In-house programmatic operating model: three models, roles, salaries and tooling

The models below are the ones we see in public sources and in contracts. Names vary; the ownership pattern is what matters.

The ANA's 2017 study described the middle path well. It found most marketers "prefer to set the strategy for their programmatic campaigns and own the contracts", while agencies "still play an important role in campaign execution" (ANA, 18 December 2017). Nine years later, the same pattern shows up in the 26% hybrid share.

In a full in-house model, the advertiser signs the DSP contracts, hires the traders and analysts, and runs every line itself. Agencies may still help with creative or research, but not with buying.

The examples are real but few. UK retailer Boots has a 27-person team that already handles programmatic buying, as well as spending on CTV, affiliate PPC and audio. Its head of performance media said the team needed to "match the cadence of the business", since a retailer "trades each day" (Digiday, 4 February 2026). Dollar Shave Club manages its entire media operation in-house with four people, according to the same report.

Full in-housing fits when buying decisions depend on data only the advertiser has. A retailer adjusting spend to daily stock levels is the classic case. It also fits when the advertiser wants bid-level evidence of every dollar, which is easier when the seat and the log-level data are its own. The fee side of that evidence is covered in DSP pricing explained.

The costs are people and scale. Georgia-Pacific's senior director of digital media named "recruiting, developing and retaining specialized talent" as one of the biggest challenges of building its team (Digiday, 9 February 2026). And an in-house team negotiates for one brand, while an agency negotiates for its whole client roster.

In a hybrid model, the advertiser brings some programmatic work inside and leaves the rest with an agency. The split usually follows channel, market or complexity. A common pattern is in-house trading for always-on performance lines in the home market, with the agency handling national TV, CTV upfronts and smaller markets.

Georgia-Pacific shows the shape. It began in-housing in 2019 and now handles all its media investment in-house except TV, which stays with OMD (Digiday, 4 February 2026). Lego is building toward a similar mix. It advertised five programmatic roles in January 2026 in the US and Denmark, while Publicis Media remains its "key strategic partner" and Publicis One handles its global media account, according to the same report.

Hybrid models keep agency scale where scale matters. Digiday notes that agencies negotiate "on behalf of entire rosters of clients", bringing a heft few in-house units can match (Digiday, 9 February 2026). The risk is a blurred line of responsibility. If both teams can touch the same seat, write down who owns which line items, which supply paths and which reports. The D4 analysis of what Netflix, Lego and P&G built shows how three companies drew that line differently.

In the third model, the advertiser owns the contracts and seats, and the agency operates them. The agency is paid a disclosed fee for its work, not a margin on media. The advertiser can change agencies without losing its data, its audiences or its deal history.

This is the ANA's 2017 recommendation in practice: own the DSP or DMP contract and "give their agency access to the platform to run the campaigns" (ANA, 18 December 2017). The same study tracked the opposite arrangement, the undisclosed model, in which the agency buys media and reports only the final price. It fell from 34% of respondents in 2016 to 19% in 2017.

Seat ownership has practical wrinkles. In Amazon's restructured platform, an agency chief executive advised brands to own their seat "sooner rather than later". PPC Land noted that under Amazon's 29 September 2026 terms, agencies with existing access got advanced DVA+ controls by default. New advertisers got only the automated tier and had to ask for more (PPC Land, 29 October 2026). A brand that moves its seat in-house may start with fewer controls than its agency had. Check access tiers before migration, and see Amazon DSP cost for the fee side.

Model 3 is the lowest-risk first step for most advertisers. It changes ownership before it changes people. If the advertiser later hires traders, they inherit a seat that already holds the history. The audit steps in programmatic agency audit show what to check before the switch.

Programmatic teams are built from a small set of roles. The table shows which ones each model needs inside the company, which can be shared with an agency, and which the agency usually supplies. It is a planning guide, not a benchmark.

Smaller teams combine roles. The ANA's 2017 study found most respondents had one person working full-time on programmatic and three more who worked on it alongside other jobs (ANA, 18 December 2017). That is a light model 3 team. A full in-house team needs dedicated traders and at least one engineer who owns the data.

No official statistic tracks "programmatic trader" as an occupation. The US Bureau of Labor Statistics publishes wages for broader occupations that contain these roles. The table maps each role to its closest occupation and shows the 25th percentile, median and 75th percentile of US annual wages in BLS 2025 data, as published on O*NET OnLine.

Treat these as ranges for broad occupations, not quotes for a specific job. They are national US figures and cover every industry. Job-specific surveys exist but are thin. Payscale, for example, reported an average of $67,852 for ad operations managers in 2026, based on only 17 salary profiles last updated on 1 September 2026 (Payscale, accessed 9 November 2026). Pay in New York or San Francisco runs higher than the national median, and we found no comparable public source for UK or EU programmatic roles (n/d). Retail media roles are covered separately in retail media jobs and salaries.

Take a minimal full in-house team of five: one head of programmatic, two traders, one data engineer and one measurement analyst. At the BLS 2025 medians above, base pay comes to $133,660 + 2 × $78,760 + $135,980 + $120,230 = $547,390 a year (O*NET, BLS 2025). That excludes benefits, payroll taxes, recruiting, tools and training, which you must add from your own HR figures.

Now divide by the programmatic budget to see how much the team must improve results just to pay for itself.

Where could that come from? The ANA's Q2 2026 benchmark shows how wide the gap is between careful and careless buyers. Higher-performing advertisers turned 52.3% of programmatic spend into quality impressions, against 31.1% for lower performers, a gap of 21.2 points (Fiducia on the ANA benchmark, 6 August 2026). The higher performers also bought across about 19,000 fewer domains and apps. The definitions behind those numbers are explained in working media vs non-working media.

