What 27 CMO Mandates Taught Me About Making AI Repeatable 

By Swetha Sree Kannan, AI Marketing Director, Toss The Coin 

The short version, if you only have a minute: 

Most AI adoption in marketing fails to scale. Not because the tools are bad, but because nobody turns the individual wins into a repeatable system. Here’s what repeatable AI infrastructure actually looks like, drawn from 27 live CMO mandates. 

A few numbers that frame it: 

• 75% revenue growth at TTC, with only 20% growth in headcount of Marketers. The traditional playbook would have demanded roughly 60% more marketers and writers to get there. 

• 150 hours of senior strategist time handed back to the team every month, freed from context assembly. 

• AI workflow setup time down from 30 to 45 minutes to under 2 minutes per session. 

• 200 to 300 AI-assisted workflows run monthly, at around a quarter of the compute cost of an unoptimised equivalent. 

• Monthly content calendars delivered 60% faster, with 40% higher organic engagement across 40+ content clients. 

Three patterns made it repeatable: build AI into your infrastructure, not your inbox. Bring it into the room with your client and encode your craft into the system before you scale it. 

I’m Swetha, the AI Marketing Director at Toss The Coin. I was given this role back in January 2025, when nobody in the industry was even writing this job description yet. I had ten years of marketing behind me, and I had taken to everything AI like a fish to water, which apparently made me the right person for a job that didn’t quite exist. 

My mandate was simple to say and hard to do: figure out how AI actually gets used inside a live CMO retainer. Not in a workshop, not as a pilot, but as their Client Partner, solving their everyday marketing problems. 

Here’s what I have learned after 27 of those mandates.  

Almost every client has had an AI win in their marketing interaction, but almost nobody has a system. And that gap is quietly costing marketing teams far more than they realize. 

What CMOs Keep Telling Me 

When I talk to CMOs about AI, the story is almost always the same. Something worked brilliantly once. A campaign came together in half the time, a brief felt sharper than usual, a piece of content outperformed everything around it… but then it didn’t hold. The next time, the magic was gone. It worked for one person and fell flat for another, and the quality wobbled week to week. 

The pattern was the same across every mandate. AI gets adopted tool by tool, person by person. One strategist finds a prompt that works and keeps it in their own notes, another builds something different, and nobody connects the two. You end up with a drawer full of individual wins that never compound into anything. 

That gap between a moment and a system is the only thing worth talking about. So let me be specific about what closing it actually looks like. 

Three Things We Built That Changed How the Work Runs 

We didn’t set out to build AI infrastructure. We set out to solve a plain problem: the agency was growing fast, client expectations were rising, and we weren’t going to hire our way out of the pressure. Something about how the work got done had to change. 

1. Stop Prompting. Start Architecting

The first thing we built was RayAI, the infrastructure layer that sits beneath every AI workflow at TTC. Before RayAI, a strategist starting a campaign would spend 30 to 45 minutes just getting ready to work: pulling brand documents, copying case studies, reformatting tone-of-voice guidelines, assembling context from six different places. Every time, for every single workflow. That isn’t AI-assisted work. It’s admin with a better tool bolted on at the end. 

RayAI ended that. It connects a vector database of each client’s publicly available knowledge (their website, published case studies, brand voice, past campaigns) to a library of pre-built prompts for the work we run most often. A strategist picks the client, picks the workflow, and Ray pulls the right context and layers it in automatically. They don’t write a prompt; they start thinking. 

The maths behind it is worth sitting with. A typical five-turn session used to burn around 90,000 input tokens, and Ray brings that down to roughly 22,500, a 75% reduction per workflow. Across 200 to 300 workflows a month, that’s over 160 million tokens saved a year. But the number I actually care about is the other one: 150 hours of senior strategist time, returned to the team every month. Not saved on a slide, but given back to client conversations, to creative thinking, to the work nobody was getting to before. 

The insight isn’t technical, it’s conceptual. AI as infrastructure, not as a query box. The system already knows the client, the strategist brings the judgment, and each does what it’s best at without ever replacing the other. 

2. Bring AI Into the Room 

The second shift was about where AI sits in a client relationship, and most agencies have this exactly backwards. The default is to use AI quietly behind the scenes and hand the client a finished output. We built the opposite. 

