Mark Colgan builds sales automation for a living. He has done it for more than 250 B2B SaaS companies, and his model is to design the system, train the team to run it, and get out of the way. So Brendan Hughes opened Episode 4 of Intent, Decoded by asking, fairly, whether Mark had turned up to the wrong show. The title was "Are we automating the wrong parts of sales?", and the guest automates sales.
Mark's answer was one line, and the rest of the episode hung off it: "My mantra is automate the process and master the conversation." The trouble, as he and MarketSizer founder Niall O'Gorman spent the next fifty minutes explaining, is that most teams have it the other way round. Humans are doing the robot work, and the AI is attempting the human conversations.
Premature Automation: Scaling a Process Nobody Has Proven
Premature automation is automating a sales process before anyone has proven it works. Mark's name for it, and his explanation of why the AI SDR wave of the last few years mostly failed to deliver and pivoted to inbound.
The arithmetic is brutal and simple. He gets brought in to talk about the shiny thing: the tools, the tech, the AI. Then he asks to see the outbound results. "Let's just say they've got a 0.1% reply rate on 500 emails." Automation will happily send the same message to 5,000 people. "But if we get 0.1% at 500 people, we're probably going to get 0.1% at 5,000 people."
Volume multiplies whatever is already there. If the targeting, the message and the proof are not working at small scale, automation just makes the failure faster and more expensive. Niall traced the habit back to cheap capital, when it was cost-effective for a while to "spam the planet" and nobody worried about the leaky bucket underneath. Then every inbox became a blur of the same email, the money dried up, and teams had to go back to the basics.
Those basics, in Mark's list: do you know who you are targeting, what the segments are, what each one cares about, which problems you solve, and how you can prove you have solved them for people like this before. Only then does timing come in. No signal guarantees the moment someone will buy, but you can improve the odds of a reply. "There's a lot of work that goes into it, which is human work, before we build out the automation and then start to leverage AI on top."
Start With the Deals You Lost
Mark's first move in a new engagement is a closed-won and closed-lost analysis, and he means both halves. Everybody studies the wins. He wants to know why deals were lost: a product gap, a missing feature, timing, or something else entirely.
The second move is the one I suspect most consultants skip. He interviews the people doing the work, from the BDRs to the managers to leadership, and asks them to show him what they actually do. Years in recruitment taught him to read between the lines, and the pattern is consistent: "no matter what the process says a rep is supposed to be following, they're always doing something slightly different."
So there are three versions of the sales process in most companies. The one in the playbook, the one RevOps built into the tools, and the one reps run every day. The gaps between them often include something more basic than process, too: reps who do not really understand the difference between a founder at a 50-person company and a founder at a 3,000-person company.
His answer is not to automate around those gaps. It is to automate the admin so there is time to close them. "We're trying to automate as much as we can to free up the time so that we can reinvest that back into the training and development of our teams."
AI Amplifies Whatever Foundations It Is Pointed At
AI does not fix bad data. It makes bad data louder. Niall's version, drawn from years in ecommerce where wrong product data sent shoppers to the wrong product and drove up returns: "If you've got dirty data within your CRM and you point a very sophisticated AI at dirty data, it's just going to make a lot of noise."
He was candid that MarketSizer sees this first-hand. A conversation about which accounts to prioritise tends to become a conversation about which CRM a team uses, how it is set up, how many duplicates it holds, which fields are empty, and whether the team logs in at all. We have written about why that happens in why CRM data is inaccurate.
The newer version of the problem is transcription. Call recorders are now writing next steps and assumptions into the CRM, and when they get it wrong, every rep and every system downstream inherits the error. A team goes from one extreme, nothing logged at all, to the other, a large volume of confident and incorrect notes. Each handoff makes it worse. "If you don't have the basics right, then that's just going to be amplified with AI."
Speed Up the Research, Slow Down the Conversation
Brendan raised something he had been seeing: teams switching parts of their AI off, because in principle it should work and in practice it was burning good leads. Mark reads it differently. Not a switch-off, but teams becoming more intentional about which steps they automate, and a much wider understanding of human in the loop.
His test is to put himself in the BDR's shoes. A thousand prospects land in a sequence. Where does the rep start? What is the talk track, the opener, the angle most likely to land with this particular person? The answer is to find the places to speed up and the places to slow down. Let AI produce a research brief with five or six possible angles. Then let the rep read it, choose one, and use their judgement on the channel.
