The full keynote from the RevOps Festival, London.
Why I'm giving this framework away
When we implement agentic AI across a go-to-market team, we always look at the same six dimensions. Not because it sounds nice on a slide, but because every failed implementation we've seen skipped at least one of them. This is the exact framework from my keynote at the RevOps Festival in London. Steal it. Run the audit on your own team.
Quick context on why I care about this so much. I come from sales. I was the fourth sales hire at Trustpilot, back when e-commerce was just getting started. We went from 50 to 5,000 customers in two years, and I brought 600 of them myself. Almost one closed deal every single day. Part of the motivation was that my first startup, an e-commerce company, went very successfully bankrupt in 11 months and left me with eight years of debt. Trustpilot had an uncapped commission model. You can connect the dots.
But the real reason I closed so much: I researched every single company before I called them. I was talking to founders of really successful e-commerce businesses, and within a minute the conversation was about their online reputation, their funnel, their world. If you can describe your customer's problem better than they can, they will assume you have the solution. You cannot do that without research.
Years later I ran a 70-person consulting business where teams did exactly that research, with a very precise process, for sales teams. In November 2023 we started experimenting with AI, and by accident we built a spreadsheet where the cells were prompts feeding other prompts. That ugly little spreadsheet reproduced two weeks of research in 20 minutes. I remember staring at it thinking two things at once: "this is amazing" and "we just killed our business." That's when we became a product company.
Today we are 12 people and 84 AI agents. I'm pretty sure that within a few years, introducing your team will mean introducing your human colleagues and your digital colleagues. And on the platform, our customers are running more than 50,000 agents per day.
So here are the six dimensions. For each one, I'll give you the audit question to ask about your own team.
Positioning and sales motion
Everything we do is account based. You might have leads and contacts floating around, but at the end of the day you are chasing accounts. An account is the company you sell to. Inside it you have one or more contacts, and between those contacts you have relationships. That little triangle defines everything downstream.
Then comes your sales motion, and this is where most implementations go sideways before they even start:
- High velocity: reps open 30 to 60 new accounts per week against a TAM of 5K to 10K companies. One or two contacts per account, $5K to $50K deals, 7 to 30 day cycles.
- Mid market: 12 to 25 new accounts per week, TAM of 1K to 5K+. Two to ten contacts, $30K to $100K deals, 30 to 90 days.
- Enterprise: no new accounts every week. A book of business of 10 to 20 accounts total, each with 10+ contacts, $50K to $500K+ deals, 90+ day cycles.
You do not implement AI the same way for a high velocity team and an enterprise team. And plenty of companies run both motions at the same time, which means two different agentic setups, not one. Your motion defines your workflows. Get this wrong and every dimension after it inherits the mistake.
Can everyone on your team say, in one sentence, which sales motion each segment runs and how many contacts per account that implies? If not, start here.
Agents, sales teams and tools
Agentic AI feels like déjà vu to me. Dan Ariely got famous during the big data years for saying that big data was like teenage sex: "everyone talks about it, nobody really knows how to do it." Replace big data with agentic AI and it's 2026. Everyone claims they're doing it, and we're not even using the same words.
So let's align on words. To us, an agent is a digital colleague. You can talk to it, the same way you can talk to your marketing colleagues or your CFO, even when you'd rather not. It can be trained, you control the training, and it improves based on it. It can run multiple tasks autonomously. And it can work with other digital colleagues and with human colleagues.
What an agent is not: a chat interface on top of an LLM. A custom shared GPT. Chat, wait, repeat, chat, wait, repeat. That's an assistant, and assistants are fine, but they are not digital colleagues.
What most teams have today: AI bolted everywhere, zero agents
Look at a typical revenue team. Every function has its own point solutions, and those tools integrate with your CRM, your sequencer, your dialer. But they push data, not context. RevOps has its own AI automation tools that the sales team never touches, so reps use ChatGPT on the side, and now you have two more silos. Every vendor has bolted AI into their product, so you have AI isolated in twenty different places and zero agents on the team.
The premise sold to reps was: more tools, more productivity. It never happened. Salespeople still spend most of their time not selling, no matter how many tools and how much bolted-on AI you stack up.
What a hybrid GTM team looks like
The alternative is agents and humans working together, structured in two layers.