TTC Campaign Architect is a conversational campaign-building tool, designed to be opened in a working session and walked through with the client, live. You build the brief together in stage one, map ICPs and messaging together in stage two, and sign off on the execution plan together in stage three. Each stage needs an explicit approval before the next one opens. These are client checkpoints by design, not by accident. 

Here’s what changes when you work this way. End-of-process revisions collapse, because the decisions were made together at the start. There’s no handoff moment where the client sees the work for the first time and suddenly has opinions. The campaign that comes out at the end is one they’ve already approved, stage by stage, in real time. 

Since rolling it out, we’ve run close to seven full B2B campaigns through Campaign Architect, end to end, with clients in the room for every stage. Work that used to crawl through multi-week revision cycles now comes together in concentrated sessions over a few days. The campaigns didn’t get worse; the relationship got better. 

3. Encode Your Craft. Don’t Just Deploy a Tool

This is the one most agencies skip, because it’s the hardest to get right, and it’s also the one that compounds the most. As our client list grew, leadership kept circling the same question: how do we keep quality from depending on which writer happens to pick up which brief? 

The answer wasn’t to hire more senior writers. It was to take how a senior writer at TTC actually thinks, the instincts, the creative moves, the way they’d break a brief apart at their sharpest, and build that into a system everyone uses. 

The result is our LinkedIn calendar workflow, a five-stage system that moves from planning through topic shaping, caption development, theme exploration, and visual copy. AI does the scaffolding at every stage: it builds the monthly plan from client source links, groups the topics, drafts the captions, generates theme and visual options. The writer reviews, approves, and rewrites where it matters. Five sections, five human decisions, nothing automated end to end. 

What changed: calendar delivery dropped by around 60% in time, and organic engagement across the portfolio lifted roughly 40% with no paid support, just sharper writing reaching the right people. The biggest gains came on posts where the theme exploration stage pushed the writer past their first instinct. That’s the whole point of the system: not to replace the creative judgment, but to make sure the writer always has something worth judging. 

And here’s the part that compounds quietly. The craft no longer lives in any one person’s head; it lives in the institution. Every new writer inherits it on day one, and every client benefits from it, whether they ever know the system exists or not. 

What It Did to the Business 

Toss The Coin grew revenue 75% FY26. But here’s the number that tells the real story. 

The industry benchmark for a B2B retainer agency is roughly one writer per 4 to 5 clients, one designer per 5 to 7, one account manager per 5 to 7. Just those three delivery roles, serving 27  clients the traditional way would require us to grow our marketer headcount by 60%. Instead we just added 20% new headcount of marketers. The infrastructure took on the load those extra hires would have carried. The context assembly, the document retrieval, the first-pass scaffolding on every workflow: that stopped being a person’s job and became the system’s job. 

That’s what AI looks like on a P&L when it’s built in rather than bolted on. It isn’t a productivity bump on a single task, it’s a real change in how many people it takes to hold quality and volume at the same time. 

The Honest Part 

None of this happened in a straight line, and I would be doing you a disservice to pretend it did. The temptation to just add a tool was always there. Every time a new model dropped or a new platform promised to fix something specific, there was a version of the conversation that ended with us subscribing. Some of those tools earned their place, and several didn’t survive contact with how the team actually works. 

The ones that stayed weren’t the most impressive in a demo. They were the ones that got embedded into the workflow and were still being used three months later, without anyone needing a reminder. 

If I had to reduce all of it to one test: don’t ask whether the AI tool works, ask whether the AI system holds. Those are different questions, and most organizations are only asking the first one. 

If Any of This Sounds Familiar 

If your AI efforts feel more like a series of good moments than something your team can actually run on, I’d genuinely enjoy thinking it through with you. Not a pitch, not a proposal, just a conversation with someone who has spent the last few years sitting inside this exact problem and has some honest thoughts about where the leverage really is. 

I am a problem solver before I am anything else. If you are a CMO who knows AI should be doing more than it currently is, reach out, and let’s figure out where the system needs to be built. 

Swetha Sree Kannan is the AI Marketing Director at Toss The Coin, a B2B marketing agency running CMO-as-a-Service mandates across India. She leads how AI gets used across the agency, from design and content to MarTech and delivery. 

Connect with me on LinkedIn, or drop a comment below. 

Storyteller

Swetha Sree Kannan

Narrative Storyteller

Content strategist by day, professional Netflix binger by night.

My Heads Up