Spotlight.ai CEO Roi Carmel on VentureFizz: What Enterprise Buyers Still Won't Build Themselves
Episode 445 of the VentureFizz Podcast covers build versus buy in the AI era, the two years of CRO interviews behind the Spotlight.ai Knowledge Graph, and why the company has stayed bootstrapped through 300% growth.
BOSTON. Spotlight.ai co-founder and CEO Roi Carmel appeared on Episode 445 of the VentureFizz Podcast with host Keith Cline, in a conversation that ranged from a childhood split across Israel, communist-era Romania and Los Angeles to a nationally ranked judo career, product leadership at HP Software, Perfecto and Cybereason, and the founding of Spotlight.ai in 2022.
The thread running through the episode is a question most enterprise software companies are arguing about internally right now. If AI makes it feel like any team can build anything, what is still worth buying? Carmel answered it with a two-part test, and returned to it repeatedly as Cline walked him through his career.
A two-part test for build versus buy
Carmel opened by warning against the pattern he sees most often: companies searching their own architecture for a place to put AI, rather than starting from a problem.
"When technology moves this fast, people sometimes forget the fundamentals: Where is the pain? Where is the organizational bottleneck?" he told Cline. Without that, he said, "you end up with a solution looking for a problem."
He then acknowledged the reality that vendors have been slow to say out loud. Buyers now believe AI gives them the ability to build in-house, and Carmel argued there is no point fighting it. Vendors should help customers accelerate what they intend to build themselves, then be precise about the narrow set of things customers will not realistically build. In his experience that set comes down to two factors.
The first is workflow integration. A point solution, where a user leaves their job to go do something and then comes back, is comparatively easy to build internally. Integration that reaches across a seller's calendar, email, Salesforce and the rest of the enterprise stack is not.
The second is proprietary data. "Proprietary data allows AI to perform actions it simply cannot do without it," Carmel said, describing the data structure a vendor accumulates as the durable defense against an in-house rebuild.
An honest conversation with three customers
Asked whether the current wave of forward-deployed engineers amounts to a return to the long consulting-led implementations of the on-premise era, Carmel agreed, and attributed it to a market still working through confusion about what it can build.
He described three enterprise customers who came to Spotlight.ai over a four-month period to say they were evaluating building some of the platform's functionality themselves. Rather than argue, the company sat down with each of them and helped define the capabilities that would determine success either way.
All three came back with a split decision: specific areas they would build internally, and a decision to expand their use of Spotlight.ai for the core engine. Carmel framed the exercise as a discipline both sides should adopt. Be clear on what will make you successful, he argued, not on what is possible, because the two are not the same list.
From the judo mat to product leadership
Carmel was born in Israel, moved to Romania as a child while his parents served as diplomats, was present during the revolution that ended communist rule there, and finished high school in Los Angeles. He competed as a judoka in both Israel and LA, ranked third in California and 20th nationally.
He drew two lessons from the sport that carry into how he runs a company. The first is failure tolerance. In judo a single mistake ends a match instantly, and unlike team sports played over four quarters, the competitor who has the better day wins. The second is closer to how he thinks about deals: judo depends on using an opponent's strength, which requires connecting to what the other side is doing rather than staying inside your own plan.
"Even if you have an offensive plan, it means nothing if you don't connect with where your opponent is moving and adjust in real time," he said.
His technical career started as a developer, and he was managing a development team within a year and three global teams within four. Two things became clear in that stretch. He had hired engineers who were better coders than he was, and he was more drawn to customer problems than to the implementation beneath them.
Killing a product on the first day at HP
At HP Software, which had acquired test automation pioneer Mercury for $4.5 billion, Carmel managed portfolios of $100 million and then $400 million, along with reseller partnerships with SAP. He described the period as where he learned how much the go-to-market engine determines whether a product matters at all.
He also told Cline about his first action at the company. HP was carrying both WinRunner and its successor, QuickTest Professional, because customers still loved the older product. Carmel asked why the company shipped two products doing the same job, and then sunset WinRunner.
