Here is a fact about sales teams that everyone knows and almost no one does anything about: the top 10% of reps are much better than everyone else, and nobody can quite explain why. Ask a sales manager and you will get a shrug and a sentence like "she's just really good on the phone." This is not an explanation. It is the absence of one, dressed up as a compliment. And it is expensive, because if you can't say what makes the good rep good, you can't teach it to the other nine.
Knode.ai, a New York company founded in 2023, is built on the mildly heretical idea that "she's just good on the phone" is a measurement problem rather than a mystery. The recordings already exist. The winning behaviors are in there, somewhere, buried in thousands of hours of transcripts nobody has time to listen to. So Knode listens to all of them, isolates the talk tracks and tactics that actually correlate with closed deals, and turns the result into something a manager can hand to a struggling rep on a Tuesday.
The company describes itself, in the flat confident way of startups that have decided on their positioning, as "agentic sales management for seller behavior change." Underneath the phrase is a fairly specific product. Knode ingests a team's sales calls, scores them, benchmarks every seller against the behaviors that its own top performers exhibit, and produces coaching recommendations tied to the specific gaps that are costing each person deals. The pitch to a sales leader is roughly: you do not have to watch the calls, and you do not have to guess what to say.
The word to notice there is own. There is a whole shelf of sales-coaching content that will tell you the seven habits of highly effective closers, and it is mostly fine and mostly generic and mostly ignored. Knode's benchmarks come from inside your business - from the calls your reps are making, selling your product, to your buyers. The "right answer" is defined by what already wins in your specific room, not by a template a vendor bought off a shelf and resold to everyone. This is a small design decision that turns out to matter, because it makes the scorecard feel fair. A rep can argue with a best-practices deck. It is harder to argue with a chart that says the people hitting quota consistently do this thing you are not doing.
The contrarian bet: coach after the call, not during it
There is a fashionable version of this product that whispers in your ear in real time. You are on a discovery call, the AI is listening, and a little card pops up: "ask about budget now." Knode went the other way, on purpose. It analyzes conversations after they happen, on the theory - which is intuitively correct to anyone who has ever tried to do two things at once while a prospect is talking - that people learn better when they are reflecting than when they are being nudged mid-sentence. Real-time assistance optimizes the call you are on. Post-call analysis optimizes the salesperson. Knode is betting the second one compounds.
It is a genuinely contrarian choice in a category racing toward live everything, and it is the kind of decision that tells you a company has a point of view rather than a feature list. It may also be right. The thing a manager actually wants is not a smarter version of the rep on today's call; it is a better rep on the next fifty.
The economics here are the quiet part. Everyone agrees reps should get coaching. The problem is arithmetic: a manager with eight reps and a full pipeline does not have the hours to review calls and write feedback for each of them every week, so in practice most reps get coached rarely, badly, or not at all. Coaching, the one thing that reliably improves performance, is rationed by manager attention. Knode's actual product is not "AI that grades calls." It is "coaching cheap enough to give everyone." That is a more radical proposition than another meeting-notes bot, and it is aimed at a real and boring cost center: the months it takes a new hire to ramp, every one of them expensive.
Who is behind it
Knode was started in 2023 by three co-founders: Riccardo Pellegrini, who runs the company as chief executive; Henry Katz, the chief technology officer; and Nicholas Horelik. The framing they use for the team - technologists, data scientists, and sales experts - is telling, because the hard part of this problem lives exactly at that seam. The machine-learning is necessary but not sufficient; you also have to know what a good sales call actually looks like, and encode that judgment without flattening it into a checklist. It is easy to build a model that counts filler words. It is hard to build one that knows which two minutes of a forty-minute call decided the deal.
The company is small and, by the standards of the genre, refreshingly unhyperbolic about it. It raised a seed round around its 2023 founding, keeps the specifics quiet, and has spent its energy on the product rather than the press release. It carries SOC 2 Type II compliance, which is the unglamorous but load-bearing detail for a tool whose entire function is to ingest every recorded sales conversation a company has - a category where "we take security seriously" has to be more than a sentence on a slide. It also plugs into the existing stack rather than trying to replace it, showing up, for instance, in the Gong Collective integrations marketplace, so teams can point Knode at conversation data they are already recording.
Which brings us to the number the company likes to lead with: more than a million sales interactions analyzed. Marketing stats are marketing stats, but this one points at something real. Below a certain volume, sales advice is anecdote - a manager remembers one call that went well and generalizes wildly. Above it, patterns become statistics, and "you should probably slow down on pricing" turns into "reps who spend ninety more seconds on discovery close at a measurably higher rate." That shift, from vibes to evidence, is the whole promise. Whether Knode delivers it at the scale it claims is the question every buyer has to answer for themselves, on their own calls. Conveniently, that is exactly the thing the product is built to measure.
The publicly referenced customers skew toward high-growth operations - payments company Payabli, on-demand marketplace HomeAglow, PowerToFly, Via, Disco, and others - the sort of teams that are hiring reps faster than they can train them and feel the ramp problem most acutely. That is the natural customer: not the enterprise with a mature enablement machine, but the company scaling fast enough that "she's just good on the phone" has become a genuine operational liability.
The competitive question
Knode is not alone in the room. The conversation-intelligence category is crowded and well funded - Gong and Chorus built large businesses recording and analyzing calls; a newer wave including Clari Copilot, Second Nature, and Hyperbound comes at coaching from the angle of roleplay and real-time assist. What separates Knode is less the raw capability - everyone can transcribe a call now - and more the emphasis. The incumbents are, at their core, visibility tools: they show a leader what happened across the pipeline, which is genuinely useful and also, on its own, a little inert. A dashboard that tells you a rep talked 70% of the time does not tell you what to do about it. Knode's wager is that the market has enough visibility and not enough instruction, and that the valuable half of the problem is the part that begins after you already know what happened.
That is a defensible place to stand, but it is also a demanding one. Instruction is harder to get right than observation, because a wrong recommendation is worse than no recommendation - it burns the one thing a coaching tool cannot afford to lose, which is the rep's trust. Tell a seller to slow down on pricing when slowing down is exactly what lost them the last deal, and you have taught them to ignore the software. The bar Knode has set for itself is not "can it read the call," which is table stakes, but "can it be right often enough that a skeptical salesperson keeps listening." That is a product problem, a data problem, and a psychology problem at once, which is roughly why the company keeps describing its team as sales experts as much as engineers.
None of this resolves in a headline. Knode is a young company making a specific, testable claim - that the pattern of winning is already in your call archive, and that surfacing it changes behavior - and the honest thing to say is that the verdict is still out, deal by deal, team by team. But the claim is at least the right kind of claim: concrete, measurable, and falsifiable on your own data. In a category that often sells the feeling of insight, Knode is selling the thing itself, and inviting you to check the math.
None of this is magic, and Knode, to its credit, does not pretend it is. The bet is smaller and sturdier than magic. It is that the answer to your sales team's biggest question - what do the winners do that the rest don't - is already sitting in your call recordings, and that the only thing standing between you and it is the time to listen. Knode's whole business is that you no longer have to. What you do with the answer, of course, is still up to you.