Programmatic is not plug and play. Open the black box and you gain control. But only if you measure it and keep humans in the loop.

Berg learned the complexity first-hand selling DSP technology for AdForm, where even then the tools were rarely used the way they were intended, because the background is so technology driven. Programmatic grew out of display, the cheap remnant traffic, and became the channel everyone went to. From there, more and more channels were connected through the DSPs, each with its own buying process: open auctions, private deals, private auctions, programmatic guarantees. The result is a system that only becomes more complex over time, where a company barely knows where to start to understand it.

Ten years ago, when MINT Square started, transparency meant pricing and the business models behind trading desks. Today advertisers understand programmatic far better and their expectations have moved. They care less about the technologies and more about the entire supply path: where the money flows across every step and vendor, and which inventory is bought in what quality. Transparency now means knowing which SSPs, targeting partners and verification partners are in play, so a brand knows where its ad is shown and that it is safe.

MINT Square tackles the mid-market, where Berg says more than 80 percent of clients are independent agencies. They usually start with Google Ads and Facebook, then expect programmatic to deliver the same output as the walled gardens. Inside that framework it is very hard to justify programmatic, so it takes patience and management awareness that you have to take the time, dig deeper and really understand it.

Berg insists there is no bad DSP, only whether it tackles the problem a client faces. Inventory is the first filter: MINT Square runs Active Agent, Adform and The Trade Desk, and a client wanting Netflix has to go to The Trade Desk. After inventory it is API connections and reporting integration that decide between platforms. Trying to run several ad techs at once simply multiplies the complexity.

Choosing the DSP is the end of the beginning. You cannot assume a new platform behaves like the last, and the expectation that performance climbs the moment you switch it on is false. Allocate budget to test it inside out, then focus on four levels: translating media planning into dynamic programmatic, understanding how the platform is built, running campaigns in parallel to compare, and the reporting. Instead of a forgettable two-day workshop, MINT Square takes clients into the boat, builds a blueprint and lets them shadow the work, so after about three months they genuinely know whether it works for them.

A typical black box is a walled garden, and Berg is clear it has advantages. It is comfortable, often managed for free, and all about the output. If you are happy putting money in and getting what you want, it is a fair deal with no wrong expectations. But if you want to know where the ad is shown, what targeting vendor is used and whether the data is first-party or second-party, there is often no answer. For a holistic understanding of the ecosystem, you have to open up to the open internet.

A lot of the work is education. The hard part is explaining the pricing to people new to programmatic: technology fees, data providers and verification partners all take a CPM, so a high share of the media goes to technology and only a fraction reaches the publisher. With a walled garden you never see how much they earn, the money just disappears into the platform. Opening up means many players are involved, so teams need a sense of the added value before they will bring it into the business.

Programmatic is not setting five parameters and letting it run. Performance needs constant optimisation circles, several times a week, which is unfamiliar to traders used to briefing an insertion order. The real challenge is a clear overview of every platform capability: choosing the right SSP, the right DMP and targeting partners, connecting and matching first-party data. It needs real situations, not just theory, which is where Berg's teams support traders from onboarding through to reporting.

The biggest gap in teams is expectation. People assume you switch the campaign on and, with AI, the machine does everything. Berg says we are not there yet. The algorithms handle a lot, but the human aspect is so important and you have to intervene. You cannot let a machine run clients' budgets unsupervised, because the complexity does not justify it. You need control of spend, pacing and whether the audience is actually being reached.

The role has moved from media planning to something far more technical, so Berg wants people unafraid of numbers and analytics who can interpret results. He uses two LLMs, with Anthropic now a major one, not to substitute people but to help them optimise the day-to-day and free up headspace from repetitive tasks like reporting. What surprises him is how hesitant even young people are to use these tools beyond simple questions. He wants people brave enough to experiment with the new SSPs, DMPs and verification partners that appear almost every month.

When a client says programmatic is not working, it is usually measurement. Advertisers max out Google Ads, social and influencers, hit a threshold where everything gets more expensive, and turn to programmatic for an answer. But it needs attribution modelling integrated across campaigns. Google Ads catches users in a last-click mindset, so clicks lead to purchases. Programmatic is a lean-back medium that reaches audiences through identifiers across the journey, so success shows up in post-view conversions that only proper tracking reveals.

Having worked in publisher sales and SSP relationships, Berg sees what buyers often miss: the publisher's aim is the highest revenue and most of the bid, and the more relevant the placement, the more they earn. Publishers now have a whole portfolio of ways to connect inventory, and the same media can be bought in different quality and position. Understanding the SSP side and holding direct relationships with sales houses unlocks deals you would never find by just clicking in the DSP.

Berg is excited about AI, though he cautions it is easy to frame AI when algorithms have existed for a long time and are simply used differently now. We are not yet at the point where you allocate a budget and it works, there are liabilities and that is too simple. But he believes trading itself will be automated, and then the value moves to relationship management with vendors, finding the best deals and the right partners rather than spending time on the trade. Not by the end of the year, but within five years he expects a solution.