The orchestration layer holds your agents, organized in three categories: research, personalization and coaching. One or more people on the team own this layer, typically RevOps. The intelligence layer is what feeds it: your ICP verticals, your persona cards, your value proposition. The same way you would train a new sales hire, you train your agents. The difference is that agents remember 100% of the training.
Instead of pushing data around, agents now push context, and that context lands where the team already works: in the CRM, the sequencer, the dialer. Your CRM stays the source of truth, and the agents keep it current. The same logic extends to the chat tools through MCP. If your reps call trained agents from ChatGPT or Claude, the output is as good as your orchestration layer, not as good as each rep's prompt. That last part matters, because the alternative looks like this: a rep pastes an account into ChatGPT and asks "is this a good fit for us?" And ChatGPT, being your best friend, answers "yes, this is the best account I have ever seen."
Audit your stack on two axes
Here is the take-home exercise for this dimension. Map every tool in your stack on two axes: multi-user vs. single-user, and data-driven vs. context-driven.
- Data-driven, multi-user (lead databases, and the CRM and sequencers around them): the databases were great for a decade. You pay a license per rep so they can filter 600 million contacts. But agents can do that filtering now, and agents don't need seats, they need APIs. That's the part of the stack you can replace. Your CRM and sequencer stay, integrated via API, as the system where context lands and reps act.
- Data-driven, single-user (AI automation tools): AI on top of data, operated by one person. This is where new silos come from. Replace.
- Context-driven, single-user (AI assistants): your team will use ChatGPT and Claude whether you allow it or not. Don't block them, integrate them via MCP.
- Context-driven, multi-user: this is the agentic workspace quadrant. Build your agents here.
TAM and agentic playbooks
Your total addressable market is not unlimited, so let's put numbers on it. Say your TAM is 5,000 accounts. Four thousand of them are not in your CRM yet. That white space is your newbiz gap. The rest are in your CRM: some working, some in opportunity, some clients. And because we don't live in la la land, most accounts don't complete that journey. They go dormant.
This is the part most sales organizations get wrong. Once they've drained their go-to-market playbook they say "we ran out of leads." You didn't run out of leads. Your leads are sitting dormant inside your CRM.
That gives you three account-based playbooks. If you run high velocity or mid market, you'll run all three. If you sell enterprise, your accounts are already in the CRM, so recycling is essentially the only playbook you run.
Playbook 1: the newbiz gap
Push accounts in from wherever they live: a LinkedIn or Crunchbase company search, an external CSV, your CRM. An Account Qualification agent does the reasoning: is this company actually ICP? Do they do the specific things that make them a fit? Then an Account Research agent scores them on your signals, with a gate: only accounts with signals found or a score above your threshold move on. A Contact Finder agent works with a Contact Qualification agent and waterfall access to contact data vendors to find the right people with verified emails and phone numbers. And a Play Copywriting agent reads everything the other agents produced and writes the outreach. The output lands in your sequencer and CRM, where your reps take action.
Playbook 2: newbiz recycling
Define dormant with rules your CRM can evaluate: does the account have an open opportunity? No. Any activity in the last X days? No. Any open task with a future due date? No. Then it's dormant, and you can recycle it: search for new contacts, requalify the existing ones, regenerate outreach based on fresh research. Same workflow shape as the newbiz gap, just pointed at your own CRM instead of external databases.
Playbook 3: ex-customer recycling
Here's a number that surprises people: 3 to 4% of your contact data goes stale every single month, because people change jobs. So run a Contact Qualification agent on your ex-customers and champions: is this person still working at the account? If yes, stop. If no, check whether their new company fits your ICP. If it does, update the CRM, move the contact, and hand your rep a warm lead. Your champion just landed somewhere new with budget and a problem they already know you can solve. The same signal is a red flag for your CS team on the account they left.
And all of this can be scheduled. Workflows and single agents run in the background: monitor account-level signals, monitor contact-level signals, check employment status monthly, create CRM alerts. Autonomous, not chat-wait-repeat.
What percentage of your CRM is dormant right now, and which playbook is systematically working those accounts? If the answer is "none," that is your cheapest pipeline.
Top of funnel, channels and activities
A quick history of outreach channels, because we keep repeating the same cycle. When I started in sales, fax was DocuSign. We had a fax machine on the sales floor, and when it started shaking, salespeople would literally run towards it hoping their signed contract was coming through. It used to be a channel. Then it died.
