TLDR: The market is flooded with AI workflows that look impressive but never move a number. In this RevGenius webinar, revenue leaders from Lusha, HubSpot, Clay, and Zapier separated what actually drives pipeline in 2026 from what just feels productive. The short version: the teams pulling ahead are not adding more tools, they are changing what they measure, grounding AI in quality data and context, and putting the right insight in front of customers at the right time.
[Watch the full session now →]
Two years into the AI era, go-to-market teams have no shortage of tools, demos, and LinkedIn-ready workflows. What they lack is a clear read on which of those things actually generate revenue and which just look good in a screenshot. That gap was the whole reason for this session: to cut through the noise and surface the plays that are working inside fast-growing companies right now.
To dig into it, Lusha partnered with RevGenius for a panel titled, “From hype to impact: revenue optimization in 2026.” It was hosted by Jared Robin from RevGenius and moderated by Elad Uzan, VP of Product at Lusha. The speakers were David Greco, Head of Growth at Clay, Lucy Alexander, Director of Agentic Prospecting at HubSpot, and Lindsay Rothlisberger, Head of GTM Innovation at Zapier.
The framing was deliberately practical. No broad predictions, no abstract thought leadership. Just what these teams are prioritizing today, the KPIs they manage, the workflows producing results, and the experiments that actually worked. The discussion moved through four themes: how the metrics are changing, why tools are not the strategy, what the highest impact workflows have in common, and what is built to last versus what is already fading.
The scoreboard is changing
In the past, go-to-market teams ran on the same numbers: conversion rates, win rates, lead-to-opportunity conversion. Those still matter. But AI shifted what is worth measuring, and several teams realized their old metrics were blurring rather than sharpening their focus.
Zapier’s team moved away from obsessing over how many AI projects they could launch and how much time those projects saved. Their number now is recovered selling capacity: did the rep get their week back, and did it show up in pipeline generated? Time saved only counts if it gets reinvested into pipeline and closed deals. They are also watching win rate on coached deals, since AI let them scale the volume of coaching touches reps receive.
HubSpot’s prospecting team landed in a similar place. Their north star is pipeline dollars generated per rep per month. Time-based metrics were not enough on their own, since there is no guarantee reps reinvest reclaimed time into higher-value activities. They considered measuring dollars closed per rep but rejected it, because too many other factors (pricing, win rates, separate close-rate workstreams) muddy that number. They wanted something more proximal to the work they actually control.
The throughline: your team’s actions reflect what you measure them on. As AI changes the work, the scoreboard has to change with it.
Tools are not the strategy
David from Clay made the bluntest point of the session: “Don’t defer creative decisions to AI. Every campaign that worked started with a human-designed message. AI scaled the distribution. Teams handing AI the keys are getting beaten.” His team designs the campaign end to end, then uses automation to scale it. The reason so much AI output is generic, in his view, is that people defer the thinking, the judgment about what resonates with an ICP, to the machine instead of doing it themselves.
The reframe the panel kept returning to: your job is to be the architect of the process, not the person typing every line. You decide what good looks like. The machine does the heavy lifting.
The highest impact workflows share a pattern
HubSpot’s biggest prospecting lever has been better prioritization. Lead scoring is not new, but AI leveled it up. Their team splits rep books into four quadrants on fit and intent, then runs what one colleague called the “rep sniff test”: looking at an account the way a human would on the internet. Does the site have lead-gen forms? Are they running digital marketing? These fuzzier questions were not accessible through static data before. Clear SLAs sit on top of each bucket, and the payoff is concentrated. As Lucy put it: “Data quality matters more than prompt sophistication. The top 30% of accounts drive 95% of revenue. We found them with signal weighting, not funnel volume.” Reps who lean into the prioritized order generate about 10 percent more pipeline dollars per month than those who do not, a lift the data science team confirms with causal analysis that controls for tenure, prior attainment, and manager.
Zapier looks for two signals when hunting for a high-value workflow. First, is there unstructured data to unlock, like meeting recordings, emails, and free-form text that can become a new data point? Second, look for where humans are acting as middleware, collecting things from different places and applying their own personal expertise to tie the pieces together. Both are strong candidates for AI. Their highest impact investment to date applies exactly this: AI facilitates the sales-to-customer-success handoff, pulling all the context leading up to the closed deal and generating the first kickoff materials and account plan. That cut customer time to value.
What the failed and surprising bets taught them
This industry runs on trial and error. A pipeline that works today can break tomorrow, not because something technically failed but because the trend moved on. Some bets surprised the panel.
Lusha’s team assumed reps wanted ready-to-go lists generated by the machine. Reps rejected it. They did not want a black box; they wanted to understand the logic and the reason behind each recommendation. If you hand a rep a contact with no context for why to reach out or what to say, they end up empty-handed. The lesson: reps want AI to do the heavy lifting, but they want to stay in control of what the model optimizes for and which signals it weighs.
Clay learned that a failed experiment is often just a timing problem. A homepage build that flopped nine months ago is now converting, because the underlying technology improved. The real cost is the opportunity cost of your time, so spend it on what your company is genuinely strong at.
Zapier’s lesson was about taste. Do not rely on AI to create your context, or everyone starts sounding the same. Human judgment matters more than ever. Throwing everything at the wall early was useful for learning, but strong ROI came from investing in an opinionated, specific, and evolving context layer.
What sticks, and what fades
Asked what survives into 2026 and what is already fading, the panel converged:
- Data quality is non-negotiable. Messy data ruins everything downstream. Freshness and accessibility matter more than any single tool.
- The context layer stays. Multiple speakers are investing here as the foundation everything else reads from.
