Topline · 20 Sep 2026 · From the week of 14 September
CRM As A Business World Model: The Future For GTM Teams? | Keith Peiris, CEO @ Lightfield
These are notes on the conversation, checked against its transcript. The episode itself has the full discussion.
In brief
Keith Peiris, co-founder and CEO of Lightfield, joins Topline hosts Sam Jacobs, AJ Bruno and Asad Zaman. Lightfield is an agentic CRM that launched in November 2025, has more than 5,000 customers and raised a $47M Series A led by Andreessen Horowitz. Keith argues that every system of record is really a forecasting tool, and that forecasting in today's fast-moving market needs a 'business world model' that captures everything rather than just fields. He also covers how Lightfield gets reliable output from non-deterministic models (rep approval of AI field suggestions, real code for math), the two reasons customers actually rip out Salesforce or HubSpot, and the hybrid seat-plus-consumption pricing it landed on after trying both extremes. The hosts debate whether the SaaS partner ecosystem is dead now that software is easier to build, with Asad arguing it is and the others pushing back.
For founders
- Keith Peiris says marginally better AI output is not enough to make a company replace its system of record; in his experience customers switch because they believe one connected model of the business improves every funnel stage, or because they want high-level scenario planning.
- After trying pure seat pricing and then pure consumption pricing, Lightfield put predictable core work in the seat/platform fee and metered only work with visible ROI (pipeline generation, AI workflow automations, scenario planning); pure consumption made customers afraid to touch anything.
- Keith describes Series A fundraising as bimodal: he doesn't see $10M–$15M Series A rounds getting done, and investors either pass or back a $100B-outcome thesis with around $50M.
- Lightfield's investor thesis was that agentic work and consolidation make deal sizes at 200–400 person companies look like Salesforce deals at 5,000-person companies, so a big outcome doesn't depend on ripping enterprises off Salesforce.
- Keith expects commoditized GTM layers such as sequencing and basic forecasting to be absorbed into the system of record, while he partners with products like Gong and Granola that take serious focus to match.
For revenue leaders
- Keith recommends that growth-stage companies have reps approve AI-suggested field updates instead of leaving everything on automatic, because without field auditability leaders won't believe the forecast.
- Have code, not LLMs, do the arithmetic in forecasts and dashboards; Lightfield's head of finance checked every formula once and converted prompts to code, which got the dashboards about 95% of the way there, and LLMs are used to flag deals on the cusp because of missing features or competition.
- Keith says revenue leaders should write down their own standards for deal reviews and have the AI follow them, rather than leaving the AI to decide what matters.
- Asad Zaman's view is that CRM data quality depends on a value exchange: reps fill in the system diligently only when it gives them something back, and reps saying it has zero ROI is his signal to switch.
- Keith says SDRs on a unified CRM pull real-time insights and case studies from customer success and see renewals ahead of time for expansion, and he calls that interconnectedness one of the two real reasons to switch CRMs.
What was said 36, most useful first
Lightfield uses its low-cost PLG plan as a lead-generation tool and upgrades customers from a $2K/year plan to a $24K/year plan once they hit product-market fit. Listen
Keith said pre-PMF companies need a simple system that collects everything, and they need 'something real' once they find PMF. Lightfield moves quite a few customers from the $2K plan to the $24K plan at that point. He describes the current ideal customer as having about 10 reps and growing toward 30, rather than companies still searching for product-market fit.
“we actually move quite a few people off of our like... 2k per year plan to the 24k one as soon as they hit PMF.”
LLMs shouldn't do math, and converting forecast formulas from prompts to real code got Lightfield's dashboards about 95% of the way there. Listen
Asad Zaman challenged whether humans checking every AI-generated forecast formula saves any time. Keith said the work was a one-time implementation: Lightfield's head of finance reviewed every formula and some prompts were converted to code. After that the team makes only fine adjustments week over week. Keith acknowledged 'some might disagree' but said LLMs are not the right tool for math.
“We converted some things from prompts to real code. I don't want LLMs doing math. I want real code doing math.”
Genuine AI errors in Lightfield happen less than once a month for his own company; the more common failure is that the GTM assumption he gave the system was wrong. Listen
AJ Bruno asked how often Keith's daily use of his own product turns up errors. Keith said cases where he asks Lightfield for X and gets Y happen less than once a month. More often, he told the system to track X and then learned from the field that the company should be tracking something else.
“what I thought I knew about our go -to -market process was proven wrong by something we learned recently.”
