Why most AI initiatives in media operations fail, and what actually works

This article is a submission by Alloyed, a global managed business solutions provider built on two decades of operational experience. Alloyed delivers what it calls cognitive process orchestration, fusing people, process, and fit-for-purpose technology to produce predictable, measurable outcomes for growing businesses, with delivery centers in Bangladesh, El Salvador, and the Philippines.

Measurable outcomes

• $10M saved per year for a nonprofit health system
• 4,000+ hours saved automating claim status
• 100% reduction in brand safety flags for a media client

Every media finance leader is dealing with some version of the same demand. Modernise the operation. Do it without disrupting the business. Do it now. 

Most are not short on tools. There is no shortage of automation vendors, and no shortage of AI platforms promising to fix the problem. What shows up instead, again and again, is that technology has been layered onto operations that were never fully understood in the first place. 

The result is a graveyard of pilot projects that never made it to production, and in some cases, AI initiatives that actually made the numbers worse.

A recent town hall webinar with the Media Financial Management Association examined why that keeps happening and what a genuinely different approach looks like. This is that argument, laid out.

The strain is real, and it is not going away on its own

Ask any media operations team what is eating their week and you will hear the same list. 

Campaign volume keeps climbing. The platform landscape keeps fragmenting. Manual processes that worked fine at half the scale are now the bottleneck. Compliance pressure has not let up. 

And the people doing the work are burned out, doing repetitive tasks they are overqualified for, while the strategic work they actually want to do sits untouched.

When webinar attendees were polled live, the answers confirmed what comes up in every client conversation. 

The top strain was split almost evenly between cost pressure and the difficulty of actually implementing AI. Not a lack of desire to modernise. A lack of a workable path to get there.

A second poll made the gap even clearer. Most organisations in the room had pilot projects running somewhere in the business. Almost none of them had made the jump to automation running in production at real scale. 

Pilots tend to happen department by department, each team experimenting on its own, which feels like progress until you try to stitch it into something enterprise grade and realise there is no foundation underneath it.

The mistake is treating this as a technology problem

Here is the pattern that shows up most often. A team gets pressure to implement AI. They pick a tool. They point it at a process. The process was never fully mapped, the data was never cleaned, and nobody agreed in advance what success looked like. 

Six months later, the initiative is quietly shelved, and the organisation concludes that AI does not work for their business. 

And many organisations have spent tens, if not hundreds of thousands of dollars, and hundreds of resource hours pulled away from more strategic initiatives, to learn this very lesson.

It is not that the AI failed. It is that the foundation was never there to support it. You would not hand someone a shovel and call that a strategy for digging a hole. The tool is not the plan. 

Understanding the work, cleaning it up, and only then automating the right pieces of it, that is the plan.

This is also why the people versus AI framing is the wrong one. The organisations getting real value are not choosing between headcount and automation. They are pairing the two deliberately. 

AI brings consistency and speed: accurate data ingestion across platforms, automated anomaly detection, compliance checks that never get tired or distracted. People bring the judgment AI cannot fake: reviewing the exceptions AI flags, applying context that is specific to a client or a vertical, making the final call before anything goes live. 

Neither replaces the other. Removing either one breaks the model.

A different way in: ingest, then analyse, then optimise

The approach that works starts somewhere most automation projects skip entirely. Before anything gets automated, ingest the work itself. 

That means documenting the workflow as it actually runs today, not as the org chart says it should run, integrating the data and systems involved, defining what the SLAs actually are, and training a team to do the work well before any technology touches it.

Only once that foundation exists does the work move to analysis. This is where every bottleneck gets mapped, the actual work gets shadowed, and the team gets honest about where the friction is. 

It is tempting to skip straight to automating something that looks broken, but a process nobody has fully mapped will produce automation that is broken in new and more expensive ways.

Optimisation comes last, not first. That is where the agentic AI, the bots, the integrations, and the API connections actually get built, tested against real business rules, and rolled out with documentation and training that keeps humans in command. 

It is a sequence, not a menu you pick items from.

Skip a step in that sequence and this is exactly where the negative return on investment that so many organisations are quietly experiencing comes from.

What this looks like when it works

A media company came to the table with a familiar set of problems. Headcount costs were climbing and the team was burning out. Make goods and revenue leakage were eating margin. Reporting was inconsistent enough that partner coordination had become its own full-time job. 

The front line of social engagement was offshored to an expert team, which produced savings immediately, and that breathing room was used to analyse and rebuild the media planning process with faster, more predictive insights. 

Brand safety flags dropped, the team reclaimed hours each month by catching discrepancies before they became errors, and campaign and billing reporting has stayed on time ever since.

The same pattern has played out at a very different scale. A large nonprofit health system wanted to move off manual processes to bring costs down. 

Rather than starting with automation, it started by assessing the actual work, building the business case, and proving the concept before scaling anything. 

The result was a standardised roadmap and a steady rollout of automations across the enterprise, with the savings following the proof rather than the hype.

At a national health insurer, the problem was narrower but no less painful: tens of thousands of claim lines a cycle being validated and entered by hand, with the errors and delays that always come with that much manual data entry. 

A pair of workflow automations later, the team had saved thousands of hours, cut data-entry errors sharply, and sped up the billing cycle. Employee satisfaction on that team rose too, which is worth pausing on. 

The people doing the work were relieved to stop doing the part of it that never should have required a human in the first place.

Scale it in tiers, not in one leap

None of this requires a single enormous automation project. Think about it in three tiers, and do not skip any of them.

Tier one is optimisation: getting more out of the tools already sitting in the tech stack, such as Excel, M365 and Google Workspace, and layering in RPA to automate what already exists, before buying or building anything new. 

Tier two brings in AI platforms and large language models to add intelligence on top of those workflows. 

Tier three is agentic AI, genuinely end-to-end workflows built with a trusted ecosystem of partners once the foundation underneath it is solid.

The organisations that jump straight to tier three without doing tier one and two first are usually the ones struggling six months later. The tiers exist because the foundation actually matters.

What comes next for AI-enabled operations

The end state worth building toward is not a fully automated operation with no people in it. It is AI-augmented hybrid teams: people around the world made significantly more capable by the AI tools they use, continuously improving the process instead of waiting for the next transformation mandate from the board.

Proven ROI is what unlocks everything else. When webinar attendees were asked what would give them the confidence to move forward with AI-enabled operations, proven business outcomes topped the list, ahead of faster implementation, lower risk, or even executive support. 

That is not a coincidence. Executive support tends to follow proof, not the other way around. Which is exactly why the business case and a real proof of concept should come before scaling anything.

The organisations that come out ahead in the next few years will not be the ones that automated the most tasks the fastest. They will be the ones that took the time to actually understand their operations first, and built AI into that understanding rather than around it.