27 Oct 2026 · 9 min read
In-House or Agency: Who Should Run Your AI Buying Agents?
Should brands run AI buying agents in-house or through an agency? For most brands the answer is a hybrid: the brand owns the agent's data, contracts, mandate and approval rights, then chooses whether its own team or an agency operates the agent day to day.
Fully in-house operation suits brands with proprietary sales data, programmatic talent and a short approval chain. Agency operation suits brands that need multi-market coverage, specialist trading skills and the negotiating weight of a large client roster.
TL;DR. The useful question is not "in-house or agency" but who owns the agent, who operates it and who answers for each deal it makes. In-house agencies are now standard at large advertisers, but programmatic media has moved more slowly than creative, so many brands will start with an agency operating an agent the brand controls. Whichever model is chosen, the mandate, the approval rights and an independent record of every agreed deal should stay with the brand.
Why is "in-house or agency" the wrong first question?
The choice looks binary because media buying has historically been bundled: whoever held the trading seat also held the data, the contracts and the relationship. AI buying agents loosen that bundle. An agent is software that works from a mandate, acts inside it and hands anything outside it back to a person, so the parts of the job can sit with different parties.
A more useful test separates three roles, which this piece calls Own, Operate, Answer. Own covers the data that feeds the agent, the technology contracts and the mandate it works to. Operate covers configuring, monitoring and tuning the agent each day, and Answer covers who approves each deal and who can produce the record of what was agreed.
Hershey's programmatic chief made the ownership case in a Digiday interview in May 2026. "To make it differentiated for a company like Hershey, inherently, we should own it. I say that because it's our data fueling it," said Vinny Rinaldi, vice president of consumer connections at Hershey.
On the agency relationship, Rinaldi added: "The agency just sits in our buying seats in our ecosystem, but they are still our agents-of-record."
Teams can rehearse the split before committing budget. Alkimi, an enterprise platform for agentic advertising, offers one worked example: its Gym lets buyers and sellers build their own agents, set the rules those agents work to and spar them against counterparty agents before any real money moves. A brand could run the same brief through an agent its own team configures and one its agency configures, then see where each stops for human approval.
How far have brands already moved media buying in-house?
In-house agencies are mainstream at large advertisers. The ANA's 2023 survey found 82% of its members had an in-house agency, up from 78% in 2018, 58% in 2013 and 42% in 2008. Among major multinationals, a WFA and Observatory International survey of 45 companies found 66% had an in-house agency and a further 21% were considering one.
Media, and programmatic in particular, has moved more slowly than creative. A 2020 Forrester analysis found that 95% of in-house agencies offered creative services but only 19% offered programmatic media, and pointed to fierce competition for digital media and data talent. The ANA's 2023 benchmark put the share of US brands with an in-house media unit at 54%, as Digiday reported in February 2026.
The direction is towards blending, not replacing. Every respondent in the WFA survey still worked with external agencies, even though half wanted to take on digital media planning and buying tasks within three years. "It's not in-housing or [external agencies] but in-housing and [external agencies]," the ANA's Bill Duggan told Digiday when the 2023 report came out.
What are the three operating models for AI buying agents?
Agency-run: the agency builds and operates the agent
In this model the agency supplies the agent, runs it inside its own stack and reports results, while the brand sets objectives and approves. Holding companies are building for it: Digiday reported in September 2026 that WPP has built and is testing its own buyer agent and that Omnicom is doing the same. Dentsu is testing media-buying agents in the US through a partnership with Newton Research, according to Digiday in June 2026.
Duluth Trading Company is an example of a brand choosing this model. Its agency partner manages agentic bidding on its behalf, and director of marketing Ellie Uberto told Digiday: "We're comfortable with it because we know that our agency knows us very well."
In-house: the brand builds and operates the agent
Here the brand's own media team configures the agent, runs it and approves its work. That needs programmatic talent and infrastructure, and Digiday's February 2026 analysis noted that brands including Lego, Boots and Georgia-Pacific have been building internal programmatic expertise. In-house teams can also be deliberately cautious about autonomy: Bayer's senior director of digital media activation, Glenniss Richards, said AI was making her team "quicker, faster, certainly more agile" but was "not owning or controlling our campaigns", as reported by Digiday.
Hybrid: the brand owns, the agency operates
The brand holds the data, contracts and mandate, and the agency operates the agent inside the brand's accounts, answerable to the brand's approvers. Hershey's set-up is close to this shape: Rinaldi described a joint business plan between Hershey, Publicis and The Trade Desk, with Hershey owning the contracts. The hybrid keeps agency expertise and scale while making any later change of operator a handover rather than a rebuild.