Marginally better AI-generated output is not a good enough reason for a company to replace its system of record. Listen
Having done several rip-and-replace projects off Salesforce and HubSpot, Keith said that better agent work looks like a reason to switch from the outside but isn't one. Given ten emails, some will be marginally better and some marginally worse, and he says that does not justify changing a system of record.
“It's like not a good enough reason to change a system of record”
Keith Peiris predicts the sequencing and forecasting layer (the Clari/Outreach category) will probably be absorbed into the system of record because it has become commoditized. Listen
Keith said a good sequencer can now be built in a couple of weeks, and so can good forecasting tools built with frontier models on top of great data. He still can't imagine Lightfield doing everything in revenue, because he sees that as infinite and says some companies will endure.
“So I think that that layer, the sort of Clari Outreach, that layer I think will probably be sucked into the system of record.”
Keith Peiris describes Series A funding as bimodal: he doesn't see $10M–$15M rounds getting done, and investors either pass or back a $100B-outcome thesis with around $50M. Listen
Keith said this holds for any Series A right now. Funded companies all have a system-of-record thesis and extreme growth, which he described as 0.1% or 99.9th percentile. Investors accept inefficiency in a land-grab moment and bet on whether the team has the resolve to win, so in his view fewer companies get much more funding. Lightfield raised $47M led by Andreessen Horowitz.
“I don't see $10 million series A as getting done anymore, or even $15 million series A. I think that the default is you don't get funded, or... You know, someone believes that there's like a $100 billion outcome here.”
Lightfield tried pure seat pricing and pure consumption pricing before landing on a hybrid, because pure consumption made customers stop using the product. Listen
Seats plus a platform fee was easiest to get through procurement, but usage at the edges ranged from 10% to 100,000%. Pure consumption backfired, with customers feeling there was a meter on every button. The seat and platform fee now covers the predictable core: AI CRM, call recorder, transcription, filling out fields, tasks and the data model. Consumption covers work with visible ROI, such as pipeline generation (list building, LinkedIn messages) and AI workflow automations like complex lead scoring.
“It's like going to a store where everything's really expensive, that our customers didn't touch anything.”
Lightfield's POC policy is a deal-size cutoff around $25K: below it, the POC comes after a signed contract; above it, Lightfield offers a POC with good qualification if it has enough implementation capacity. Listen
Keith said he spent a few hundred dollars on tokens running about 100 versions of the question of whether to do POCs before or after contract. The answer he reached: under roughly $25K, signing the contract comes first, which enables the POC. Above that, with good qualification, they 'absolutely should' do one, assuming enough implementation capacity to do it well.
“So below a certain deal size, you know for us it's around 25k. It's part of the contract you have to sign to to enable and then beyond that with good qualification we absolutely should.”
Lightfield's average contract size is about $25K–$30K, and its sweet spot is Series B companies with commercial teams of 20 to 100. Listen
Keith gave the $25K–$30K average with some outliers, plus some large deals where the company is still working out implementation and deployment. He described the best-fit customer as a Series B company, or one finding that HubSpot isn't supporting the work and forecasting it needs. He said inbound leads get a rep response within a couple of hours under the team's SLA.
“Right now it's sort of in the 25 to 30k with some outliers.”
Every system of record, CRM included, is at heart a forecasting tool, and that Salesforce won because reps didn't use Siebel and leaders couldn't trust the pipeline. Listen
Keith frames ERP and CRM history as being about forecasting. In his account, because reps didn't use Siebel, revenue leaders couldn't trust the pipeline forecast, and Salesforce broke through by gathering data that made forecasts more reliable. He uses this to explain why Lightfield is built around forecasting and prediction.
“all of these systems of record, whether it was ERP or first CRM, they're actually all about forecasting. And I think the reason that Salesforce was able to break through was that sales reps didn't use Siebel.”
Forecasting has become too complex for pipeline history alone and now needs a broader model of the market and the business. Listen
Keith points to new products shipping every couple of weeks, falling model costs, and segments being written and rewritten. He argues that forecasting in this environment needs signals about the world and about what companies are doing and building, which is something larger than a CRM. That thinking is why Lightfield frames its product as a 'business world model' that starts from the CRM.
“I think sort of forecast and prediction in today's world is far more complicated than just looking at your pipeline for the past month, the past quarter.”
An AI-era CRM should update itself through APIs and capture everything, not only structured fields, because you can't know what questions you'll ask later. Listen
Keith names two data-structure differences from HubSpot and Salesforce. First, the system should update itself automatically by default through APIs, which he says conventional CRMs weren't built for. Second, it should capture all interactions rather than only what fits into fields.
