Where sales agents stop and people still start

· 5 min read
AI-generated image: Where sales agents stop and people still start
AI-generated image

Three reports crossed the wire today, and they all circle one question: how much of selling can software run on its own, and where does a person still have to step in. Each describes agents absorbing part of the sales motion while the last, highest-stakes act stays with a human, which is exactly the line a buyer needs to find before signing anything.

Start with the shape of a sales process, because the news only makes sense against it. Most teams run a funnel with distinct jobs. Prospecting, often handled by a sales development rep, finds and warms leads. First-line support answers the early questions. Customer success keeps existing accounts healthy. And somewhere near the end, an account executive closes: agrees terms, handles the final objection, and gets a signature. These jobs look similar from a distance, but they are not equally hard to automate. The earlier ones are mostly about volume and pattern. The last one is about judgement under pressure, with money on the table and a human on the other side who can walk away.

If you want to see where your own funnel leaks before you add any software to it, we have written separately about how to find where your business loses leads. That groundwork decides which stage an agent could actually help with.

An AI that co-closes but does not close

The most useful report today comes from a team running agents across its own revenue function. By their account, agents had taken over most of their sales development work, much of first-line support, and a lot of customer success work, but none of them could close [1]. That is a precise map of the current boundary. The tasks that are repetitive, high-volume, and forgiving of a wrong turn have moved to software. The task where one wrong sentence ends a six-figure conversation has not.

The headline detail is that the same system co-closed two six-figure deals in a single week. Read that word carefully. Co-closed is not closed. It means the agent did real work inside a deal that a person ultimately carried over the line. For a business choosing tools, this is the honest version of the promise. You are not buying something that replaces your best closer. You are buying something that does the preparation, the follow-up, and the chasing around that closer, so the human time goes where judgement is needed.

The practical test follows from this. When you trial a sales agent, do not ask whether it can close. Ask what it removes from the hours before and after a close. If it drafts the follow-up, keeps the thread warm, and surfaces the right context at the right moment, it is earning its place even though a person still signs the deal. We have made a related argument about support, where the point of automation is to handle the predictable and route the rest, in what AI auto-reply is for.

The context gap in the pipeline

The second report names a problem that every sales leader will recognise. A new tool called SciFin starts from the claim that every sales organisation runs on a picture of the pipeline that nobody fully trusts [2]. That is the quiet failure underneath most forecasts. The numbers in the pipeline review are a blend of real signals and rep optimism, and the two are hard to separate until a quarter closes and you find out which deals were ever real.

SciFin is described, in its own words, as a sales tool, even where its marketing reaches for a bigger idea [3]. The reported funding behind it is substantial: the founder, who previously started Nutanix and Cohesity, raised forty-four million dollars for it. That figure tells you investors believe the pipeline-trust problem is large and unsolved. It does not tell you the product works for your team, and you should not read it that way.

The concept worth taking from this, regardless of which tool you choose, is the context gap. A forecast is only as good as the evidence under each deal. When a rep says a deal will close, is that backed by an email where the buyer agreed a date, or by a hopeful read of a good meeting. A tool that narrows the context gap pulls the evidence together so the forecast rests on what actually happened rather than on mood. You can start closing that gap today with discipline, not software: write down why each open deal is where it is, and what single thing has to happen next. The tool is only worth buying when doing that by hand stops scaling.

This is the same instinct that makes a shared record valuable everywhere in a business, which is why we keep returning to the idea of one place that holds the full history of a conversation, as in shared team inbox, what it changes. A forecast you can trust and a reply you can trust both come from the same thing: context that is written down rather than remembered.

A working session on what is not working

The third item is smaller but points at something the first two share. A gathering is planned at Granola's HQ in San Francisco on December 8, described as a real conversation about agents for go-to-market, and time to hang out after [4]. The phrase that matters is in the framing: the session is about agents for go-to-market, what is working and what is not.

That last clause is the useful part. Most public talk about sales agents reports only the wins. A session that sets out to cover what is not working is more valuable to a buyer than any launch, because the failures tell you where the current boundary really sits. If you are evaluating this category, the questions raised in rooms like that one are the questions to put to any vendor. Which stages do your agents handle without supervision. Which still need a person. What breaks, and how often.

What a buyer should take from the day

Put the three together and a consistent picture appears. Agents have moved deep into the top and middle of the funnel. They prepare, chase, answer, and assemble context. They have not taken over the close, and the teams furthest along say so plainly [1]. The pipeline itself is still only as trustworthy as the evidence under it [2], and the honest conversations in this field are now about limits as much as gains [4].

So when you choose a tool this quarter, buy for the stage where the work is real and repetitive, and keep a person on the stage where judgement decides the outcome. Measure the tool by the hours it removes around a deal, not by a claim that it can run the deal alone. That is the same measure we applied in yesterday's roundup, the SaaS daily briefing for 13 November, and it holds here too. The agents are good at the many small steps. The close is still yours.

Sources

  1. [1] Our AI VP of Revenue Can’t Close a Deal Yet. But It Co-Closed Two Six-Figure Deals in One Week. — SaaStr
  2. [2] SaaStr AI App of the Week: SciFin. The Nutanix and Cohesity Founder Raised $44 Million to Build AI for Sales Teams. — SaaStr
  3. [3] SaaStr AI App of the Week: SciFin. The Nutanix and Cohesity Founder Raised $44 Million to Build AI for Sales Teams. — SaaStr
  4. [4] Come Meet Us Live at Granola HQ in SF: Granola’s VP of Sales and I Will Be There to Talk Agents for GTM, What’s Working and What Isn’t — SaaStr

Want a person to look at your website and social pages? Get a free check-up, written by a real person within 2 business days.

The 360REV newsletter

What is actually changing across productivity software, written for operators and cited to sources. No more than one email a day.

Double opt-in — we send one confirmation link and nothing else until you click it. Unsubscribe from any edition. We never sell or share your address.