I sat down with Ben Carden from Revenue Flow recently to talk about how AI is changing go-to-market. We covered a lot of ground, from my early days in recruitment boiler rooms to what outbound will look like in a couple of years. But one theme kept coming back: the split between operations and judgement, and how most sales teams have it backwards.
Here’s what I mean.
The 60/40 problem
Salesforce’s State of Sales report puts it bluntly: reps spend around 60% of their time on things that aren’t selling. Building lists, enriching data, logging activity, updating the CRM, enrolling prospects in sequences. All important work. But none of it is the work that actually generates pipeline.
The other 40% is where deals happen. Reading a conversation. Knowing when to push and when to pull back. Asking a question the prospect didn’t expect. Deciding that this account isn’t worth pursuing even though it looks good on paper. That’s judgement work. And it’s the stuff that turns average reps into great ones.
The tragedy is that most SDRs never get enough practice at the judgement work because they’re buried in the operational stuff. They spend their mornings in spreadsheets and their afternoons copying data between tools. By the time they pick up the phone, they’re already behind.
This isn’t new. I’ve been in B2B sales for over 15 years. I started in a recruitment boiler room in the UK where the day was mostly dialling, getting through gatekeepers, and sometimes literally walking into offices to ask for the decision maker. I ran 100 lead researchers at TaskDrive, a business that only existed because the operational grind was so time-consuming that companies would outsource it. I co-founded Speak On Podcasts and ran classic “predictable revenue” outbound. And now as a consultant, I see the same pattern in every company I work with.
The tools have changed. The ratio hasn’t. That’s what needs fixing.
We automated the wrong things
Here’s the irony. AI has given us the ability to automate almost all of the operational work. List building, enrichment, data verification, first-draft emails, CRM updates. The technology exists to hand all of this to machines.
But most teams have done the opposite. They’ve kept humans on the operational work and handed the judgement work to AI. Reps are still manually building lists and logging calls while AI writes their “personalised” outbound emails with fake references to the prospect’s university or local restaurants.
We’ve automated the human and manualised the machine.
This is the whole thesis behind The Effective Seller: automate the process, master the conversation. Put AI on the intelligence work (the stuff that follows rules) and free up humans to spend more time on the judgement work (the stuff that requires context, empathy, and trust).
What programmatic prospecting used to look like
I was doing what I call “programmatic prospecting” back in 2020, well before Clay and the current wave of tools. It meant having several tabs open, downloading CSVs, uploading them somewhere else, and maybe using Zapier to string a few things together. It worked, but it was brittle and time-consuming.
I actually found a Google Sheet I built in 2019 recently. It was a “Company Intelligence Tool.” You’d type in a domain and it would generate clickable links to Google Trends, Crunchbase, LinkedIn Jobs, Capterra. One row per tool. Click each one manually. At the time, I was genuinely proud of it.
That was six years ago.
Now with Clay, you run all of those enrichments in columns, not rows. No clicking. The data comes to you, structured and ready. Take it further with Claude Code and you can build skills that run as agentic workflows. No dashboard. No tabs. No copying between tools.
The speed of change is staggering. MCP went from 2 million monthly SDK downloads when Anthropic launched it in late 2024 to 97 million by March 2026. Over 10,000 public MCP servers exist. The infrastructure for API-first, dashboardless sales tools didn’t exist two years ago. Now it’s the default for every new tool being built.
What took me a full day in 2020 can be done in minutes now. But the strategic thinking hasn’t changed at all. You still need to know who to target, what to say, and when to say it. The execution got faster. The strategy didn’t get easier.
The SDR role isn’t dying. It’s splitting in two.
Everyone keeps saying AI is going to replace the SDR. But look at what’s actually happening.
Anthropic is actively hiring salespeople. The companies building AI are hiring humans to sell it. Because they understand that the conversation itself (the discovery, the trust-building, the judgement) is the part AI can’t do.
Clay just built their first ever SDR function. They’re calling them “ClayDRs,” and these reps spend their time on high-quality, bespoke outreach to enterprise accounts. That’s not the traditional SDR grind. That’s strategic selling.
What’s dying is the traditional SDR model: high volume, templated sequences, measured purely on activity. What’s replacing it is something more interesting.
I think we’ll see two roles emerge. The GTM Engineer, who builds and manages the machine (workflows, enrichment, signals, automation), possibly in a pod structure alongside sellers. And the Full-Cycle Seller, who uses the machine to manage the entire journey from prospect to close.
The handoff between SDR and AE is dying because the machine handles the top of funnel and the seller handles the conversation. But the SDR role as a training ground for sales careers isn’t going away. It’s just that what you’re training on changes. Less data entry and volume. More conversation skills, discovery, and strategic thinking from day one.
Signals are easy to build. Knowing what to do with them is hard.
I’ve been pushing the idea of signal stacking for a while now. Combining things like job changes, funding rounds, and hiring patterns to figure out exactly when to reach out to someone. That used to be gut feel. A great rep just “knew” when to call. They’d notice a LinkedIn post, hear something at a conference, see a job listing. The problem was they could only track maybe 20-30 accounts in their heads.