“I think you want to capture everything versus what's just in the fields because Like, who knows what question you're going to ask next month, you know?”
Lightfield customers that sell to enterprise use partial capture rather than recording everything. Listen
Asked whether capture-everything CRMs only suit inside sales, Keith said customers selling to enterprise record calls and product behavior and may use Granola in on-site meetings. No customers record dinners, but he says partial capture still gets them meaningfully closer than a traditional CRM does.
“they record calls, they record in the behavior, or sorry, they record product behavior, they maybe use Granola, you know, in their on-site meetings.”
Lightfield started with a PLG motion for seed and Series A startups, then used those customers as references to move up to larger SMB and mid-market. Listen
Keith said Lightfield was built for two years and at launch was only ready for seed or Series A companies. So it started product-led, served those customers closely, and some of them grew from no AEs to 100 AEs. Lightfield then used them as reference points to move into larger SMB and mid-market, positioning itself as 'the best CRM when you don't have one' and moving people off traditional SMB CRMs as they scale.
“we built the system for two years and when we launched it, it was really only ready for, you know, a seed series A company. So we started with a PLG motion”
Keith Peiris treats AI reliability as a 'harness problem', building on the assumption that models will always be non-deterministic. Listen
Asked how a probabilistic technology can support forecasting, Keith said Lightfield assumes models will stay non-deterministic. For most growth-stage customers, Lightfield does not tell them to leave everything on automatic: important fields are set so reps approve Lightfield's suggestions, which keeps them auditable. For processes like pricing, discounting, qualification and account research, he says the models need a lot of context to be reliable.
“If you don't have auditability of your fields, you'll never believe your forecast.”
Keith Peiris sees the LLM's role in forecasting as surfacing insight, such as which deals were on the cusp because of missing features or competition, not doing the arithmetic. Listen
Keith separates the deterministic numbers, which are handled in code, from the qualitative insight LLMs add. His example is looking past last quarter's closed count to ask how many deals nearly closed or slipped because of missing features or competitors.
“how many of them were sort of on the cusp of actually being able to close because of missing features, your competition and so forth.”
AI can improve deal reviews only if the revenue leader first sets explicit standards for how they should run. Listen
Asked whether AI could improve deal reviews, Keith said he thinks so, with a caveat. Great revenue leaders have strong opinions on what to pay attention to and what to push reps on, and he says that can't be left to chance or handed to a model to work out alone. The leader defines the standards, and the model follows them through skills, knowledge and automations.
“I think you have to sort of set the standards. for how you want Deal Review to run, and then you can get the model to follow, you know, your skills, knowledge, automations on the way you want to run your team.”
Asad Zaman's trigger for switching CRM/ATS is reps saying the tool gives them zero ROI, along with the system looking old and not working well with AI. Listen
Asad, whose firm STA was switching systems again, gave three reasons. The main one is reps saying out loud that the tool they spend time in gives them no value, which tells him the data in it will be poor. The others were that it looks and feels old, like driving a 25-year-old car, and that it doesn't play well with AI.
“it's when your reps start just vocalizing how the thing that they spend a lot of time in has zero ROI for them.”
CRM data quality depends on a value exchange: reps put good information in only when the system gives them something useful back. Listen
Asad said he doesn't want reps filling out forms, but the information gathered is only high quality when people get value in return. His recruiting example: a recruiter should be able to drop a job profile into the ATS and get back everyone the firm spoke to in the last three months who matches it. He says this matters most when not everything can be recorded automatically.
“So once your people feel like they're getting value, then they're willing to eat shit and put the information in”
One of two real reasons customers switch to Lightfield is a belief that modeling the whole business in one place improves every stage of the revenue funnel. Listen
Keith's example: if the CRM knows your happiest customers, that helps you find new ones. SDRs using Lightfield pull real-time insights from customer success and real-time case studies from the happiest customers, and they see renewals ahead of time for expansion. Customers have to believe that interconnectedness is worth something.
“A lot of SDRs that use Lightfield, they're pulling real -time insights from customer success. They're pulling real -time case studies from their happiest customers”
The second real reason customers switch CRMs is higher-level scenario planning, where data modeling and completeness matter most. Listen
Keith lists the kinds of questions involved: whether to build a product feature next quarter, whether to invest in a segment, whether to change the sales process. He says Lightfield found that for these questions the modeling and data completion really matter, which is what motivates customers to consolidate onto one system.
“we found that for that, those sorts of questions, the modeling really matters. The data completion really matters.”
