Back to blog

    AI in go-to-market: what to automate and what to own

    AI in GTM. July 23, 2026. 4 min read. Automate the research and the busywork. Own the relationship and the judgment. The teams winning with AI in go-to-market didn't hand it the relationship. They handed it the research and kept the human where trust is built.

    AI in go-to-market: what to automate and what to own

    Automate the research and the busywork. Own the relationship and the judgment. In practice that means let AI do the account research, list building, call summaries, first-draft content, and pattern-finding across your data, and keep a human on the actual selling, the pricing calls, the escalations, and any decision the buyer would feel differently about if they knew a machine made it. The reason to draw the line there isn't caution for its own sake. It's that the data on fully autonomous selling is in, and it's brutal. The teams winning with AI in go-to-market didn't hand it the relationship. They handed it the research and kept the human where trust is built.

    What does the data actually say about automating the sale?

    The autonomous AI-SDR bet, the one that promised to replace the human seller end to end, collapsed in public. 11x.ai, the venture-backed poster child for autonomous outbound, was reported by TechCrunch in March 2025 to be running 70 to 80 percent customer churn, with internal messages putting retention closer to 20 to 30 percent. The company disputed the reporting and put its own retention at 79 percent. Both numbers are on the record, and the distance between them is its own kind of signal. Zoom out and the pattern holds: Gartner expects more than 40 percent of agentic AI projects to be canceled by the end of 2027, on escalating cost, unclear business value, and inadequate risk controls. The market pivoted hard to a hybrid model. AI does the research, a human owns the relationship.

    That's the receipt behind the rule. Nobody is arguing AI has no place in GTM. The argument that lost was that AI could own the part of the sale where a person decides whether to trust you. It can't yet, and pretending otherwise cost a lot of teams a year of pipeline.

    This tracks a broader gap. 87 percent of marketers now run generative AI in at least one recurring workflow, but only about 6 percent of companies can point to more than 5 percent of their profit coming from it. AI is table stakes now, not an edge. It amplifies a motion that already works and exposes one that doesn't. Bolt it onto ghost motion and all you get is ghost motion moving faster. The full breakdown of that gap is in Why isn't AI improving our pipeline?, and the pattern itself is at ghost motion.

    So what should you actually automate?

    The research layer, aggressively. Account and prospect research, list enrichment, call and meeting summaries, first-draft content, and the thing AI is genuinely great at that most teams underuse: finding the pattern in your own data. Stop asking AI to write more emails and start asking it which of your closed-won deals share a signal your written picture of who you're selling to is missing. That's leverage. The rest is volume.

    The test for whether something is safe to automate: would the buyer feel differently about the interaction if they knew a machine did it? A summarized call, no. A pricing negotiation, yes. Automate up to that line and stop.

    What does a human still have to own, and how do you govern it?

    Judgment, relationship, and anything money- or trust-touching. But ownership isn't just a list of tasks a human keeps. In 2026 the serious-operator conversation has moved to governance: who can let an agent influence pricing, lead routing, or outreach cadence, and exactly where human approval kicks in. That's the real ownership question. Not "do we use AI," but "where does the human sign off before the agent acts." Gartner's cancellation causes read like a governance checklist nobody ran, and inadequate risk controls is one of the three reasons those projects die.

    Draw that boundary explicitly and write it down, the same way you'd document any other part of the system. An agent that can research and draft with a human approving anything customer-facing is a force multiplier. An agent with no approval gate touching pricing or cadence is a liability you haven't priced yet.


    The teams that win the AI-in-GTM race won't be the ones with the most tools. They'll be the ones with the clearest underlying motion, the kind AI can actually accelerate, and a written line between what it runs and what a human signs off on.

    The GTM Signal Check shows you where your motion is solid enough to automate and where automating would just scale the mess. Sixty minutes, no pitch.

    Run the GTM Signal Check before you buy the next tool.


    By Eric Glass, Founder, HG Digital. HG Digital is a full-stack GTM firm for growth-stage B2B: we do the GTM work, you run the business.

    Common questions

    What does the data say about automating sales?
    The fully autonomous bet collapsed in public. TechCrunch reported in March 2025 that 11x.ai, the poster child for autonomous outbound, was running 70 to 80 percent customer churn, and the company disputed that reporting and put its own retention at 79 percent. Both numbers are on the record. Gartner expects more than 40 percent of agentic AI projects to be canceled by the end of 2027. The market pivoted to a hybrid model where AI does the research and a human owns the relationship.
    What should you automate with AI in go-to-market?
    The research layer, aggressively. Account and prospect research, list enrichment, call and meeting summaries, first-draft content, and pattern-finding across your own data, like asking which of your closed-won deals share a signal your written picture of who you're selling to is missing. That is leverage. The rest is volume.
    What does a human still have to own?
    Judgment, relationship, and anything money- or trust-touching. The real ownership question in 2026 is governance: who can let an agent influence pricing, lead routing, or outreach cadence, and exactly where human approval kicks in. Draw that boundary explicitly and write it down. An agent with no approval gate touching pricing or cadence is a liability you have not priced yet.
    How do you decide what's safe to automate?
    The test is whether the buyer would feel differently about the interaction if they knew a machine did it. A summarized call, no. A pricing negotiation, yes. Automate up to that line and stop.
    Run the GTM Signal Check