I asked Keith Gooberman whether the industry needs another DSP. Most founders answer that question with a deck. He answered it with a confession about his own business model.
Keith Gooberman runs Pontiac, a demand-side platform he built without venture money, without cookies, and with a fee he says is several percentage points thinner than the big players. He is a mechanical engineer out of Union College who came up as a media trader. He talks about bidders the way other people talk about plumbing, and he means that as a compliment to plumbing.
Does the market need another DSP? He started with the size of the prize: the fees alone, he says, are a business worth hundreds of millions of dollars to the companies at the top. “There's definitely room. There's room for competition. There's room for people with new ideas doing things differently.”
Then he said the thing nobody pitching a DSP says out loud. Raising money to build one has dried up, and he understands why. “It is hard to raise money for, because the model is difficult to get right.”
And then, later, describing his own model of single-digit fees and a long wait to profitability: “If I went into a VC room and I told them the plan, they would tell me it's basically not going to work. And they'd be right, unless we use the power of time and a couple of balls bounced our way. And I think that's what's happened now.”
That is the most honest sentence a DSP founder has given me. The plan works only if you are patient and a little lucky, and the only reason he can afford patience is that nobody gave him money that comes with a clock.
The battleship problem
His case against the incumbents is technical, and it is worth understanding, because it is the whole reason Pontiac can charge less.
The legacy DSPs were built around cookies. When a DSP sees a user's first impression from an SSP, he explains, it often does not bid at all. It uses that impression to drop the SSP's cookie ID into a match table so it knows who the user is. The second time, it bids. And before the bidder can respond at all, it has to look up which audience segments that user falls into, so it knows which campaigns want them.
“Those lookups, this is a tremendous amount of computing power. Every bid costs the DSP a sizable amount of computing power.” To make remarketing work, you need to see something like 30 million ads a second, around the world, at all times. “These things make the maintenance of the infrastructure, the core bidder, very expensive.”
Now take the cookies out. Bid only on deals the SSP has already shaped. His example: a sports deal targeted to greater Minneapolis, running 6 p.m. to 10 p.m. “The SSP can send that, but you still need a DSP to upload the creatives, to pace, to be able to buy, maybe some optimization.” A deal like that might be under 2,000 queries per second. You are not running a battleship anymore. You are running the paddle at the auction.
That is his bet on curation. As more decisioning moves to the sell side, buyers want a lighter DSP. And a lighter DSP, with lower infrastructure costs, can take less of the dollar. “We're taking less of the dollar than the next guy, so publishers are like, yeah, this DSP is better than that one, because we make more of our money.”
The Coca-Cola question got a better answer than I expected. If The Trade Desk is Coke, Pontiac is not artisanal kombucha or bodega knockoff cola. “We're a little bit more of the flavored seltzer water.” Same carbonation, fewer calories. But he was clear that to get Disney, NBC and the other big CTV sellers to plug in, the seltzer still has to be bottled like a full soda brand. “We have to be super legit, super clean. We can't be doing anything trying to steal audience data, like God forbid.”

