownership of the definition. That insight—governance is change management, not data management—has guided every governance initiative I’ve led since, including the data quality frameworks I built at AT&T years later.
Sample SQL — Advertiser Churn Propensity
SQL — Advertiser Renewal Risk Scoring
-- Score advertiser renewal risk 60 days before contract expiration
WITH advertiser_activity AS (
SELECT
a.advertiser_id,
a.account_name,
a.renewal_date,
a.account_manager_id,
a.segment,
COUNT(DISTINCT c.campaign_id) AS active_campaigns,
SUM(c.monthly_spend) AS total_monthly_spend,
AVG(c.click_through_rate) AS avg_ctr,
SUM(c.phone_calls_attributed) AS total_attributed_calls,
MAX(c.last_performance_review_date) AS last_review_date
FROM dim_advertisers a
LEFT JOIN fct_campaigns c
ON a.advertiser_id = c.advertiser_id
AND c.campaign_status = 'ACTIVE'
WHERE a.renewal_date BETWEEN DATEADD('day', 0, CURRENT_DATE)
AND DATEADD('day', 60, CURRENT_DATE)
GROUP BY 1, 2, 3, 4, 5
),
rfm_scores AS (
SELECT
advertiser_id,
DATEDIFF('day', last_review_date, CURRENT_DATE) AS recency_days,
active_campaigns AS frequency,
total_monthly_spend AS monetary
FROM advertiser_activity
),
churn_risk AS (
SELECT
aa.advertiser_id,
aa.account_name,
aa.renewal_date,
aa.account_manager_id,
aa.segment,
aa.total_attributed_calls,
aa.avg_ctr,
rfm.recency_days,
rfm.frequency,
rfm.monetary,
-- Risk score: higher = more at risk
CASE
WHEN rfm.recency_days > 45 THEN 3 ELSE 0 END +
CASE
WHEN rfm.frequency < 2 THEN 2 ELSE 0 END +
CASE
WHEN rfm.monetary < 500 THEN 2 ELSE 0 END +
CASE
WHEN aa.avg_ctr < 0.015 THEN 1 ELSE 0 END +
CASE
WHEN aa.total_attributed_calls = 0 THEN 2 ELSE 0 END AS churn_risk_score
FROM advertiser_activity aa
JOIN rfm_scores rfm USING (advertiser_id)
)
SELECT
advertiser_id,
account_name,
renewal_date,
account_manager_id,
segment,
churn_risk_score,
CASE
WHEN churn_risk_score >= 7 THEN 'HIGH'
WHEN churn_risk_score >= 4 THEN 'MEDIUM'
ELSE 'LOW'
END AS risk_tier,
total_attributed_calls,
avg_ctr,
monetary AS monthly_spend
FROM churn_risk
ORDER BY churn_risk_score DESC, renewal_date ASC;
The Churn Model That Taught Me About Feature Engineering
My first churn model used only spend data and contract tenure. It was directionally right but barely outperformed a simple rule-of-thumb. The breakthrough came when I talked to experienced account managers and asked: “When you know an advertiser is going to cancel before they tell you, what do you see?” Their answers weren’t in any database: “They stop returning my calls.” “They didn’t ask for their performance report.” “They asked about contract terms.” I translated those signals into measurable proxies: days since last performance review, number of unanswered account-manager emails, whether they’d downloaded their campaign report in the last 30 days. Adding those engagement signals improved the model’s precision by 22 percentage points. The best features in a business model often come from human expertise, not data exploration. I now treat every domain expert as a feature-engineering partner.
Results
Data Governance Impact
Data quality incidents reduced 34% within six months of governance framework launch; 47 canonical metric definitions adopted division-wide eliminating reporting discrepancies; data stewardship model reduced billing reconciliation disputes by 41%.
Campaign Analytics Impact
Campaign reporting time reduced from 2–3 days to same-day; 95.4% of active campaigns tracked vs. ~61% under manual process; multi-channel attribution model revealed print was driving 31% of verifiable customer contacts, preserving a product line that data would have otherwise wrongly condemned.
Advertiser Retention Impact
Churn propensity model enabled proactive outreach to 280+ high-risk accounts per renewal cycle; renewal campaign conversion improved 12%; advertiser lifetime value increased by an average of $1,800 per retained account through upsell recommendations driven by segmentation analytics.
The Foundation That Built Everything After
Looking back, AT&T/YP Advertising was one of the most formative experiences of my career. Not because of the technology—we were using SQL Server, SSRS, and early Power BI—but because of the problems. Data governance before “data governance” was a fashionable term. Attribution modelling before every platform had a built-in attribution tab. Churn prediction for a product whose own industry was uncertain about its future. The ambiguity forced me to develop rigour: if you can’t trust your data, you have to build the infrastructure to trust it before you can analyse anything. That instinct—fix the foundation first, then build the analysis—has shaped every engagement since. The enterprise data governance work I led at AT&T years later? I already knew the playbook. I had written the first draft of it here.
Challenges & Lessons
- Governance is Change Management: The technical work of building a data dictionary is straightforward; the hard work is getting cross-functional stakeholders to agree on definitions and actually use them. Facilitation and ownership matter more than documentation quality.
- Attribution Requires All Channels: An attribution model that misses a whole class of conversions (phone calls, in-person visits) will produce confidently wrong conclusions. Always ask: “What channels aren’t we measuring?” before publishing attribution results.
- Domain Experts Are Feature Engineers: The best predictive signals often live in the minds of experienced humans, not in the database. Interviewing domain experts before building models consistently surfaces features that pure data exploration misses.
- Fix the Foundation First: Analysis built on untrustworthy data is worse than no analysis—it produces confident wrong decisions. Investing in data quality and governance infrastructure before scaling analytics pays compounding returns.