The arithmetic gives a rule of thumb, not a verdict. At $50 million, a team that lifts working media by one point covers its base pay. At $5 million, it would need to lift working media by about 11 points, which is a large share of the whole performance gap. Smaller budgets therefore lean toward model 3, where you gain ownership without carrying a full payroll.

Every model uses the same tool categories. The difference is whose name is on each contract. Our recommendation, reflected in the last column, follows the ANA's advice to keep contracts and data with the advertiser.

Fees differ by vendor and contract. Platform fee structures are compared in DSP pricing explained and, for the largest independent DSP, in The Trade Desk fees. Before signing, run a structured comparison with the DSP RFP template, and make sure every contract grants access to log-level data.

The operating model lives in contracts, not in org charts. Whichever model you pick, five clauses decide whether it works as intended. They matter most in models 2 and 3, where an agency still touches the seat.

These clauses follow directly from the ANA's long-standing advice that marketers should own contracts and understand the trade-offs of undisclosed buying (ANA, 18 December 2017). Ask your legal team to check them against local law and your existing master services agreement.

A clear rhythm stops the most common hybrid failure, two teams changing the same line item. Set it out in one page and review it every quarter.

• Daily: traders manage pacing, bids and supply on the lines they own. Nobody edits another team's lines.
• Weekly: the head of programmatic reviews spend, cost per outcome and supply quality across all seats, in-house and agency.
• Monthly: finance reconciles invoices against platform logs. Differences go back to the vendor or agency in writing.
• Quarterly: leadership reviews the split of work between in-house and agency, and moves lines if the evidence supports it.

The rhythm also tells you when the model is wrong. If the in-house team spends most of its week on trafficking and tags rather than on decisions, it is understaffed for full in-housing, and a move back toward model 3 may be cheaper.

Use the table to see where your situation points. If most of your answers land in one column, start there. If they split, model 3 is the safe default, because it keeps your options open.

One factor overrides the rest: contract ownership. Whatever model you choose, hold the DSP contract and the data in your own name if you can. That is the step every source on this page agrees on, and it is the one that makes the other models reversible. The in-house programmatic hub links the related decision pages.

Transitions fail when ownership, people and data change all at once. Sequence them.

1. Audit the current set-up. List every seat, contract, fee and data feed, and who owns each one. The programmatic agency audit gives a checklist.
2. Move the contracts first. Put DSP and data contracts in the advertiser's name. This alone moves you to model 3.
3. Secure the data. Turn on log-level feeds into your own warehouse and keep them for at least one full planning cycle.
4. Check access tiers. Confirm that the new advertiser seat has the same controls the agency had, as the Amazon example shows (PPC Land, 29 October 2026).
5. Hire the core before the traders. A head of programmatic and a data engineer first; traders once the data flows.
6. Run in parallel. Let the in-house team take a defined slice of budget while the agency keeps the rest, and compare cost per outcome.
7. Hand over in stages. Move lines in groups, with a written split of responsibility for anything still shared.

Timing depends on hiring and contract terms, and we found no reliable public benchmark for how long a transition takes (n/d). Plan at least one full quarter of parallel running, so seasonal effects do not decide the comparison.

Agentic buying is now part of the in-housing debate. Digiday reported in February 2026 that CMOs see it as a way to "cut down agency fees or cut agencies out entirely". The same piece called agentic media buying "for now, an idea in utero" (Digiday, 9 February 2026). By April, Omnicom said it had executed live agent-to-agent buys for several clients (company claim) (Digiday, 29 April 2026).

Agents may shrink the trading workload, which helps small in-house teams. They do not change the ownership question. An agent working inside an agency's seat produces the agency's data. An agent working inside your seat produces yours. What is actually live, and how to test the claims, is covered in AI agents for media buying.

1. Association of National Advertisers, "Marketers Expanding In-House Capabilities For Programmatic Buying: ANA Study" (18 December 2017). https://www.ana.net/content/show/id/pr-2017-expanding-inhouse
2. Association of National Advertisers, "The Resilient Rise of the In-House Agency: 2026 State of In-Housing Report" (June 2026; summary page). https://www.ana.net/miccontent/show/id/rr-2026-06-resilient-rise-in-house
3. Digiday, "Lego is building out an in-house programmatic team" (4 February 2026). https://digiday.com/media-buying/lego-is-building-out-an-in-house-programmatic-team/
4. Digiday, "The case for and against bringing programmatic in-house" (9 February 2026). https://digiday.com/media-buying/the-case-for-and-against-bringing-programmatic-in-house/
5. Fiducia, "ANA Programmatic Transparency Benchmark: Q2 2026" (6 August 2026). https://www.fiducia.eco/post/ana-programmatic-transparency-benchmark-q2-2026-working-media-efficiency-reaches-a-record-45
6. O*NET OnLine, National wages, 11-2011.00 Advertising and Promotions Managers (BLS 2025 wage data, accessed 9 November 2026). https://www.onetonline.org/link/localwages/11-2011.00
7. O*NET OnLine, National wages, 11-2021.00 Marketing Managers (BLS 2025 wage data). https://www.onetonline.org/link/localwages/11-2021.00
8. O*NET OnLine, National wages, 13-1161.00 Market Research Analysts and Marketing Specialists (BLS 2025 wage data). https://www.onetonline.org/link/localwages/13-1161.00
9. O*NET OnLine, National wages, 15-2051.00 Data Scientists (BLS 2025 wage data). https://www.onetonline.org/link/localwages/15-2051.00
10. O*NET OnLine, National wages, 15-1252.00 Software Developers (BLS 2025 wage data). https://www.onetonline.org/link/localwages/15-1252.00