On Thursday, Aziz Ben Allala of Syrena put a question to our panel that I'm still chewing on. AI is joining the revenue team, he said, so who's accountable for the number?

My answer is a person. And the best thing AI can do for that person is get them more conversations with buyers who actually want to talk.

The question on stage

The panel was "How Do You Scale Revenue?" at Syrena and Deloitte's summit at Level39 in Canary Wharf. I shared the stage with Tom Healy, head of fundraising at Haatch, Jonathan Daglish from Deloitte and Tom Andrews from Airspeed, with about 110 founders, revenue leaders and investors in the room.

People buy from people

As long as humans are the economic buyer, the best use of AI in your revenue engine is scaling human connection.

Someone signing a big contract wants to judge the person on the other side of it. AI can do most of the work that gets you to that meeting: the account research, the prep before a call, the follow-up notes afterwards. But the deal gets made when two people talk.

So the test for any AI in a revenue team is simple: does it get your sellers more conversations with the right buyers, better prepared each time?

The AI SDR trap

A lot of AI sales tools are built for the opposite job. They build a cheaper, worse copy of a salesperson and point it at the biggest list they can find.

My question for anyone buying one: if human SDRs aren't the right approach for your business, why would worse AI ones be?

Look at what these tools brag about. Inbox warming exists as a feature because sending at volume gets punished, by spam filters first and then by the people reading. When a product's selling point is getting past those filters, I treat it as a red flag.

Hunter's 2026 State of Email Outreach report, based on 31 million emails sent by Hunter users in 2025, puts numbers on it. Sequences sent to 21 to 50 people got a 6.2% reply rate, against 2.4% for lists of 500 or more, so the small list with a real reason to write is the one that gets answered.

At Lua we build growth systems that learn your own playbook and get more autonomy as they prove they can run it. An agent starts with the research and the drafts, a person checks the work, and the agent takes on more as the results justify it.

Point tools against one layer

The other route is to buy a separate AI tool for every corner of the business: one for outbound, others for call notes, support and forecasting. Do that and you'll lose to an 18-year-old with a Claude Code subscription.

Each of those tools has a slice of what your company knows about its customers. The context that makes a salesperson good (what this buyer asked last quarter, why the last deal stalled) ends up split across several vendors, and none of them has the whole picture.

Gartner predicts that by 2028 AI agents will outnumber sellers 10 to 1, and that fewer than 40% of sellers will say agents improved their productivity. It has a name for the mess, too: agent sprawl.

Companies need to own their intelligence layer. That means one place where the customer context, the playbook and the agents work together, on data the company controls. Lua is what they build it on.

Our own revenue team

We've been building this for our own revenue team at Lua, and it starts in an unglamorous place: who already knows whom. Before the agents suggest a cold email, they map the warm routes: the teammate who worked with the buyer's COO, or an investor on their board. Most weeks there are more warm routes than anyone has used.

They draw up a weekly list of around 20 accounts and note why every one matters now, whether that's a new leader, a new contract or a new office. For each account they research the buying group, draft the first message and suggest the play. Then they stop, and nothing goes out until a person reads it and switches it on.

Over time they're built to learn which plays work for which buyers, and keep the ones with evidence behind them.

Human moments

Being accountable for the number means knowing which moments need a person.

The first meeting is one. Buyers deciding whether you're right for them want a seller there, and I'd want the same if I were buying. Pricing is another, because it's a judgement about the relationship as much as the spreadsheet. And the sensitive moments go to a person every time: the deal that's wobbling, the customer who's unhappy, the renewal under pressure.

Around those moments, agents do the legwork. They research the account before the first call, write up the follow-up after it and update the CRM.

So who's accountable for the number? A person is. Our agents do the research, find the route in and write the first draft, then wait for someone to switch it on.

Thanks to Aziz, Clem and the Syrena team for having me.

If you're working out what AI should do in your own revenue team, book 20 minutes with Hugo, our VP Sales.