Insights · Article · November 17, 2026

GTM Advice I'd Push Back On: Outbound and Selling After the Signature.

Part two: outbound and what happens after the signature. Why AI personalization stopped working, what replaces the sales relay, and who owns the customer once revenue follows usage.

Larry Chao ·

A crew member with binoculars on the wing of a cargo ship's bridge as a pilot boat pulls alongside at sea.

Two things changed at the same time. AI made sending almost free, so every buyer's inbox is full of messages that look personal. And usage and outcome pricing moved much of the revenue past the signature, so a closed deal is the start of the money, not the end of the sale.

Together, they move the advantage to three places: timing, the conversation, and who owns the customer after the contract. The first post in this series covered inbound and the foundations underneath it. This one covers outbound and selling after the signature, with the same format: the common advice, where I'd push back, and what I'd do instead.

1. "Personalize every message with AI."

The standard playbook is to have AI pull a prospect's college, their latest post, and their company news, then write a first line that could only be sent to them. A few years ago that stood out. Now every tool does it, and buyers have learned the pattern.

Conor Paulsen of Uptown described a sales leader's version of this:

"Mentioning someone's alma mater in the first line is the most automated thing you can do on LinkedIn."

His example is a single-rep test, not a benchmark, but the direction is consistent with what other practitioners report. Kelly Arndt, summarizing a session by Brendan Short, put it plainly: AI personalization is, at best, flashy. At worst, it is inaccurate, or the same thing every competitor is sending.

What buyers respond to is timing and relevance. A message that reaches someone the week they started looking for a solution, with an offer that helps, beats a clever first line sent at the wrong time.

What to do instead: spend your effort on the signal (why this company, why now) and the offer (what you can give before asking for time). Keep messages short. Skip the compliments and the paragraph about yourself.

2. "Outbound is a numbers game."

The numbers-game version of outbound says more lists, more sequences, more sends. When sending was expensive, volume was a real advantage. Now anyone can send a thousand messages a week, and buyers have stopped answering. Teams that relied on bought lists and cold sequences list them among their least successful experiments.

The alternative is smaller and more deliberate. In his signal-based "micro campaigns," Brendan Short suggests three steps. Find a signal that shows a company is in enough pain to look for a solution, such as a company posting publicly about a new compliance requirement. Limit the campaign to 50 to 150 contacts, few enough that you know who each person is and why you are writing. Then lead with an offer that gives value before asking for time: a benchmark, an audit, a comparison.

Kelly Arndt summed it up in four words:

"Timing + relevant offer = pipeline."

Measure the right thing, too. Replies show interest. A paid pilot shows demand. A campaign with a high reply rate and no pilots hasn't proven anything yet.

What to do instead: start from the trigger, not the data. Name the event that means a company might buy, find where you can see it, and run one small campaign per trigger. Expand only the campaigns that turn into paid work.

3. "AI SDRs can replace your SDRs."

The first wave of AI SDRs promised to write to strangers in cold accounts at scale. Adam Schoenfeld's review of how that went is blunt: low response rates and high churn. One vendor retired its "Stop Hiring Humans" campaign and hired a human BDR, and some customers publicly fired their AI SDRs.

The second wave works better because the job is narrower. These agents qualify buyers who already raised a hand, using approved facts, clear rules, and a calendar, then hand off to a person to close. Vercel turned a top SDR's process into an agent in six weeks, for about $5,000 a year. ElevenLabs qualifies inbound leads in a median of four minutes instead of two days.

The lesson is not that agents don't work in sales. It is that they work best on bounded jobs with a clear definition of done.

What to do instead: give agents the inbound and follow-up work first. Keep a person on the first cold touch, especially for the accounts that matter most. If you are a founder, write down how you qualify a lead before you automate it; that document is the agent's job description.

4. "Let agents run GTM end to end."

The pitch for full autonomy is appealing: agents find accounts, write and send messages, handle replies, and book meetings while the team sleeps. Some of that should run unattended. All of it shouldn't, at least not yet.

A useful way to decide is to sort the work by what a mistake costs. Monitoring, enrichment, scoring, drafting, and sorting replies can run on their own, because an error costs a correction. A first message to a named account, anything sent to a target logo, and any claim that would be expensive to retract should be reviewed by a person first. And for the twenty or so accounts that matter most, a person should write the first message, because the whole point is that it wasn't automated.

The boundary should move on evidence, not confidence. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, on cost and weak controls. Widening what agents do as their error rate drops is how you avoid being in that group.

The same logic applies to customers buying an AI product. Autonomy is their decision, granted step by step: observe, then assist with approval, then a bounded task, then a wider scope. Each step needs evidence, and with usage pricing each step up is also how revenue grows.

What to do instead: write down what fires without you, what a person reviews first, and what a person writes. Review the list monthly against the error rate.

5. "Build the SaaS sales relay early."

The standard SaaS sales team is a relay. The SDR qualifies leads and books a meeting. The AE discovers, demos, closes, and hands off. The customer success manager onboards the customer and prevents churn. The account manager looks for expansion. Many founders copy that structure as soon as they can afford a second hire.

Jacco van der Kooij of Winning by Design points out why it existed. The relay was built to sell subscriptions at high volume and low cost, when inbound leads were plentiful and the money arrived at signature. Both conditions are fading. Inbound has dropped for most companies, and with usage pricing, the AE who closed the deal is gone by the time the revenue shows up. Something gets lost at every handoff, and expansion is found late, by someone new.

