The Exported Mind

In January, Omnicom stood on a stage in Las Vegas. Omni, their revamped, AI-powered platform was the centrepiece of the new Omnicom's story at CES as the connective layer across strategy, creative, media, commerce and measurement. The reason to believe in the merger.

By July, reportedly most of the people who built it worked for someone else.

Eight months ago, on the day the Omnicom-IPG deal closed, I wrote that the merger had built "a bigger horse as the world upgrades to tanks." Last month, in The Standing Gallop, I published a field guide to the three mechanisms that made a flat company report 6.1% organic growth. I thought the machinery was fully mapped. Then, on Friday, Adweek found a fourth mechanism and something far more important hiding underneath it.

According to documentation obtained by Kendra Barnett at Adweek, Omnicom transferred at least 468 employees (most of them the product and engineering staff who built and operate Omni) to Endava, a third-party technology services contractor, in moves executed quietly across June and July 2026. Roughly 50 further staff were laid off. The transfers span the US, UK, India and Malaysia. There was no announcement. The story surfaced through obtained documents, five to six months after the CES showcase, and some of the transferred employees describe a chaotic transition and question the strategy of separating a keystone platform from the people who develop and support it.

Omnicom, on the record, calls it a multiyear partnership "designed to increase engineering capacity and accelerate delivery," and says it retains ownership of the Omni product, its AI innovation, data science, client relationships, and intellectual property.

One honesty note before we go further. This entire fact base traces to a single investigation. It is documentation-based, it carries on-record company statements, and its numbers are far too specific to be speculation, which is why I am writing on the basis that it is correct, and why I expect broader coverage and a fuller Omnicom response this week. And their clients are reading the same reporting I am.

So, in advance of that, here is what I predict that response will miss, because almost all of the early commentary has missed it.

The obvious take is outrage: the platform sold as the company's future, built by someone else's employees. I am not going to write that piece, because I think the decision itself is correct.

Endava is not a body shop. It is a NYSE-listed digital engineering firm with global delivery centres and a formal services partner of OpenAI. Every serious software company scales with delivery partners. WPP builds Open through partnerships with Google Cloud, Microsoft and Meta. Accenture has made an empire of building other people's technology. "We own the roadmap, partners build" is how most enterprise platforms on earth actually work.

More than that: Omnicom, like WPP (despite the recent Enterprise Solutions announcement), does not have the enterprise technology backbone to build products at this level themselves, and there is something genuinely grown-up about acknowledging it. Design the product. Own the roadmap. Optimise it for clients. Let specialists do the building, the hosting, the updating. I have argued for years that the holding companies should be honest about what they are and are not, and this, at least, is honest. If I advised WPP, I would tell them to consider the same move.

So if the outsourcing is defensible, and frankly orthodox, why does this story matter so much?

Because of what it reveals about the one thing you cannot outsource.

In 2015, Publicis completed the acquisition of Sapient for $3.7 billion. I was inside the company for what followed, and I will tell you what I believe that acquisition actually bought, because it was not, primarily, the engineers.

The engineering population that came with Sapient and was built out from it runs to some twenty thousand people and I have cited the public figure before. But headcount was never the asset. The asset was that an engineering mindset reached the leadership of the company.

Engineers think systemically. They think in products, not projects. They think about how you build something once and sell it many times; how revenue streams change shape around a platform; how a capability serves every client rather than being optimised, lovingly and unprofitably, for one. That way of thinking arriving at Management Committee level, arguing with the agency instincts in the room, sometimes losing and gradually winning is what turned Publicis into a structurally different company from every other holding group. Epsilon becoming CoreID rather than another agency P&L. CoreAI as an operating layer rather than a demo. Those were not engineering deliverables. They were leadership decisions that only a leadership shaped by engineers could take.

