Omnicom has spent the year telling clients that its AI-powered Omni platform is the keystone of the merged Omnicom–IPG offering, showcasing the rebuilt system at CES in January. In June and July, Adweek reported in a 20 August exclusive, it quietly transferred large portions of the teams that built it to an outside contractor. At least 468 staffers moved to technology services firm Endava across the US, UK, India and Malaysia, most of them product and engineering talent behind Omni. Separately, roughly 50 US employees of the Omni Platforms product and engineering division received layoff notices on 9 June.
Omnicom’s framing, via a spokesperson, is a multiyear partnership with Endava designed to increase engineering capacity and accelerate delivery, with the platform itself, client relationships, data science and AI intellectual property staying in-house. The context makes the economics legible. Since closing the IPG deal, Omnicom now targets about $1.5 billion in merger savings over 30 months, double the original $750 million ambition. Around 4,000 job cuts were announced when the deal closed, with outsourcing and offshoring named among the savings levers. Moving the people who build your flagship platform onto a contractor’s payroll is that lever, pulled.
None of this is scandalous on its own terms — technology outsourcing is how most large enterprises run engineering, and Endava is a substantial, publicly listed firm that does this for banks and insurers. The tension is in the marketing. Omni is sold to clients as proprietary, differentiating, strategic; the hands that build it are now, in large part, a vendor’s. Publishers should resist the temptation to enjoy an agency holding company’s awkward week and instead read the move for what it says about the buy side they sell into.
01What the outsourcing says about where holdco money is going
The transfer resolves a question the industry has been circling since the merger closed: is the holdco AI story an investment programme or a margin programme? The honest answer this move suggests is both, in sequence — and margin is winning the near term. A company treating its platform engineering as a durable competitive weapon typically concentrates and retains that talent; a company treating it as a cost line to be delivered at contractor economics makes exactly this move. Omnicom would say capability is unchanged and delivery will accelerate, and that may prove true. But the signal to the market is that the differentiation is meant to live in the data, the IP and the client relationship — not in the engineering organisation. The build is becoming a commodity input.
For publishers, the relevant chain runs like this. Holdcos are consolidating spend decisions into automated platforms — Omni now sits on Acxiom identity data and first-party assets and increasingly mediates targeting, planning and measurement for a vast share of brand budgets. Those platforms are simultaneously being engineered for cost. A buying system maintained at contractor cost inside a $1.5 billion savings programme is a system tuned for efficiency metrics. It favours automatable supply paths, standardised measurement and algorithmic allocation. The human planner who understood why a premium publisher’s context was worth a manual line on the plan is precisely the kind of cost such programmes remove. Publishers who cannot be found, evaluated and bought by the platform on the platform’s terms will increasingly not be bought at all.
There is also a stability question worth logging. Agency platform roadmaps now depend on a three-party arrangement — holdco strategy, contractor delivery, merger-integration economics — during the exact period when agentic buying, curation and AI-driven planning are being wired into these systems. Publishers doing bespoke integrations with holdco platforms should expect slower exception-handling and less appetite for custom work while that settles.
02Why this matters for publishers
| The buy side is industrialising, and this is what industrialisation looks like | Spend decisions are moving into platforms engineered at contractor economics. Inventory that requires human judgement to appreciate is inventory the cost programme is designed to stop paying for. |
|---|---|
| Holdco AI capability is thinner than the CES demo implied | The proprietary-platform story now runs on outsourced engineering. Publishers should weight what these platforms demonstrably do today over roadmap claims when deciding how much integration effort to spend on any one holdco stack. |
| Merger economics will shape buying behaviour for 30 months | A $1.5 billion savings target is not abstract: it reaches media supply paths, fees and vendor rationalisation. Expect pressure toward consolidation of partners — fewer, bigger, more automatable relationships — across every holdco conversation. |
| The precedent travels | If the largest holding company can hand its platform engineering to a contractor mid-AI-race, so can the others — and so can the ad tech vendors publishers depend on. "Who actually builds and maintains this?" is now a fair diligence question in any platform relationship you rely on for revenue. |
The Adweek story reads as an embarrassment about marketing versus reality, and it is partly that.
03What publishers should do
04The bottom line
The Adweek story reads as an embarrassment about marketing versus reality, and it is partly that. But the durable lesson for publishers is structural: the buy side is being rebuilt as industrial software, delivered at contractor cost, inside the largest savings programme the agency world has run. That software is where publisher revenue decisions increasingly get made. Sympathy or schadenfreude are both wasted responses. The useful one is to meet the machine on its terms — legible, well-signalled, packaged inventory — while building the direct brand relationships that do not route through a holdco cost programme at all. The agencies are deciding what they are willing to stop owning. Publishers should be equally deliberate about what they refuse to outsource: the audience relationship, the data, and the evidence of their own value.