Why every new site brings more office work
Every face a media owner sells sets off a chain of tasks. Someone has to confirm it is free, add it to a proposal, include it in a contract, order the print, send the installer, receive and check the posting photo, deal with any issues, invoice the agency, settle with the site owner and collect payment. At the end, someone prepares the Campaign report.
That work grows with the number of faces and Campaigns. When a media owner adds a hundred sites, it also multiplies the availability checks, the work orders, the photos to review and the settlements to calculate. That is why growth tends to stall in the office before it stalls on the street, because selling more faces takes more people to manage them.
AI agents change that relationship for part of the work. An agent is a program that receives a goal, consults the data and tools it has access to and carries out the necessary steps within the limits the company sets.
What research says about automating administrative work
McKinsey has been measuring automation potential by activity type for almost a decade. In 2016 it estimated that data collection and processing had a technical potential above 60%, compared with 18% for activities that require applying expert judgement to decisions and planning, and 9% for managing and developing people.2
Generative AI widened that potential. In 2023 the same firm calculated that, combined with other technologies, it could automate activities that take up 60% to 70% of employees' time today.3 In November 2025 the McKinsey Global Institute estimated that technologies already demonstrated could, in theory, automate activities equivalent to about 57% of US work hours: 44% would fall to agents and 13% to robots. The report presents the figure as the technical potential for change in what people do, warns against reading it as a forecast of job losses and stresses that people will remain vital to make these technologies work well.1
Much of a media owner's office work consists of collecting and processing data: availability, sizes, dates, photos and amounts. Negotiation and pricing decisions belong to the categories with the least potential.
A map of a media owner's tasks
We have sorted eleven operational tasks, plus three that depend on the sales team, into three groups. "Automatable now" means an agent can do the whole task with clear rules and reliable data. "With oversight" means the agent prepares the work and a person reviews and approves it before it goes out. "Stays with people" covers what depends on relationships and commercial judgement.
| Task | Group | What the agent does and what the person does |
|---|---|---|
| Inventory and availability | Automatable now | Keeps each face's availability up to date from the contracts and answers agency queries. |
| Production and printing | Automatable now | Prepares the print order with sizes, copies and delivery date for each Campaign, and follows it up. |
| Proof of posting | Automatable now | Checks each photo against the booked face, date and location, and flags anything that does not match. |
| Invoicing | Automatable now | Generates the invoice from what was delivered, with issues already deducted. |
| Reporting | Automatable now | Builds the close-out report with the Campaign's faces, dates and proof of posting. |
| Proposals | With oversight | Puts the proposal together with available faces and agreed prices. A person reviews it before it goes out. |
| Contracts | With oversight | Drafts the contract from the accepted proposal. A person with signing authority signs it. |
| Posting and installers | With oversight | Groups faces by area and proposes routes and work orders. The operations manager approves them. |
| Issues | With oversight | Spots the issue and proposes a fix or a credit. A person decides. |
| Site owner settlements | With oversight | Calculates what each site owner is due under their contract. Finance approves it. |
| Collections | With oversight | Matches incoming payments and sends reminders. A person decides when to chase or renegotiate terms. |
| Client relationships | Stays with people | The agent prepares the client's history. The sales team owns the relationship. |
| Negotiation | Stays with people | The agent supplies forecast occupancy and past terms. A person negotiates. |
| Pricing decisions | Stays with people | The agent can model scenarios. The decision rests with sales leadership. |
What an agent can do now
Tasks in the first group follow known rules and work with data that already exists. An agent can keep each face's availability up to date from the contracts and tell an agency which faces are free on given dates in a given area. It can prepare print orders with the sizes and number of copies for each Campaign, check every posting photo against the booked face, date and location, generate the invoice from what was delivered and build the close-out report.
What needs a person to review
In the second group the agent does most of the work, but the result commits the company to a client, a supplier or a site owner. A proposal carries prices and terms, a contract is binding, a settlement with a site owner moves money and a credit for an issue reduces what is collected. Here the agent prepares the draft and a person reviews it before it goes out.
It is worth writing down when that review is needed: an amount threshold, a new client, a change to agreed terms or an issue affecting several faces. In posting, the operations manager also knows constraints that sit in no system, such as a difficult access point or a pending permit.
What stays with the team
Three tasks fall outside an agent's reach. Client relationships with agencies and advertisers are built on trust and knowledge of the client. Negotiation requires reading the other side and deciding what to concede. Pricing decisions combine the media owner's strategy, its position in each area and its relationship with each agency.
In these tasks the agent prepares the person who decides, with the client's history, the forecast occupancy of the faces requested or what was agreed in previous years.
What an agent needs to work
If availability lives in a spreadsheet, contracts in emails and photos in weekly folders, the agent will have to guess which photo belongs to which face. The precondition is an operational database: inventory with a shared taxonomy, contracts with their faces and dates, and proof of posting and invoices linked to each Campaign.
Company adoption is still early. In McKinsey's 2025 survey on AI, 23% of respondents say their organisation is scaling an agentic AI system somewhere in the business, and another 39% have begun experimenting with agents. In any given function, the share scaling them is no more than 10%.4
Are scaling an agentic AI system somewhere in the organisation.
Have begun experimenting with agents.
Are scaling them in any given function.
Where to start
- Measure where the time goesHow many hours the team spends on each task on the map in a normal month.
- Put inventory data in orderEvery face with its reference, location, format and availability in one system.
- Start with a task from the first groupAvailability or proof-of-posting checks are usually good candidates.
- Write down the review rulesWhich amounts, clients or changes require the agent to ask for approval.
- Review the results every monthErrors caught, time saved and cases that can move to sample-based review.
Where OS.ooh fits
OS.ooh is the operational base an agent needs. Inventory is recorded with a shared taxonomy, each proposal and contract is closed with each agency at its agreed price and whatever is sold is locked, with no double bookings. Proof of posting arrives with photo, GPS and time linked to the Campaign, and the invoice is drawn from what was delivered. On that base, agents can be deployed to prepare proposals, work orders or settlements, with the right person reviewing according to each company's roles and permissions.
Sources
- McKinsey Global Institute, "Agents, robots, and us: Skill partnerships in the age of AI", 25 November 2025. mckinsey.com
- McKinsey Quarterly, "Where machines could replace humans—and where they can't (yet)", 8 July 2016. mckinsey.com
- McKinsey & Company, "The economic potential of generative AI: The next productivity frontier", 14 June 2023. mckinsey.com
- McKinsey & Company, "The state of AI in 2025: Agents, innovation, and transformation", 5 November 2025. mckinsey.com