Then Aaron Ross wrote Predictable Revenue, which evangelized the SDR model and the email channel. Back then, receiving an email was exciting. People thought I was a genius because I could guess someone's email address from their LinkedIn profile. Today there's a cheaper ZoomInfo every single day. Everybody has the same data. There is no edge left in access.
Then COVID pushed everyone onto the only channels available, email and LinkedIn, because we didn't have direct dial data yet. Both got noisy. And then AI arrived and, honestly, it has done more damage to the email channel than anything positive: very cheap GenAI copy on top of very cheap data. For buyers it's so much noise that even a genuinely relevant cold email is unlikely to be seen. Google and Microsoft had to fight back and protect their users, so deliverability got harder on top.
The new blue ocean is the phone. But there's a catch: we now have a generation of salespeople who did not grow up with a landline. My youngest brother is 11 years younger than me. He literally doesn't have the green call button on his phone home screen. He has never called a restaurant to book a table. And then sales leaders look at this generation and say they're lazy. They're not lazy. They've never used the tool. Train them to be confident on the phone and they can be extremely good at it.
The seven reasons you have a pipeline issue
None of them are new, and none of them are about AI:
- Quality of your list (accounts and contacts)
- Quality of your outreach (email copy, talk track)
- Timing (calendar-driven or signal-driven)
- Spam issues (mailbox or phone number reputation)
- Activities (number of calls per week)
- Connecting ratio (dials vs. connects)
- Average talk time (connected vs. talk time)
90% of the time, it's a list quality issue. Fix the list first, then work down the stack step by step.
And the metric almost nobody looks at: talk time. We measured this across our customers. If a cold call passes one minute of conversation, with the right researched company, you have a 50% chance of booking a meeting. People keep their guard up for the first 20 seconds of a cold call. If those 20 seconds are research-based, about them, the guard drops. Then you have to win the next 40 seconds. That's the whole game.
Data-driven sales vs. context-driven sales
The old way to define an ICP: 1,000+ employees, industry is manufacturing. That's data. The new way adds the checks that actually predict fit, and gives them to agents:
- Selling 3D software to manufacturers? Check they manufacture a technical product that needs visualization, then research whether they launched a new product or exhibit at trade shows.
- Selling transfer pricing data? Check for active cross-border intercompany flows, then look for hiring in transfer pricing or tax roles.
- Selling customs compliance software? Check the company is a trader, importer or exporter, but not a broker, then look for tariff exposure or geography expansion.
Data-driven outreach was the mad libs email: "Hi first name, congrats on data variable." Every buyer can call it out instantly, and putting AI on top of it fails 100% of the time, because AI cannot personalize on data points it doesn't understand. Context-driven sales flips it: agents qualify, research, and store the context they collect. That context powers two kinds of outreach, evergreen qualification-based plays and signal-based plays. And it powers the phone: a research-based cold call opener, research-based talking points, calibrated pain questions, all written in your tone of voice by the Play Copywriting agent, because the research was already done and documented by the agents upstream.
Put together, this is how you fix the pipeline issue with a team of agents: cherry-pick the best leads with qualification agents and contact finders. Prioritize with the signal score from account research. Increase your connect rate with a phone waterfall plus TitanX, which predicts which number is most likely to pick up and gets 20 to 30% connect rates on priority-one numbers. Then increase talk time by combining the account research with the play copywriting. Less volume, way more quality, and you have the luxury of cherry-picking who you call.
Change management
This is the dimension that decides whether everything above actually happens. Most failed AI implementations, GTM or not, fail on change management. Not on the model, not on the tooling. On habits.
The order matters more than anything:
- 1. Learn how to work with agents (today). Unlock usage first. Reps working with existing, trained agents on real accounts.
- 2. Build new habits (today). What do I do with my agents on Monday morning? What do they run over the weekend?
- 3. Create and tweak agents (next). Only once one and two are in place.
I can tell you the order matters because we made the mistake ourselves. We onboarded customers starting at number three, "let's build your custom agents together," and it didn't work. When we flipped it and started with one and two, long-term adoption got dramatically better.
The other half of change management is being explicit about which habits die. Building lists with industry and company-size filters plus manual research on top: gone, 100% outsourced to agents that do it better, faster, and trained on your ICP. Generic signal search: goes to Account Research. Contact search by job title: goes to Contact Finder. Manual CRM entries: gone, and honestly this is the best time ever to work in RevOps, because agents can finally keep the CRM updated in the most perfect way. Account plans: generated from the research. And everything the agents produce gets turned into something a rep can act on.
