We sat down with Benjamin Cox, VP of AI Strategy & Operations at Rakuten Advertising, to unpack how his team is operationalizing AI agents across the business—covering everything from build vs buy and vendor selection to change management, measurement, and a hub-and-spoke delivery model.
Rakuten Advertising operates an affiliate marketing network, connecting large advertisers (e.g., luxury retail, global financial institutions, tech, apparel, travel) with publishers (loyalty, content, influencers, and more). Because their edge depends on massive-scale data analysis, prediction, and finding optimal partnership matches, Ben sees AI as a natural force multiplier—both in core analytics and in the “white glove” services that improve performance for customers.
1) Start with the right definition: agents ≠ chatbots
Ben draws a clear line:
-
Chatbots: conversational, human-in-the-loop tools that help you think and draft.
-
Agents: systems that take on workflows—handling repeatable tasks end-to-end or semi-automating steps with humans involved at chosen points.
His framing: if the business does “the same task a million times,” that’s where agents can augment, queue, or automate work to create scale.
2) Crawl phase: don’t throw AI at your hardest problem
Ben warns against a common corporate reflex: start with the biggest, thorniest problem.
Instead, his crawl-phase playbook is:
- Pick low-risk, high-impact “low-hanging fruit”
- Keep scope tight and controlled
- Treat early efforts as calculated experiments to prove value
- Do use-case selection and people enablement in parallel (he argues doing both well “10x” improves odds of success)
3) Win adoption with behavioral science, not mandates
Ben’s adoption lens is blunt: people won’t use agents just because an organization wants them to.
He targets two motivators:
-
A) Painkillers — Automate the tasks people dread—the recurring work that makes them think, “I can’t believe I’m still doing this every week.”
-
B) Rewards — Use agents to make work more fulfilling, like deeper research and strategy—with a “consultant next to you” feel.
If you miss both, people “fight the machine,” and adoption stalls.
4) The market problem: vendor overload (and why consolidation matters)
Ben calls out the noise: for any given use case, there may be many vendors, and just selecting tools can become its own burden.
His practical stance:
- He predicts industry consolidation and M&A over time
- Internally, he aims to cover needs with a small number of core AI applications (he mentions thinking in the range of “one to five”)
- Aim to keep the stack stable to reduce constant relearning—because teams are already forced to re-learn “AI” frequently as the space turns over quickly
5) Measurement: prove value early, or you’ll never scale
Ben splits success metrics into two buckets:
-
Tangible (easier): Dollars saved, dollars generated, clear “countable outcomes” (e.g., contracts closed)
-
Intangible (harder): Hours saved, quality improvements (e.g., better research), and the hardest part—translating “better work” into real throughput (partnerships, investments, build-vs-buy decisions)
His rule: choose early use cases you can actually validate. Once you prove a few wins:
- You build credibility
- Demand grows
- The org flips from “finding projects” to “turning projects down”
6) Change management: normalize failure and manufacture champions
Ben treats experimentation as a core operating discipline:
- He jokes he wants an OKR for: “name three AI projects you tried that failed.”
- He reinforces that model-building is probabilistic—projects can fail for business, legal, or cost constraints even with good intent.
Then he scales adoption through people:
- Identify influential champions inside non–AI-native teams
- Look for intellectual curiosity and a willingness to tinker
- Bet on “progressive experimenters” in client services, marketing, ops—not just technical experts
7) Skills ladder: meet people where they are
Ben describes a tiered learning path that brings people along, instead of demanding instant expertise:
- Chat tools (entry point)
- Workflow automation (requires process mapping + implications)
- AI agents (no longer require coding background to build)
His point: LLMs compressed the learning curve. Non-technical users can now draft a PRD, hand it to a coding agent, iterate, and get to a demo—something older “intelligent automation” promises rarely delivered.
8) Operating model: hub-and-spoke, with embedded analysts
Ben favors a hub-and-spoke approach:
- A core team handles heavier R&D and major builds
- Embedded analysts sit inside business units (marketing, client services, etc.) to learn the domain and apply AI where it matters
His principle: “I don’t want to be the single author of any agent.” But he still wants governance on the big projects—visibility, monitoring, measurement, and control.
9) Distribution: adoption lives (or dies) in daily tools
Ben’s adoption rule is simple: if agents aren’t where people already work, usage collapses.
Rakuten Advertising uses agents heavily in existing collaborative tools, because it reduces friction and the “another scary tool” problem.
10) The “after state”: automation-first as a default mindset
Once agents are in production, Ben says the biggest change is how teams think:
- They’ve automated enough that new services start from an automation-first posture
- The starting question becomes: “What if this had to scale to 100% of customers?”
- Then they work backward to decide what must remain human-led (human touch, oversight, exceptions)
- They even invested in business process mapping training so teams can describe workflows clearly enough to improve them with agents
Ben’s starter advice for leaders
- Set the cultural expectation: this is the direction
- Start with weekly pain points that improve quality of life
- Be realistic and expect failure
- Build one narrow agent first—then compounding makes the next ones easier
- Stay measurement-driven, and don’t repeat the same failed experiment ten times
- Accept you’ll always feel behind; progress comes from iteration and testing
That’s how Rakuten Advertising is turning decades of affiliate-marketing expertise into scalable, agent-powered execution across the business.