A CPO Replaced Her £500/Month Sales Admin With an AI Agent. The Savings Were Immediate.

Claire Vo is a 3x Chief Product Officer, founder of ChatPRD, and host of the How I AI podcast. She's not someone who falls for hype. When AI agents first appeared, she was one of the loudest sceptics. Her first install deleted her family calendar. It took eight hours to set up.

Six months later, she runs nine AI agents across three computers. One of them—an agent she calls Sam—replaced a sales admin she was paying £500 a month.

This isn't a theoretical article about what AI could do. This is what it's actually doing, right now, for a real business.

What Sam actually does

Sam is a sales agent. Not a chatbot. Not a generic assistant. Sam has one job: manage the sales pipeline. Specifically:

• Lead research. Sam scans LinkedIn, company databases, and industry news to find prospects that match Claire's ideal client profile.
• CRM updates. Every interaction gets logged. Every lead gets scored. Every follow-up gets scheduled. No manual data entry.
• Email drafting. Sam writes personalised outreach emails based on what it knows about each prospect—their company, their role, their recent activity.
• Follow-up sequences. If someone doesn't reply, Sam sends a follow-up at the right interval. If they do reply, Sam flags it for Claire to handle personally.

Before Sam, Claire was paying a virtual assistant 10 hours a week to do exactly this. The VA was good—but limited by human speed, human working hours, and human consistency.

The real economics

Let's be honest about the numbers.

Before (human VA): £500/month for 40 hours of work. Available during business hours. Quality depended on the day. Needed training, managing, and checking.

After (AI agent): £30–50/month in API costs and infrastructure. Available 24/7. Consistent quality every time. Needed setup and occasional tuning.

That's a 90% cost reduction. But the bigger win isn't the money—it's the time Claire got back. She stopped managing someone else's work and started focusing on closing deals herself.

It's not magic. It's painful to set up.

Claire is brutally honest about this. In her words: “I won't sugarcoat it: it's a pain to set up. It is not hands-off. But the value is so high, I am willing to go through the pain.”

The setup challenges are real:

• Configuration takes hours, not minutes. You're not installing an app. You're defining an agent's personality, permissions, tools, and boundaries.
• Security needs careful thought. You don't give an AI agent your main email password. You create a separate account with limited permissions. Progressive trust—reading only at first, then drafting, then sending.
• It will make mistakes. Claire's calendar agent deleted her family calendar in the first week. Things break. You need patience and a willingness to debug.

This is exactly why most UK business owners don't do this themselves. The value is clear. The implementation is the bottleneck.

What Anthropic's own sales team discovered

Here's something fascinating. At Anthropic—the company behind Claude—their internal sales team started using AI agents before anyone told them to. Jenny Wen, Design Lead at Anthropic, described it on a recent podcast:

“There were a few people who were die-hard Claude Code users, and they were using it to generate leads lists, come up with scripts for calls. It blew my mind, because I didn't even comprehend that you could use it for those things.”

Even inside an AI company, the sales use cases surprised people. The people closest to the ground—the ones doing the work—found the automation before the product team designed it.

What this means for your business

If you're running a service business in Hampshire or anywhere in the UK, you probably recognise the pattern. You or someone on your team spends hours every week on:

• Updating your CRM (or the spreadsheet pretending to be one)
• Writing follow-up emails that sound personal but aren't
• Researching prospects before calls
• Chasing leads who went quiet
• Creating quotes and proposals from templates

Each of these tasks is a candidate for an AI agent. Not a chatbot—a proper agent with access to your CRM, your email, and your business rules.

The “ugly product-market fit” test

Claire uses a term I love: “ugly product-market fit.” It means the product is painful to use, full of rough edges, and sometimes breaks—but the value is so high that people keep using it anyway.

That's where AI agents are right now. The technology works. The ROI is real. But the setup experience is terrible for anyone who isn't technical.

This gap—between “it clearly works” and “I can't set it up myself”—is exactly where a done-for-you service makes sense. Someone who understands the tools builds it for you in days, not months.

How to start (without deleting your calendar)

• Pick one task, not five. Claire's biggest mistake was trying to throw everything at one agent. Start with your most time-consuming, repetitive sales task.
• Use progressive trust. Let the agent read-only first. Then let it draft. Only when you're confident, let it send.
• Separate accounts. Never give an AI agent access to your primary email or calendar. Create dedicated accounts with limited permissions.
• Budget £30–50/month for infrastructure. API costs, a small server, maybe a CRM integration. That's your running cost after setup.
• Expect two weeks of tuning. The first version won't be perfect. By week two, it'll be better than your VA.

The bottom line

Claire Vo went from sceptic to running nine agents because the economics are undeniable. One agent alone saved her £500 a month and 10 hours a week. Across her entire agent team—that's the equivalent of a full-time employee she doesn't need to hire.

For UK service businesses doing £500K–£5M in revenue, the question isn't whether AI agents work. It's whether you set them up yourself (painful) or get someone to build them for you (fast).

Sources: Claire Vo on Lenny's Podcast, “From skeptic to true believer” (2026). Jenny Wen (Design Lead, Anthropic) on Peter Yang's podcast, “How Anthropic Builds Products with Claude Cowork & Claude Code” (2026).