Mark Roberge made the historical point in "Why AI Is Bringing Back the Full Cycle Seller": before the SaaS era, one salesperson set their own meetings, closed their own deals, and managed their own customers. The role was split because no one person could do every stage well. AI takes on much of the research, preparation, and admin, which makes one owner possible again. As Michael Hartman, who leads GTM at Anthropic, put it:

"AI is undoing 20 years of sales specialization."

What to do instead: make your first sales hire someone who can open, close, deliver, and expand an account. Split the role only when deals repeat and the volume truly needs it.

6. "Hire a sales leader to take sales off the founder."

This advice usually comes with good intentions: the founder should focus on product and fundraising, so hire someone who knows how to sell. The problem is timing. Before deals repeat, there is no process to hand over, only the founder's judgment.

In the early days, the founder learns things no one else can. You run discovery, demos, and the first negotiations yourself. You write down the buyer's words, objections, and reasons for losing. You test the same offer with similar customers and separate what repeats from what was a one-off. You hand off only when someone else can reproduce the process. AI can prepare the call, but the founder still needs to hear the answer.

Maja Voje (The GTM Strategist) and Alex Lindahl (Clay) put the principle well in their GTM engineering roadmap:

"Do it manually first and earn the right to automate."

The same applies to hiring. Do it yourself first, and earn the right to delegate.

What to do instead: keep selling until the process is written down and someone else can run it. When you hire, hire for the full cycle (see the previous section), and tie part of their pay to usage and expansion, not only to bookings.

7. "The deal is done at signature."

Under perpetual licenses, the seller was paid at signature and moved on. Under subscriptions, revenue was spread over the contract, but seats were bought up front and the risk was renewal. Under usage and outcome pricing, little revenue arrives at signature. It follows adoption. If the customer doesn't use the product, there is no revenue, regardless of what the contract says.

That changes what selling is. Jacco van der Kooij borrows a line from Amazon's engineering history. In 2006, Werner Vogels ended the practice of developers throwing code over the wall with "You build it, you run it." Jacco's version for sales is six words:

"You sell it, you own it."

He describes how owning the customer after the sale changed his own selling: he stopped selling false hope, he looked for the outcome instead of the pain, and he told customers about the conditions and assumptions up front.

The time to do this comes from AI. Winning by Design estimates that a seller has about 1,600 selling hours a year, with less than 30% spent with customers. Taking back much of the rest (research, preparation, proposals, admin) is what makes it possible to stay with accounts after the close.

What to do instead: decide who owns each account after signature before you close it. Give that person the usage data and the time to act on it: adoption check-ins, risk flags, and expansion conversations.

8. "Default to seat pricing."

Seat pricing is the habit SaaS taught us: count the users, multiply by a monthly price. Cost-plus pricing is the habit hardware taught us. Neither fits a product whose value comes from work it does on the customer's behalf.

Start from the value instead. Economic value is what a customer can do with you, minus what they can do without you, in dollars per customer per year. Then choose what to charge for. Bessemer's AI pricing playbook describes four options for an AI agent:

Charge perHow it worksTrade-off
Agent or seatA flat subscription, like hiring a personThe customer carries the idle risk
ActivityMetered on actions or tokensTracks your costs; the customer feels every token
OutputPer finished deliverable, such as a processed claim"Done" has to be defined up front
OutcomePaid on a measured business resultNeeds attribution and cost control

Four questions help choose. Do you save the customer money or make them money? Can you anchor the price to a KPI they already track? Is usage predictable, or do you need caps and tiers? And can you control the outcome, or only part of the workflow?

Whatever you choose, the baseline becomes the meter. Pick a unit the buyer already counts, such as tickets, invoices, or defects. Measure today's volume, cost, and error rate before you deploy. Agree in writing how the unit becomes a bill, and add a cap or floor if usage swings. If you can't measure the baseline, you can't defend the price.

What to do instead: price on the difference you make, in a unit the customer already counts, and review it with them every month against the baseline.

Wrong too early, not wrong forever

None of this advice is wrong forever. Volume outbound still works for some high-velocity products. Specialized sales roles make sense once deals repeat and volume demands them. Seat pricing fits some products well. The mistake is reaching for them before the business is ready, or without noticing that the conditions they were built for have changed.

If you take one thing from this post, make it ownership. Decide who owns each customer after the signature, give them the time AI frees up, and measure them on what the customer actually uses. The rest follows from that.

Sources

  • Conor Paulsen, Uptown, LinkedIn post, November 2026 (a single-rep test, unverified)
  • Kelly Arndt, LinkedIn post summarizing Brendan Short's micro-campaign session at the Demand Collective summit, November 2026
  • Adam Schoenfeld, review of AI SDRs (2026), including the Vercel and ElevenLabs examples
  • Gartner, forecast that over 40% of agentic AI projects will be cancelled by the end of 2027 (2025)
  • Jacco van der Kooij, Winning by Design, LinkedIn series on the full-cycle seller, October to November 2026, including "You sell it, you own it" and the Michael Hartman quote
  • Mark Roberge, "Why AI Is Bringing Back the Full Cycle Seller," September 2026
  • Maja Voje and Alex Lindahl, GTM Engineer roadmap 2027, November 2026
  • Bessemer Venture Partners, "The AI Pricing and Monetization Playbook", 2026

By Larry Chao, Sovereign Pivot

← Back to Insights