AI Writes Your Cold Emails. It Should Not Pick Your Accounts.
A recent GTM argument holds that AI has not delivered on its promise to sales teams, and the diagnosis lands close to home. The tools work. They draft cold emails that read fine. They build account lists in seconds. They summarize every call before the rep leaves the room. None of that work, by itself, wins a single deal.
The deal-winning call sits upstream of all of it. Which accounts do we chase this quarter, and why these and not the others. That is the decision that moves the number. It is also the decision teams are quietly handing to an AI vendor whose reasoning nobody can see and nobody owns.
The easy work got automated, the leverage did not
Look at where AI actually shows up in a GTM stack. First-touch email. Sequence personalization. List enrichment. Call transcription and summary. These are the tasks that scale cleanly, and they are exactly the tasks that sit downstream of the only choice that matters.
A perfectly written cold email to the wrong account is wasted motion at high speed. A flawless call summary of a deal that was never winnable is a tidy record of a loss. The volume work got faster. The judgment work, deciding where to point the volume, stayed flat. In many teams it got outsourced to a scoring model that ranks accounts and nobody questions.
That is the move worth watching. The account list is no longer a strategy artifact a human defends. It is an output a tool produces. And when a decision becomes an output, the human stops owning it.
”Why now” is a claim, not a score
A good account decision carries a reason. This logo is buying because they just raised, because a competitor signed last month, because their new VP ran our category at her last company. Those are claims a person can state, defend, and be wrong about in a way the team learns from.
A scoring model gives you a number. The number is built from signals nobody on the team chose, weighted in proportions nobody can recite, against an outcome the vendor defined. When the quarter misses, the model has no answer for what it was thinking. You cannot correct a reasoning you never saw. The “why now” became a black box, and the cost shows up two quarters later as a pipeline built on logic nobody can reconstruct.
This is the same trap engineering teams hit with coding agents. An agent that writes code fast is useful. An agent that silently decides what to build, with no human accountable for the choice, is a liability dressed as productivity. GTM is now living the sales version of that problem.
Accountability does not transfer to the vendor
Here is the part teams skip. When you let a vendor’s model pick your accounts, you have not removed the decision. You have moved the accountability somewhere it cannot live.
The vendor is not on the hook for your number. They did not sit in your pipeline review. They will not be in the room when the board asks why the quarter missed. The model produced a ranked list, the team worked the list, and when it underperforms there is no human inside the company who can say “I chose this and here is the reasoning I would revise.” Accountability that cannot be located is accountability that does not exist.
This is the boundary we keep coming back to in sales governance, where deals still close on trust between people (see The Leaner AI Sales Org Is a Governance Org). The pattern is consistent. AI can own the volume that scales. A named human has to own the judgment that compounds, and account selection is the highest-leverage judgment in the funnel.
Keep the model, keep the owner
This is not a case for ripping out the scoring tool. A model that surfaces signals a human would miss is genuinely useful, and the speed is real. The fix is cheaper than that.
Treat the account list as a recommendation, not a verdict. Put a person between the model’s output and the team’s effort, and require that person to attach a stated reason to each account that made the cut. Not a score. A sentence a human will defend in the next pipeline review. The model proposes; the human decides and signs their name to the decision.
When the quarter is reviewed, you then have something to learn from. You can read the reasons that worked against the reasons that did not, and tune both the human judgment and the signals you feed the model. A black box gives you none of that. It gives you a ranked list and a shrug.
Do this now
Open your current target account list and ask one question of it: who decided these, and can they tell you why each one made the cut. If the honest answer is “the tool ranked them,” you have ceded your highest-leverage decision to a vendor with no stake in your number. Name an owner for the list this week. Require a one-sentence “why now” per account, written by a person, reviewed at quarter close. Let the model do the cold emails and the summaries. Keep the call on which accounts to chase where accountability can actually live.
This analysis builds on Why AI for GTM Hasn’t Delivered (Cam Wright, June 2026), a commentator’s argument that AI automated GTM busywork without moving the decisions that win deals.
Victorino Group helps teams keep accountable human ownership over the AI-assisted decisions that move revenue. Let’s talk.
All articles on The Thinking Wire are written with the assistance of Anthropic's Opus LLM. Each piece goes through multi-agent research to verify facts and surface contradictions, followed by human review and approval before publication. If you find any inaccurate information or wish to contact our editorial team, please reach out at editorial@victorinollc.com . About The Thinking Wire →
If this resonates, let's talk
We help companies implement AI without losing control.
Schedule a Conversation