- They used to spend 30 to 40% of their time filtering and cleaning lists. Now they get a daily cherry-picked list of ICP-fit, persona-fit leads with the signals already researched.
- High list quality means high confidence and faster ramp-up. Quota pressure used to degrade list quality right when reps needed it most.
- Research-based personalization means a higher connect-to-conversation rate, and quality pipeline means quota. That is the outcome you sell internally.
Future proofing your team
Once the first five dimensions are in place, ask three questions about whatever setup you've built:
- Can the agents work across the entire customer journey? I've mostly talked about top of funnel here. But when a full-cycle rep hands an account to CS, the CS team should keep working with the same agents, trained on the same intelligence layer. Same for the rest of the journey.
- Can the agents access your first-party data? Product usage, website interactions, everything happening inside your company, not just what's visible on the internet.
- Can your agents interact with external agents?
That last question is a prediction I'm confident about. Sales is a two-way street, and everyone in the room at these events is optimizing our side of the street. On the other side, buyers are building their own agents right now. A buyer will brief a procurement agent: "Find a CRM. Budget 150K. Timeline 30 days. Must integrate with HubSpot." That agent contacts eight vendors in parallel, sends RFPs, sits through demos, and ranks the candidates.
It's going to happen. Agents will show up in your discovery calls and say: "Hi, I'm an AI agent. Skip the small talk, this is my checklist." And they can do 800 calls per day. Imagine that volume hitting your reps' calendars.
We built our agents for that world from the get-go. A buyer-side agent reaches out, our Account Research and Qualification agents build context and qualify or disqualify, the agents run the discovery between themselves, and when there is a mutual fit, the humans come into the loop to close. That's how you absorb agentic buyer volume without overwhelming your team and without losing leads.
Run the audit on your own team
Six dimensions, one honest audit: your positioning and motion, your agents and tools, your TAM and playbooks, your top of funnel, your change management, and your future proofing. Score yourself on each. In our experience the gaps cluster in two places: the stack audit in dimension two, and change management in dimension five.
90% of the time, a pipeline problem is a list quality problem. Fix the list first, with agents trained on your ICP, then fix the rest step by step.
If you want a structured version of this exercise, we turned the framework into a free self-assessment: the Agentic Readiness Score. It walks you through the six dimensions and shows you where you stand against best practice, and what to fix first.
Frequently asked questions
What are the six dimensions of an agentic GTM team?
Positioning and sales motion, agents and tools, TAM and agentic playbooks, top of funnel and channels, change management, and future proofing. We audit all six before implementing agentic AI on any go-to-market team, because a gap in one dimension usually cancels out the gains from the others. A team with perfect agents and no change management still fails. A team with great habits and a data-only stack still sends mad libs emails.
What is the difference between an AI agent and a chat assistant like ChatGPT?
An agent is a digital colleague: trained on your ICP, personas and value proposition, improving based on that training, running multiple tasks autonomously, and working with other agents and with humans. A chat interface on top of an LLM is an assistant. Assistants are useful, and your reps will use them regardless, which is exactly why they should be connected to your trained agents via MCP rather than left to freestyle prompts.
How do I audit my GTM stack for the agentic era?
Map every tool on two axes: multi-user vs. single-user, and data-driven vs. context-driven. Seat-based data tools can be replaced, because agents need APIs, not licenses. Your CRM and sequencer stay as the system of action, integrated via API, with agents pushing context in and keeping the data current. AI assistants get integrated via MCP. Your agents get built in the multi-user, context-driven quadrant. The Agentic Readiness Score runs this audit for you across all six dimensions.
Why do most agentic GTM implementations fail?
Change management. The technology is rarely the problem. Teams start by building custom agents before anyone has learned to work with agents at all, so nothing sticks. The order that works: learn to work with existing agents first, build new daily habits second, create and tweak custom agents third. We made that mistake ourselves, started at step three, and saw adoption improve dramatically once we flipped the order.
What is context-driven sales?
Data-driven sales filters on static data points, employee count, industry, region, and sends templated outreach with data variables. Context-driven sales uses agents to research each account against the criteria that actually predict fit, stores that context, and turns it into research-based openers, talking points and calibrated pain questions. Data tells you a company has 500 employees. Context tells you they just opened a new region, are hiring for the role you sell to, and why now is the moment to call.