Removing the overlap let customers build long-term strategies with confidence, he said, making the case that clarity is usually worth more than the friction it creates.
He went on to Perfecto Mobile as Chief Strategy and Chief Product Officer, where he judged that open-source frameworks such as Selenium were commoditizing test automation and repositioned the company as a cloud lab of real mobile devices. Perfecto scaled to $70 million in revenue as category leader before being acquired by Perforce. Carmel then took the value framework and MEDDPICC qualification process his team had built there to Cybereason, where he was part of a management change that repositioned the company and helped scale it past $100 million in ARR.
Fifty CROs and a knowledge graph
The same problem followed him across Perfecto, Cybereason and advisory roles at companies including Tufin. Organizations had strong value propositions and no reliable way to operationalize them across a sales team.
Carmel spent two years interviewing more than 50 chief revenue officers before building anything. What he found was a consistent split in how sales leaders describe their own work: one side devoted to relationships and trust, the other to analyzing the deal against a playbook and reading signals. The second side degraded as teams scaled, because reps are hired and rewarded for the first.
He partnered with two co-founders, one with a background in natural language processing, to build a data structure that mapped enterprise sales methodology, value selling and industry-specific signals. They did not set out to build a knowledge graph and only later used the term. That graph now holds more than 40 million correlated signals, and when large language models arrived roughly 18 months into the work, they supplied an interface capable of traversing it in real time.
Where MEDDPICC stops
Cline raised MEDDPICC, the qualification methodology that originated at PTC. Carmel called PTC the mecca of enterprise sales leadership and named John McMahon, who sat on the Cybereason board, and Keith Butler among the leaders who built the discipline around it. By his estimate, 80% to 90% of enterprise teams that follow a formal qualification methodology now use MEDDPICC.
He then made an argument that sets up how Spotlight.ai scores deals. MEDDPICC is a risk qualification framework and the primary lens for finding risk, but it is not the only one. It does not account for deal velocity, or for a senior stakeholder who enters a deal late and opposes the purchase. Carmel described a deal as a multi-dimensional problem in which MEDDPICC is one important dimension among several.
What the platform does for reps and for managers
Carmel described the product through the framework the company runs on: listen, understand, decide, act.
For a rep, the platform flags active risks, supplies the discovery questions for the next meeting based on what is already known about the deal, and generates the customer-facing assets and business cases the rep would otherwise have to assemble. For a manager, it addresses what Carmel called the central problem of sales management, which is separating evidence from opinion.
"We have reps standing up and saying this deal is coming in. It looks really good. That's an opinion," he said on the episode. Spotlight.ai ingests calls, emails and Slack messages, maps them against the Knowledge Graph, and color codes evidence separately from opinion, giving managers a bottom-up forecast with traceability down to the sentence spoken on a call.
Asked whether the platform depends on Gong, Carmel said it integrates with existing recorders including Gong and Chorus, and supplies its own where a customer has none.
Bootstrapped, 300% growth, zero turnover
Spotlight.ai raised a small pre-seed round from angel investors including Tom Schodorf, the former CRO of Splunk, then funded its growth from customer revenue. The company grew 300% year over year last year with top-tier net retention, and Carmel said he revisits the decision every quarter.
He argued that bootstrapping produces a more accurate product, because a company without the option of throwing money at a problem has to solve it. The metric he said he is proudest of is that Spotlight.ai has had zero employee turnover since it was founded.
On the broader question, he was direct. Capital is a means, not an achievement. Acquiring customers, growing the company and delivering retention are the achievements.
Reading, archery and a morning walk
Carmel described his own AI use as conversational and personal. He takes a 30 to 40 minute walk most mornings with ChatGPT voice mode running through his AirPods, researching whatever he is curious about that week, from real estate markets to competitive archery equipment, and finds a structured summary waiting for him afterward.