- Strong individual contributors rise. Expect flatter management structures and more pay for ICs who can reason through workflows.
- Standalone vibe-coded apps fade. Building random apps people have to visit separately loses to AI that meets humans where they already work, triggered by real events inside the normal sales process. As Lindsay described what actually stuck: “Standalone tools that needed a new login got abandoned. The workflows that stuck were already inside reps’ tools: signal fires, draft lands in the outbox, MEDDPIC writes back to CRM. No new tab.”
- Infrastructure endures. Data layers, orchestration engines, and core systems like the CRM are hard to build. If you are rebuilding them yourself, you are defocusing your business.
Elad closed on the thread running through all of it: “The teams moving fastest didn’t add AI first. They removed the manual step between data and action. AI came after the process was clean. Get it backwards and you’ve automated a broken workflow.” It takes real investment to get that foundation right, and the teams that make it are the ones seeing it pay off.
FAQs
These are the questions viewers asked during the live session, with answers drawn from the discussion.
Where can I learn more about the selling-capacity metric and how it’s measured?
The idea is to measure whether time saved by AI actually converts into revenue, not just hours back. The practical version is recovered selling capacity: did the rep get their week back, and did it show up in pipeline generated and closed deals? A related approach is pipeline dollars generated per rep per month, chosen specifically because it sits closer to the work reps control than dollars closed, which gets distorted by pricing and win rates. To validate the lift is real, one team runs causal analysis that controls for rep tenure, prior attainment, and manager.
What is the delivery format of the sales-to-CS handoff? Is it a deck?
It is more than shared notes. The workflow takes all the context about the customer leading up to the closed deal and generates the actual first-kickoff materials and the account plan for the person taking over the account. The point is to hand over a strong starting point, not raw information the receiving rep still has to interpret.
What should the handoff pattern look like?
Pull together the unstructured context (enrichment data, conversations, emails) that a human used to stitch together manually, then generate the next step rather than just a summary. The principle that came up repeatedly: do not ship intel alone. Ship the intel plus the action it enables, so the receiving person can act immediately instead of making sense of everything themselves.
How does the HubSpot setup actually work? Is it direct rep access via MCP or something more involved?
Reps do not interact with the context layer directly. A single engineering team owns and maintains a context layer that stores insights and reads from the CRM. Reps use tools and UI that read from that context layer and surface it in their workspace. So the architecture is: context layer behind the scenes, rep-facing tools in front of it, rather than reps querying the layer themselves.
What has been hardest about building a context layer that stays current and is legible to both AI and humans?
The honest answer from the panel is that there are few best practices yet, and it is a continuous process, not a one-time build. Things change constantly, so the layer has to be maintained or the layers built on top of it break one by one. One team structures theirs in three tiers that are maintained differently: a foundational layer (ICP, use cases, company strategy) that rarely changes, a middle layer (playbooks, rep process) that changes more often, and a data layer of what is working that changes constantly.
Can you give an example of useful unstructured data beyond call transcripts?
Yes. Support tickets, free-form CRM fields, emails, and prompts like “what are you hoping to get out of this conversation” were all cited. “How did you hear about us” answers can be categorized at scale. Closed-lost reasons can be parsed into clean categories like specific product gaps. On the prospecting side, one speaker asks AI to tell the story of how a deal was created in five sentences, which produces a far richer picture than quantitative fields alone.
How do you operationalize AI so it is context-relevant and time-relevant inside the normal sales process rather than a standalone tool?
Build around triggers and actions. A signal on a lead or account triggers an agent to research the account, decide who to reach out to, draft the outreach, and tee it up in the tool the rep already uses so they apply judgment. Over time more of this becomes autonomous, but the near-term win is integration into existing workflows, not a separate co-pilot that still forces reps to copy, paste, and stitch things together.
A context layer sounds simple until you scope it. How do you structure one without boiling the ocean or letting it rot?
Start narrow. Scope it around a few specific jobs to be done (for example, post-call follow-up, note-taking, and CRM write-back) so you can validate the model before scaling it. Break the context into layers with a metadata schema so agents reference the right pieces the right way. To fight rot, treat maintenance as ongoing and experiment with decision logging that lets agents help keep the context updated. One team noted that the prior two years were spent proving which use cases context actually helped, which is what later told them what to prioritize getting into the layer.
Have you seen email and support ticket data directly improve deal scoring or forecasting?
This is exactly the unstructured-data play the panel championed. The pattern is to convert non-deterministic text (support tickets, emails, free-form fields, closed-lost notes) into clean categories and signals that feed prioritization and scoring. Categorized closed-lost reasons feed product marketing and re-engagement, and richer deal stories improve the picture beyond quantitative data. The strong recommendation was to consolidate this into one trusted system, typically the CRM, so the answers stay reliable.
Given how much the role has changed, how is AI actually enabling reps, since presenting intel does not automatically enable them?
Correct, intel alone does not enable a rep. The fix is to ship the end-to-end workflow, not just the insight: here are the signals on this account, here is how they make money, here is what to say on the phone, and here is the task already created for you to call. It also has to be wired into performance management and coaching, so when something is off, the manager is alerted and it shows up in the one-on-one and coaching notes.
Did anyone explore linking sales incentive compensation to AI workflow adoption?
This was raised but not something the panel reported doing through comp directly. The closest analog discussed was tying adoption into performance management and coaching rather than pay: surfacing when reps do or do not work out of the prioritized, AI-supported workflow, and folding that into manager coaching and one-on-ones. The broader signal was that organizations are increasingly paying strong individual contributors more, which rewards the people who operate this way.
Keep reading:
