Can AI Really Help Agencies Scale Effectively?

Full video transcript

How’s it going everybody? Darby here with Genai University, joined by Rachel Woods, the CEO at Divvy Up Agency, where she helps companies scale up their own agency using AI without losing the personal touch. Rachel is here as a keynote speaker, part of the Scale with AI Summit series, to share how she’s helping agencies become unstoppable by leveraging AI to grow their business again without losing the personal touch.

Rachel: We run Divvy Up, which is an AI operations agency. And I always joke the hardest part about that is not the AI side, it’s the operations side. People don’t realize how much, when you have unlimited time, unlimited effort, unlimited little AI bots and agents running around, how much you can really solve operationally. We work on that every day with our clients, and we focus on agencies mostly.

At the summit, Rachel will share the AI playbooking method: taking SOPs or processes and, instead of people needing to do each step, teaching AI agents how to do those steps. There’s a spectrum, from full-on agents to simpler prompts in ChatGPT. The main thing is how to take business processes and break them down into stuff AI is good at, so you can start getting work off your plate.

Even though Divvy Up’s work is fully focused on agencies, Rachel says all of their methods are completely applicable across any type of business — anyone who runs a service business or is limited by the number of hours in the day. That includes people trying to grow their business and people who already have things figured out and want to spend less time on mundane work.

Rachel then opened her talk: “Today I’m so excited to share our AI playbooking method. It’s the thing that we’re known for in our agency.” The talk’s nickname: how to define all the hats but not have to wear them. She noted that people using AI daily or weekly typically see about a 10 to 30 percent productivity boost, and challenged the audience to think bigger — two times, ten times, fifty times more productive.

She framed scaling’s core problem as time: even with proven demand for a product or service, there isn’t enough time to do everything needed while scaling. AI’s real value isn’t just saving time — it’s creating a system that gives you what functions like unlimited time. The people winning with AI aren’t just finding productivity gains here and there; they’re delegating to AI: giving it clear instructions on how to do work the way they want it done. As you identify high-value processes and delegate them to AI via playbooks, AI executes the work without taking your time, and you reinvest that freed time into developing and delegating the next process — a compounding cycle.

She shared her background: she worked on Facebook’s AI research team on medium language models before it was widely popular, then used AI to scale business processes at her last company (marketed internally as “automagic,” not AI, before ChatGPT existed). After that company, she started talking about AI online about a month before ChatGPT launched, which led to creating two businesses: Divvy Up, the agency, and the AI Exchange, a community sharing this material.

Rachel identified what she considers the biggest misconception about AI: most people use it to get up to average at things they’re bad at — legal work if they’re not a lawyer, copywriting if they’re not a copywriter. But AI can also be made good at the things you’re already good at. She explained that AI is trained on a huge dataset and, given a prompt, predicts the most likely or average response based on that data. Feeding it a different dataset — your expertise, proprietary data, anything outside the generic average — steers its output toward reflecting your standards instead. This remains true even as AI improves, and will remain true at AGI: output is only as good as the instructions and data provided.

She compared teaching AI to hiring a smart intern: capable, but lacking context, so specific instructions (templates, examples, guidelines, customer research) produce better results than a vague request. Divvy Up calls the process of creating and codifying those instructions “playbooking” — turning a manual process into something AI can handle systematically, repeatedly, and at unlimited scale, similar to writing an SOP (standard operating procedure).

She outlined the three steps to build an AI playbook: first, plan the process — define the goal, the inputs AI needs, and what the output should look like, then break it into numbered steps. Second, write AI instructions for each step — a prompt, an automation step (for example in Zapier), or an agent, depending on the tool. Third, test, refine, and improve the prompts and process over time, treating early versions as MVPs rather than waiting for perfection.

She then walked through a real-world use case: scaling client management. The first step is mapping out everything you might want to scale — preparing agendas, sending meeting recaps, adding ad hoc tasks to a project management tool, saving insights for future strategy or upselling, monitoring for churn risk, sending updates between calls, following up on open loops, proactively suggesting optimizations, and running QBRs. From that map, you group related tasks into one cohesive playbook. Her example combines meeting recaps, task creation, insight-saving, and churn-risk monitoring into a single process.

She detailed the steps: first, an internal recap analyzing the transcript for things not shared with the client — touchy topics, surprising questions, tasks the team may not want to commit to, the vibe of the call, and any churn-risk signals. Second, a client-facing recap in the agency’s brand voice, covering wins, decisions, and action items — functioning almost as marketing of the work done. Third, saving those recaps and automatically sending the client-facing version, optionally with a human review step. Fourth, having AI identify tasks mentioned in the meeting and auto-add them to a project management tool using the agency’s tagging and time-estimate conventions.

She noted several downstream benefits: this system can catch scope creep by flagging out-of-scope requests for team review, and can support quarterly business reviews by storing all client touchpoint data somewhere AI-accessible so QBR prep becomes fast. Once built, the same playbook helps standardize onboarding, since new team members immediately use it and their recap quality matches the agency’s standard, and they can study the playbook itself to understand the reasoning behind the standards.

She extended the example further: adding a step to flag “case study moments” during meeting analysis, then running a monthly playbook that gathers those flagged moments and drafts case studies from them. She also mentioned a content use case: taking a transcript from a sales call or keynote, having AI identify post ideas, researching each with a tool like Perplexity, and drafting posts — which can connect into a broader content workflow.

The key principle throughout: every time you delegate to AI and save time, reinvest that time into delegating the next process, rather than just banking the savings. This produces the “unlimited time cycle,” which compounds over several loops into a significant business advantage. Rachel closed by contrasting the old way of working — developing expertise, then doing the same projects repeatedly with modest efficiency gains — with the new way: developing expertise, then writing an AI playbook so AI executes those projects at scale while you reinvest the freed time into developing the next expertise or process.

She challenged viewers to pick a strategically smart process, outline it as a playbook, write prompts for each step even if just in a document, continuously improve the prompts, and use the freed-up time to build the next playbook. She invited people to reach out on LinkedIn or by email to share what they unlock using the method.