How do in-house and agency models compare? A decision table
Score each factor for the brand in question. Answers mostly in the "Points to in-house" column suggest in-house operation, answers mostly in the "Points to agency" column suggest agency operation, and a split suggests the hybrid.
Factor | Points to in-house | Points to agency |
|---|---|---|
Data feeding the agent | Proprietary sales, CRM or retail data that an agency will not hold | Planning and audience data the agency already supplies |
Technology contracts | The brand already owns its DSP and data contracts | The agency holds the seats and contracts today |
Negotiating weight | One large budget and direct publisher relationships | The brand benefits from an agency negotiating for a whole client roster |
Talent | The brand can hire and keep programmatic and data science staff | Specialist staff are hard to recruit or retain |
Markets and channels | A few markets and channels | Many markets, languages and channels |
Approval chain | One or two named approvers inside the brand | Approvals already run through the agency with client sign-off |
Pace of change | The mandate changes often and follows sales decisions | Briefs are stable and plans run quarterly |
Organisational readiness | A large organisation already running agents alongside staff | A smaller organisation still early with agents |
On negotiating weight, Digiday's February 2026 analysis noted that agencies negotiate on behalf of entire rosters of clients, bringing a scale few in-house units can match. On readiness, IAB Europe research reported by Digiday found 86% of executives at companies with more than 500 staff ran their most advanced agentic systems with people and agents planning and executing work together, against 48% at smaller companies.
What should stay with the brand whichever model it chooses?
Four things should not move with the operator. The first is the mandate, the written limits on budget, price, brand safety and approvals that the agent works to (see how to write a media buying mandate). The second is approval rights, including the decision to move an agent through the five stages of earned autonomy: Observe, Recommend, Draft, Human-approved action and, only once trust is earned, Bounded automatic action.
The third is proof of authority. Sellers need to confirm that an agent is authorised by the brand it claims to represent, and the first standard to ask any operator about is IAB Tech Lab's Agentic Advertising Management Protocols (AAMP). The fourth is the deal record: the brand should be able to retrieve what each agent agreed without asking the operator, so that changing agency does not mean losing the history.
What should a brand ask an agency that runs agents for it?
Five questions settle most of the ownership risk before a contract is signed:
Who owns the agent's configuration, instructions and the data it learns from if the relationship ends?
Which decisions need brand sign-off, and at what thresholds?
Can the brand see the agent's decision log, not just a weekly summary?
Where is each agreed deal recorded, and can the brand retrieve it independently?
How do sellers check the agent's authority to act for the brand?
For a fuller pre-launch list, see 25 checks before an agency lets an AI buying agent go live. For the platform requirements that differ between agencies and brand-direct buyers, see agentic advertising for agencies vs brand-direct buyers.
Where does a neutral marketplace fit in the operating-model choice?
The operating model decides who runs the agent, but it should not decide who holds the record. Alkimi's Marketplace is built on that principle: it works within IAB Tech Lab's AAMP framework today, runs AAMP identity checks on connecting agents, is being built to be compatible with AdCP and leaves humans with approval rights at the deal level. Agreed terms go into a DealSheet, written at agreement, timestamped and held by both parties, which either side can retrieve independently, so a brand that moves from agency-run to hybrid or in-house keeps the full history of what its agents agreed.
Frequently asked questions
Should brands run AI buying agents in-house or through an agency?
Most brands are best served by a hybrid, owning the data, contracts, mandate and approval rights while choosing who operates the agent. Fully in-house operation suits brands with proprietary data, programmatic talent and a short approval chain.
What does an agency add when it runs AI buying agents?
An agency adds trading expertise, multi-market coverage and the negotiating weight of a large client roster. It also takes on the work of hiring and keeping specialist programmatic staff.
What should a brand never hand over to an agency running its agents?
The brand should keep the mandate, the approval rights and independent access to the record of every agreed deal. Keeping those three makes a change of agency a handover rather than a rebuild.
Can a brand move from an agency-run agent to an in-house one later?
Yes, provided it owns its technology contracts, data and mandate from the start. Without that ownership, a later change of operator means rebuilding the agent rather than handing it over.
Do smaller brands need an agency to use AI buying agents?
Not necessarily, but IAB Europe research found smaller organisations behind larger ones in running agents alongside staff. A smaller brand without programmatic staff will often get further with an agency-operated agent that works to a mandate the brand writes itself.