He shoots target archery, a sport he took up roughly 15 to 20 years ago and deliberately never competed in, choosing to keep one pursuit free of his own competitiveness. His book recommendations were Never Split the Difference by Chris Voss, which he connected back to judo and the need to read the other side, and The Hard Thing About Hard Things by Ben Horowitz. He learned to surf for the first time this past summer.
Three Questions to Ask Before You Build It Yourself
Carmel's build versus buy test reduces to a short diagnostic that any executive holding an in-house AI mandate can run against a specific project.
Does it have to live inside the workflow? A tool a user opens, uses and closes is a reasonable internal build. One that has to hold state across calendar, email, CRM and the rest of the stack is a different order of problem, and the integration is the part that fails.
What proprietary data does it need, and do you have it? Carmel's position is that proprietary data lets AI do what it otherwise cannot. A model pointed at your own CRM inherits whatever that CRM contains. If the capability depends on signal you do not hold, building it does not create the signal.
Are you listing what is possible or what will make you successful? The distinction Carmel drew with three customers evaluating a rebuild. The two lists look similar early and diverge sharply, and only one of them predicts the outcome.
What Leaders Should Take Away
Fundamentals decide where AI goes, not the other way around. Find the bottleneck first, or you ship a solution looking for a problem.
Integration and proprietary data are what survive the build-versus-buy conversation. Everything else your customer can and eventually will build.
Clarity is worth more than the friction it costs. Carmel sunset a beloved product on his first day at HP so customers could finally plan.
MEDDPICC is the primary lens for risk, not the whole deal. Velocity and a late-arriving detractor are real risks it was never built to catch.
Request a demo
Spotlight.ai runs MEDDPICC qualification, deal inspection and bottom-up forecasting autonomously, grounded in a Knowledge Graph of more than 40 million correlated enterprise sales signals.
About Spotlight.ai
Spotlight.ai is the autonomous deal execution company. Headquartered in Boston, it replaces manual qualification, inspection and forecasting with an AI agent squad that listens to every buyer interaction, qualifies deals against the customer's own playbook, generates customer-facing assets and produces evidence-based, bottom-up forecasts. The platform has autonomously qualified and executed more than $8B of pipeline and is SOC 2 Type 2 certified.
About the VentureFizz Podcast
The VentureFizz Podcast is hosted by Keith Cline and features interviews with tech founders and investors. Each episode covers the guest's background story, their current company and startup advice.
Listen to the full conversation
FAQs about Spotlight.ai and Episode 445
Who is Roi Carmel?
Roi Carmel is the co-founder and CEO of Spotlight.ai. He previously served as Chief Strategy and Chief Product Officer at Perfecto Mobile, Chief Product Officer at Cybereason, and as a product executive at HP Software following its acquisition of Mercury.
What is the build versus buy framework discussed in the episode?
Carmel argues that customers will build point solutions themselves, so vendors should be clear about the two things customers will not build: non-trivial integration across a user's full workflow, and proprietary data that lets AI perform actions it could not otherwise perform.
What is the Spotlight.ai Knowledge Graph?
It is a proprietary data structure that maps enterprise sales methodology, value selling and industry-specific deal signals. It holds more than 40 million correlated signals, which the platform uses to qualify deals, identify risk and forecast from the bottom up.
Why does Spotlight.ai say MEDDPICC is not enough on its own?
MEDDPICC is a risk qualification framework and the primary lens for finding risk in a deal, but it does not capture deal velocity or a negative senior stakeholder who arrives mid-cycle. Spotlight.ai treats a deal as a multi-dimensional problem in which MEDDPICC is one dimension.
Is Spotlight.ai venture funded?
Spotlight.ai raised a small pre-seed round from angel investors including Tom Schodorf, the former CRO of Splunk, and has since funded growth from customer revenue. The company grew 300% year over year last year.
Does Spotlight.ai replace Gong?
No. Spotlight.ai consumes recordings from existing conversation intelligence tools including Gong and Chorus, and provides its own recorder for customers who do not have one. It adds the qualification, asset generation and forecasting layer on top.
Comments