Agents at work: Harnessing AI for intelligent post-sales ops with Boostr’s Patrick O’Leary

AI agents aren’t science fiction anymore. They’re showing up in the real workflows where publishers lose the most time: plan ingestion, IO revisions, trafficking, pacing, and optimization. The trick isn’t “add a chatbot.” It’s embedding intelligence at the exact moment and place work happens, so teams get hours back without changing how they work.

Rob spoke with Patrick O’Leary, Founder & CEO at Boostr, about his view of agent adoption, and how it’s grounded in real deployments inside media companies.

Rob: For many publishers, post-sales still runs on spreadsheets, email, and manual entry. What does using AI agents to tackle these tasks actually look like in practice?

Patrick: Context is everything. Agents need to understand who the user is, what record they’re on, and which fields, files, or systems the task touches. If the agent isn’t in the user’s workflow, it’s forgotten. What’s more, if it lacks data context, it makes weak decisions. That’s why we favor embedded agents; intelligence that lives inside the tools the teams are using and performs the heavy lift behind the scenes (including prompt generation), so the user doesn’t have to.

A good example is an embedded Contact Enrichment Agent that reads an email, spots missing or outdated CRM details, and proposes updates, so the user can apply them with one click – no tab-switching, no manual entry. That’s key to the adoption.

Rob: For publishers who already have order management systems (OMS), where can AI agents add real value? What gaps do agents fill that traditional OMS typically don’t address?

Patrick: Think “adjacent automation.” Traditional OMS platforms are great at structured workflows but leave a lot of swivel-chair tasks around them: applying IO revisions across line items, reconciling changes, chasing approvals, or normalizing formats from agencies. Agents extend the OMS boundary, automating those repetitive but critical steps and turning the OMS into an agentic hub for post-sales. That’s where teams feel the real lift – fewer manual touches, fewer errors, and faster cycle times.

Rob: What's the practical difference for publishers between embedded agents vs external agents? When should a publisher lean toward embedded agents, and when would external agents make more sense?

Patrick: Embedded agents sit inside your system of record. They have native context, higher security, and fewer adoption hurdles. External agents can be flexible, but they often fight limited APIs, weaker data access, and clunky user journeys. Our rule of thumb: if a task starts in a system of record (CRM, OMS, trafficking), go embedded. If you need cross-system orchestration and the APIs are robust, an external agent can help, but it still needs context, governance, and a clean adoption path.

Rob: Of course, there’s a ton of chatter that AI will replace human jobs in ad ops and sales teams. What's your actual experience so far? What happens to the roles, workloads, and workflows when these agents get introduced?

Patrick: Agents augment humans when they’re the right tool for the job. They’re great for deterministic, high-volume work and for teeing up decisions, but they’re not a fit for every task. For processes that must be 100% accurate (billing is the classic example), you still want human oversight because LLMs can hallucinate and mess up the math. The smartest approach is “guided first, autonomous later.” Train, supervise, and dial up autonomy as trust builds.

Rob: When publishers bring AI agents into those post-sales workflows, what kind of real-world results have you seen? Can you share examples of efficiency or accuracy improvements that a publisher would care about?

Patrick: Early results are promising. One beta test processed a 50-line-item IO revision in approximately two minutes versus the standard 30 minutes manually – a 15x improvement – while keeping humans in the approval loop. Multiply that across the week, and you free up hours for pacing checks, client communication, and optimization strategy. We’re seeing similar gains on plan ingestion and change-management flows where agents normalize formats, map fields, and detect inconsistencies before they hit ad ops.

Rob: AI agents are supposed to improve over time, not just solve immediate tasks. For publishers, what does that continuous improvement actually look like, and how do you make sure these agents get smarter, not just older?

Patrick: Treat agents like new teammates: give feedback, reinforce what worked, and correct mistakes so they don’t repeat. Under the hood, that means capturing user approvals/edits as structured signals, versioning task recipes, and choosing the right tool for the job (LLM when it helps, deterministic code when it’s faster/safer).

It’s equally important to keep prompts and orchestration invisible to the user – push learning into the workflow, not into a separate prompting skill the team has to master. That’s how you get compounding quality gains without adding cognitive load.

Rob: What are the practical risks of bringing AI agents into publisher workflows? Where have you seen publishers stumble, and how can they avoid those pitfalls?

Patrick: There are two big ones. First, adoption: if agents are “bolted on” and force users into a separate UI with manual prompting, they’ll be ignored. Embed them at the point of action so the agent does the prompting for the user.

Second, accuracy: roll agents into processes with tolerance for minor error and a clear approval step. Leave revenue-critical, math-sensitive tasks for guided modes with human sign-off until the agent earns trust. Start with high-frequency, high-friction tasks to build that trust.

Rob: Looking ahead two to three years, how do you expect AI agents to change the day-to-day of ad ops teams? What’s the biggest shift publishers need to prepare for right now?

Patrick: For now, build a roadmap of manually repetitive tasks across post-sales and go after the top time sinks. Don’t wait for a grand unifying model; deploy targeted agents that remove keystrokes and context-switching today.

In the mid-term, many tasks currently offloaded to outsourced teams will move to agents, especially change management and standard trafficking steps.

In the longer term, the OMS will likely become explicitly agentic – your post-sales command center that orchestrates people, processes, and agents across the revenue org.

Ready to make your OMS agentic?

The fastest ROI in AI is removing the manual, error-prone, steps your team repeats every day. Start with guided agents embedded in your post-sales workflows, measure the time they give back, and expand from there. Your team will spend less time wrangling spreadsheets and more time delighting advertisers. Start a conversation with Boostr or book a demo to see how embedded AI agents can unlock meaningful efficiency across your revenue teams.

This is content created in paid partnership with Boostr. We only feature partners who we believe bring real value to the publisher community.