How a multi-market broadcast media enterprise is replacing a fragmented, decades-old advertising technology stack with a connected, partly-automated operating model — without losing local-market judgment.
When a company's advertising technology stack has grown into five or six disconnected systems over a decade, where do you automate first — and how do you tell the difference between a process problem AI can solve today and a platform problem no model can solve until a vendor changes?
The organization's advertising operations ran on a stack assembled over many years: a customer relationship management system recently replaced company-wide, a separate media sales and traffic-research system, a separate digital ad-trafficking platform, and a legacy data warehouse in the process of being decommissioned. Each system did its job in isolation. None of them talked to each other well, and the manual work of connecting them had quietly become the actual operating model.
That showed up in a few different ways at once: hundreds of thousands of advertising copy instructions processed manually every year; a compliance and content-review function straining to keep up with regulatory and licensing requirements; a digital order workflow that depended on a large offshore team for tasks a modern pipeline could largely handle; and an executive reporting process built around a static report nearing two hundred pages, refreshed too infrequently to catch problems while they were still small.
Some of this was a model-capability problem — work AI could take on immediately. Some of it was a platform problem — automation blocked not by what AI could do, but by a legacy system with no way for other software to talk to it. Treating both as the same problem would have wasted a year on the wrong fixes.
The engagement moved in parallel across the parts of the stack that were ready for automation now, while building the evidence base for the harder, more political decisions still ahead.
Documented the current-state technology stack end-to-end — CRM, media sales and traffic research, digital ad trafficking, and the legacy data warehouse behind it — alongside the manual workflow steps stitching them together. This produced a shared, accurate picture that had not existed in one place before.
For high-volume manual work like copy-instruction processing, the limiting factor was largely automatable format handling, not judgment — a model-capability problem. For other workflows, the limiting factor was a legacy platform with no external interface for other systems to connect to — a platform problem that automation alone could not solve.
Deployed machine-learning-assisted processing for copy instructions and automated compliance screening for regulatory and licensing review, while keeping a human reviewer in place for edge cases — political disclosures, licensing-restricted content, and anything the model flagged as uncertain rather than resolving it silently.
Redesigned a fully manual, offshore-staffed digital order workflow into a semi-automated pipeline, moving quality control earlier in the process — closer to where data is entered rather than after the person who entered it has moved on — instead of only after the fact.
Quantified inventory that was going unvalued against the organization's revenue base and identified how much inventory was drifting into low-value remnant categories by default — turning an operational modernization project into a revenue conversation the organization's leadership could act on directly.
This is a live, active engagement. The figures below reflect validated pilot results and current-state findings rather than a completed enterprise-wide rollout.
| Workstream | Result to date |
|---|---|
| Copy instruction automation | 72% of a high-volume annual workload automated via machine learning, piloted across three markets; the remaining share is blocked by a source-system format limitation, not model capability. |
| Compliance & content review | Reduced required review staffing by roughly 70% in the pilot market, with human review retained for regulatory and licensing edge cases. |
| Digital order workflow | Modernized from a fully manual, offshore-staffed process into a semi-automated pipeline, targeting roughly a 10x efficiency gain. |
| Revenue & inventory visibility | Identified a significant pool of advertising inventory going unvalued against the organization's revenue base, with a majority of inventory drifting into low-value remnant categories by default. |
| Executive reporting | Began replacing a static, nearly 200-page legacy report with a near-real-time sales-intelligence platform, cutting data latency from a daily batch cycle to roughly an hour. |
Large organizations rarely have one advertising-technology problem — they have several, layered on top of each other over years of vendor decisions and local autonomy. The fastest path to value is not picking the flashiest automation opportunity; it's sorting the portfolio by what's blocked by model capability versus what's blocked by a platform, and sequencing accordingly.
Not every automation blocker is a reason to wait for AI to get better. Some are a reason to have a different conversation — with a vendor, with an executive sponsor, or with the people who built the workaround in the first place.
Let's figure out which blockers are AI problems and which ones aren't.