Now AI can surface those signals across thousands of accounts simultaneously. The detection is automated. But not every sales rep is “great,” and that’s where it gets interesting.
The skill shifts from “can I spot the signal” to “do I know which signals actually matter for my specific product and market?” Not all signals are equal. A funding round is meaningful for some products and irrelevant for others. A new hire matters, but only if you wait two to three weeks before reaching out, not day one.
Signals alone don’t convert either. I’ve seen teams build beautiful signal workflows that fire into Slack channels. And nothing happens. Because the rep doesn’t know what to do with the signal. No context, no suggested angles, no reason to call beyond “something changed.”
The real skill now is signal design: which signals to track, how to combine them, what context to wrap around them, and what action the rep should take. That’s judgement work. And it’s more valuable than ever.
Data quality is the new differentiator
Here’s something that should make every sales leader uncomfortable. I recently looked up Clay’s revenue across different data platforms. The estimates ranged from zero to $36 million. Same company. Wildly different numbers. And these are the platforms sales teams rely on for targeting decisions.
As we hand more research over to AI, data quality becomes the single biggest differentiator between good outbound and bad outbound. If your enrichment is wrong, your targeting is wrong. If your targeting is wrong, your messaging is irrelevant. And then you’re just sending bad emails faster.
Someone has to check. Either the rep does a manual sanity check (which defeats the purpose of automation) or you build validation layers into your workflow. Waterfall enrichment helps. Cross-referencing sources helps. But nothing is 100%.
This is actually an argument for keeping humans in the loop, not removing them. AI handles the volume. A human needs to spot-check the quality. Trust but verify.
Where the line moves next
Sequoia made this point that what counts as “judgement” today eventually becomes “intelligence” tomorrow. The line keeps moving as AI gets better. Two years ago, writing a personalised email felt like judgement. Now it’s intelligence work that AI handles.
So what moves next?
I think it depends on the complexity of the purchase. For products with a narrow scope and a low price point (under $1,000 a month), AI will handle the entire buying process within a couple of years. Agents buying from agents. The evaluation is straightforward, the risk is low, and the decision doesn’t require much context.
But the moment there’s nuance, it stays human. When the purchase involves integration with existing systems, when there are change management implications, when multiple stakeholders need to align, when the wrong decision costs you six months of rework. Those are judgement calls. AI can surface the options and do the research, but a human needs to weigh the trade-offs.
The dividing line isn’t really about price. It’s about consequence. Low consequence, narrow scope, AI handles it end to end. High consequence, complex implications, humans stay in the loop.
For outbound specifically: prospecting for low-ACV, self-serve products will be almost entirely automated. Prospecting for enterprise deals with long sales cycles and multi-threaded buying committees? That’s going to need humans for a long time. The conversation itself is where trust gets built, and trust is what closes complex deals.
What I teach now versus two years ago
Two years ago I was teaching people how to use Clay tables, how to set up enrichment waterfalls, how to build lists step by step. Now I’m teaching people how to think about AI-assisted workflows, MCP connections in Claude, and agentic processes in Claude Code.
I’ve stopped teaching manual list building in the traditional sense. The “go to LinkedIn Sales Navigator, export to a CSV, clean it up in Google Sheets” workflow is dead. The tools do this now.
What I teach more of now (and have always taught): campaign strategy, ICP definition, how to think about signals, and how to build the context layer that makes AI actually useful. The foundations haven’t changed. But the emphasis has shifted from “here’s how to use the tool” to “here’s how to think about the problem so the tool can solve it.”
The biggest shift: I now spend as much time teaching people what not to automate as what to automate. The premature automation trap. Teams buy the tools before they’ve fixed the foundations. They scale a broken process and wonder why nothing improves.
If you’re still doing outbound the old way
If you’re still running manual lists, templated sequences, and a big SDR team, the first thing to change isn’t your tools. It’s your understanding of the problem.
Map your rep’s actual workflow. Sit next to them and document every step they take to research a prospect and send an email. Every click, every tab, every export. I guarantee there are 5-10 steps that could be automated or eliminated.
Don’t start by buying a tool. Start by understanding the problem. Most teams buy Clay or Apollo and then try to figure out what to do with it. Flip it. Figure out what your reps are wasting time on, then find the tool that solves that specific problem.
What’s left that only a human can do?
Curiosity. Asking the question nobody expected. Reading the room on a call and knowing when to push and when to shut up. Building trust over time. Saying “this isn’t the right fit for you” and meaning it.
Those are human skills. And as AI floods the market with more automated touches, they’re becoming more valuable, not less. The supply of AI-generated outreach goes up. The value of genuine human connection goes up with it.
Is it the best time or the worst time to be in B2B sales? Both. Best time if you’re willing to adapt. The reps who can use the machine and master the conversation will be worth more than ever. Worst time if you’re still relying on volume and hoping the phones save you.
Automate the process. Master the conversation. That’s the future.
If you’re a founder or sales leader trying to figure out where AI fits into your outbound, I can help. I work with B2B SaaS teams to fix the foundations, build the systems, and train the skills that make outbound actually work.
Book a call and let’s talk about what’s broken and what to do about it: yellowo.co.uk/schedule