What does not work is treating personalisation as a box to tick. "I see that you like cats. I like cats too. By the way, do you want to book a demo?" Mark's word for it is fake personalisation, and his best example was a friend who is a lifelong Arsenal fan receiving an AI-written email congratulating him on Tottenham's win in the north London derby. It did not go down well.
Niall split sales teams into two camps. A small number of AI-native reps run sophisticated sequences and automated replies with real success. Most reps arrive at work, plug into the system, and want help doing their job while keeping ownership of it. What frustrates them is not AI itself. It is AI that slows them down or gets in their way. Leslie Venetz made the same argument from the buyer's side in Episode 2: a signal earns curiosity, not permission.
Context Is a Spreadsheet Nobody Has to Read
Context engineering is giving AI agents the rules and guardrails they need to work inside. Brendan's question was whether it is the next big topic, now that the marketing team and the sales team no longer sit side by side and pass context over the desk. Mark's view is that context was always needed. It just matters more once work is handed to agents.
He breaks it into two layers. Company-level context: your ideal client profiles, the segments within them, the personas in each segment, their pains and motivations, how your value maps to each one, and the proof that you have solved the same problem for a company like theirs. His advice on format is liberating. Put it in a spreadsheet, in one big matrix. "It doesn't matter that it's ugly. It doesn't matter that it looks overwhelming, because you're not the one reading it."
Then playbook-level context. Marketing went to four events in October and came back with four lists of scanned badges. Dropping those lists into Salesforce and asking reps to work them is hard, especially for a rep who was not there. The rep can look up the event website, but that tells them almost nothing. The playbook needs to say why the list exists, not just what is on it.
Niall added a third layer: feedback. The people who say no, or unsubscribe, or never open the email at all, are context too. Sometimes the proposition was fine and the timing was fine, and the prospect had simply been hit ten times by an automated sequence. If nobody reviews why accounts did not respond, and nothing feeds that back into the engine, a single avoidable point of friction quietly kills deals that should have worked.
What Stays Human
People buy from people, and relationships run on deposits and withdrawals. That was Niall's starting point when Brendan asked what stays with humans for the foreseeable future. The nuance AI still struggles with is reading the emotion in a situation. Is this person angry with me, with the company, or with something completely unrelated, like a sick child at home making it impossible to concentrate?
He did credit AI where it is working. His example was cancelling a SaaS product he was furious with, through an offboarding conversation run on Ada, and coming out with a better opinion of the brand than he went in with. "It's only a shame that the product itself didn't work as well." His point was that AI is excellent at the repeatable work that enables a conversation. What it lacks is the person who has knocked on a hundred doors, heard 99 noes, and can still smile and find a way through.
Mark's example of the human in the loop was coaching. Across the four or five organisations he is working with at the moment, he said about 70% of meetings are booked from cold calling, so the work is improving reps' data, talk tracks and confidence. He uses AI to analyse every call recording weekly and break down objections and call dispositions. When it shows that a rep is giving up at the first objection, the next step is not a training module. It is a conversation: what does that objection mean to you, how are you internalising it? Half the time, the prospect just wants to get off the phone, and it is not personal.
Tenured reps handle this better. When a prospect says they use a competitor, an experienced rep does not trash the competitor. They ask a question that plants a seed of doubt, something the rep knows the competitor is weak at, and the conversation moves on to the next stage.
AI Is Closer to Ready for Inbound Than Outbound
The difference is whose intent started the conversation. When a buyer reaches out to book a demo or ask about an integration, Mark said, they mostly want the answer and do not mind much how it arrives, as long as they are not tricked into thinking an AI is a person. When a seller reaches out, the bar is much higher. That is why so many AI SDR companies have pivoted to inbound.
Niall agreed. Many of us have been conditioned to talk to support agents like Fin by Intercom, as long as a human is easy to reach when we get stuck. Robo-dialling on outbound is a different experience entirely. It cannot adjust to a tone of frustration the way a human SDR would. Doing outbound well is about preparing the human for success, and making sure every sequence matches what has already happened with that person across every channel.
Mark's most practical tip of the episode belongs here. Most of his clients run power or parallel diallers to hit activity numbers. He has them group the call block by persona or job title, then by signal or angle, so a rep dialling twenty prospects is having roughly the same conversation twenty times. Instead of a facilities manager, then a head of operations, then a founder, it is one pattern, repeated and improved. Reps finish the block more confident, because they have built up the reps.