Omnicom outsourcing the build is fine. Companies with deep engineering leadership outsource builds every day, and the products get better. What worries me is the other half of the equation. The brief. The architecture of the product. The judgement about what Omni should become: what gets productised, what gets priced, what gets killed. That judgement has to live inside the company, at the top of the company, and it is exactly the capability the Omnicom-IPG merger conspicuously declined to buy. They bought more media. More creative. More of the past. I asked at the time why anyone would buy more of the services being automated instead of the capability to build what comes next. Friday's story is the answer arriving.

The Standing Gallop documented three mechanisms that let a standing horse report a gallop: draw the perimeter, report the gross number, let cost play the part of growth. File this one fourth: headcount that exits without a redundancy event.

Four hundred and sixty-eight people left Omnicom's employment. Around fifty registered as layoffs. The other four hundred-plus will never appear in a WARN filing aggregation, a layoff tracker, or a "jobs lost" headline. They were transferred, not cut. When I return to my cumulative job-reduction arithmetic in the December scorecard, remember this channel exists, because it means every public tally is an undercount.

And read the carve-out list again, the way an engineer would. Omnicom tells us what it kept: the product, the AI innovation, the data science, the client relationships, the IP. A company's list of retained crown jewels is a self-diagnosis and it tells you precisely what leadership believes the valuable capabilities are. The capability to build is not on the list. They have told us, in their own words, that they do not consider making things a core competence of the world's largest advertising company.

Now sit in the client's chair (like I do, though thankfully not an Omnicom client), because this is where the story stops being about Omnicom at all.

If you appointed Omnicom or chose to stay through the merger's disruption, partly on the strength of the Omni commitment, Friday's reporting deserves a place on your next agency-management agenda. Not because contractors can't build enterprise software. They demonstrably can. But because of the logic Omnicom itself has just validated.

The greatest threat to any media agency today is not another media agency. Not even close. It is the platforms (Google, Meta, Amazon, and a list that now includes OpenAI) building natural-language, client-facing tools that let advertisers create content and manage media directly. The agency's defence against that disintermediation was supposed to be the proprietary technology layer: the agnostic port that sits above the walled gardens and orchestrates across them. Omni was Omnicom's version of that defence.

And Omnicom has now shown everyone how that layer is actually made. You keep the strategy, the design authority, the relationships and the IP and you brief the building out to a specialist contractor. Which is a perfectly sound operating model. It is also, word for word, a model available to any client of scale. The same contractors will take the client's call. Endava's entire business is serving enterprises directly.

The honest counter is that orchestration is genuinely hard, that neutrality across platforms has value, and that most clients do not want to run a technology estate. All true. But the neutrality argument rests on the credibility of the technology behind it and last October I wrote that Omni was already "bolted together through acquisitions, generating no meaningful client lock-in." When the middle layer's own justification for outsourcing its platform is indistinguishable from a client's justification for skipping the middle layer, the premium gets harder to defend every quarter.

If this arrangement works, Omni's roadmap should visibly accelerate with faster releases, better product and demonstrable client value over the next two quarters. Endava is capable of exactly that. If it happens, the pragmatists were right, the partnership delivered what the press statement promised, and my mindset argument was romanticism. I will say so.

But I don't expect to. Not because the engineers left but because of who is left to brief them. My view since the day this merger was announced has not changed: a company that responds to the platform era by buying more of the agency era, and responds to the technology question by renting it, will not fail dramatically. It will do something quieter. It will continue to underperform against Publicis, against the platforms and against what its own scale should make possible until the gap becomes the story no perimeter can redraw.

The people on that Las Vegas stage in January were selling a platform as the future of the company. The people who built it now work somewhere else. Both of those things are true at once, and only one of them was announced.

You can export the builders. You cannot export the mindset that writes the brief.

Brandflow is written by Justin Billingsley, who has spent his career on all three sides of the industry's table: senior client, global agency leader, technology founder. First published 23 August 2026 in the Brandflow newsletter on LinkedIn.