When Outbound Works, Something Else Breaks
A question from the audience asked about an "AI pile-up": AI helps sales produce far more feedback for the product team, and product becomes the bottleneck. Niall widened it. What happens when outbound actually works, and the team has more conversations than it can handle? Success exposes the weak points further down, from trial to close and beyond.
Mark had seen exactly that at a client last year, where customer success and onboarding ran out of hands. The reflex answer was to hire, but a new hire takes two to three months to get fully up to speed. What they did instead was manage expectations first, scheduling onboarding calls or pilots a week or two out, then run the same process review on onboarding that Mark had run on outbound. It turned out the onboarding process had not been evaluated or changed in eighteen months to two years. Fixing it freed up capacity.
Cold Email as Distribution, Not a Demo Request
Brendan confessed to what most of us do with a cleverly personalised cold email: think "clever AI", then delete and block. Mark's response is that smarter teams are rethinking what cold email is for.
The first send is still about relevance: the right person, the right message, the right time. But when there is no reply, the account is often still a good one. It fits the ICP, it scores well, and it showed a signal that matches past closed-won business. Rather than re-enrolling the prospect in a reworded version of the same sequence after forty-five days, those teams move them into a nurture track built like marketing. What problems does this persona have, and how can we help solve some of them even if they never buy from us?
Mark's own version: he could give away a cold email template, because the reader still has to work out targeting, deliverability and the rest of the sequence. If someone does not reply, "I'm not going to take it personally. Maybe the timing wasn't right." He wants them to see his name every forty or fifty days, so that when the problem arrives, they think of him. The catch is that this is a long game, and BDRs measured on monthly meetings and quarterly quota find it very hard to play.
Knowing When Not to Reach Out
Signals are not scarce any more. Restraint is. Brendan handed Niall the last word on timing, and Niall's answer was that the hardest skill now is deciding when not to act on a signal. Teams are bombarded with them across multiple platforms, and the temptation to dial ten more because there is a number to hit is constant. "Knowing when not to reach out is a practice in and of itself."
His question back to Mark was who protects that discipline once the consultant leaves. Mark's answer: it depends on the organisation. Sometimes RevOps, sometimes marketing, sometimes a global head of sales development. Usually it is the champion he worked with, the person who already understood that things needed to change.
Then Mark offered a tip that, as Niall pointed out on air, sounded like MarketSizer had paid him to say it. (We did not.) When a prospect says they use a competitor, train reps to find out when they started and when they renew, and put that in the CRM as a trigger. One of his clients found that most customers churned from a particular competitor between six and eight months in, which made that window the moment to reach out. "It's the holy grail. It's very hard to find."
That is the problem MarketSizer works on: subscription intelligence, which detects subscription lifecycle events such as likely trials, switches and removals from observed vendor installations, and estimates renewal windows from when a product was first detected. Niall's caveat was that the information only helps if it is used responsibly. Deliver it in a way that does not leave the prospect wondering whether they are being watched, respect the boundaries of the person you are contacting, and do not trash the other vendor, because it backfires. The timing play itself is covered in our guide to winning back churned customers on competitor renewal timing.
The data and the system are two halves of the same job. As Niall put it in closing, a company can have the data, but without the systems and capability to act on it, it will not have much impact. Which brings it back to Mark's line. Automate the process. Keep the conversation for yourself.
Intent, Decoded is MarketSizer's podcast series on the mechanics of purchase intent and what it actually takes to use it well. Watch Episode 4 on YouTube, listen on Spotify, or catch the full LinkedIn Live recording here. Mark Colgan runs Yellow O and is on LinkedIn. Follow the show on Spotify or YouTube to catch future episodes, or visit the podcast hub for everything in one place.
Recommended reading
- Episode 4: Are We Automating the Wrong Parts of Sales? - the full conversation, with chapters, quotes and the video.
- Why CRM Data Is Inaccurate - the dirty-data problem, and why AI makes it louder.
- Sales Outreach Not Getting Responses? - what to fix before you add volume.
- What Is Subscription Intelligence? - the data behind competitor tenure and renewal timing.
- B2B Winback: Recover Churned Customers on Competitor Renewal Timing - the timing play behind Mark's closing tip.
- When Every Tool Has the Same Signals - the Episode 3 recap with Jack Porter, on signals as a multiplier.
- From Signal to Trust - the Episode 2 recap with Leslie Venetz, on earning the right to a buyer's time.
- Most GTM Teams Use Intent Data at the Wrong Time - the Episode 1 recap.
- Intent, Decoded - the podcast hub - every episode in one place.