One host argued that AI tools telling reps exactly what to do or say next misunderstand sales, and that the companies building them will fail. Listen
The host said this approach comes from technical people who have convinced themselves sales is deterministic. In complex sales, the same move leads to different results in different scenarios. They suggested the approach might work for customer support but not sales, and predicted 'those companies go down in flames.'
“One thing that a lot of people are doing that I think is pretty dumb is telling the reps what to do next.”
High-growth companies want best-of-breed, but 'best' may come from deep, shared understanding of customers rather than having every point-tool feature. Listen
Lightfield's ICP is high-growth tech companies, and Keith says they all want best-of-breed. He asks whether best-of-breed means every Outreach sequencer feature, or a system that understands the customer and sales process well enough to do the best agent work. His analogy is Microsoft Office: Teams and PowerPoint may not be the best individually, but shared data and directory access make the suite work.
“I think the bet that a lot of people are making now is actually if you have If you have the best modeling of your customers and your prospects, that makes those other features better than the points.”
Many GTM features are now just automations and skills running on the system of record, and some customers have built CPQ inside Lightfield. Listen
Keith admitted Lightfield doesn't yet have the best commission planning tool. His broader point is that customers are building CPQ in Lightfield themselves, using automations, conditional fields and forms, rather than buying a separate product.
“a lot of these features are now just automations and skills that run on your system of record in the new world.”
The SaaS partner ecosystem strategy is dying because easier building leaves less reason to partner. Listen
Asad frames it as build, buy or partner: companies partnered when they couldn't afford to buy and couldn't build. As building gets easier every month, companies build more themselves, so there is less room for partnerships. He says that makes it risky to found a company designed to be an ecosystem partner, because the platform may build your product in a weekend.
“if building has become so much easier you're going to build more of the things so it leaves less room for partnerships”
Asad Zaman predicts that within a year or two, companies will distill standalone products with compute, turning many standalone companies into features. Listen
Asad's example is that a company could spend a million dollars of compute to distill a complete replica of a product like Granola. He concludes that much more product will be needed to win a category. The other hosts called the argument 'smart sounding' but wrong.
“a lot of things that are standalone companies will put on features and the amount of product you're gonna need to have. on a company to be able to win a category is going to just be that much more.”
One host countered that 10x-better products still win even when copying is easy, pointing to Granola's growth despite being simple to replicate. Listen
The host called the distillation argument smart-sounding but divorced from how buyers value something ten times better. They said Granola could already be copied cheaply, yet it grows fast because of product team, distribution, virality or customer empathy. They added that its bot-free recording and pre-call web research come from how deeply it understands the user's workflow.
“So at the end of the day, building a company and building products is still really fucking hard.”
Keith Peiris partners with Gong and Granola instead of trying to match them, and keeps Lightfield's focus on being unbeatable in a bake-off as a system of record. Listen
Keith said Lightfield's call recorder works well for a large SMB but is not as good as Gong's analytics and coaching, so Lightfield runs an open platform, partners with Gong and has a Granola integration. He said building Gong or Granola to their quality 'takes serious focus.' Lightfield's next six months go to the business world model: ingesting signals, building out sales processes, and role-based access control on fields so agents can operate.
“we're sort of all in. on the system of record, and our goal is to be unbeatable in a bake-off on that note. And that doesn't really give us time or energy to go build the best Granola.”
Even a fast-growing company buys software rather than vibe-coding it, because it wants someone else to own support and maintenance. Listen
Keith said Lightfield buys a lot of software even as a high-growth company. The reason is that it wants a vendor to own supporting and advancing the product, so it doesn't fall apart six months later when there are 30 reps. He was dismissive of the trend of people on Twitter saying they'll vibe-code their own CRM.
“part of the reason we want to buy software instead of building it ourselves is because we want someone else to own... supporting it, advancing it, you know, making sure that it doesn't fall apart six months from now when I've got 30 reps.”
Competitors racing to ship the ~100 features needed to run a GTM team often end up with 'Frankenstein' apps that reps can't use. Listen
A host had observed that GTM platforms bolt features onto a left-hand menu without connected workflows. Keith agreed that everyone is racing to the roughly 100 features required to run a go-to-market team. He said making them performant and coherent together is really hard, and many players produce apps that in theory prospect, forecast and hold a CRM data model but aren't usable by a rep.
“create this Frankenstein app. You know, with a hundred features that in theory could prospect, in theory could forecast, in theory has a CRM data model, but is not a thing that a rep can use.”
Keith Peiris's pitch to Andreessen Horowitz was that agentic work and consolidation make Lightfield's deals at 200–400 person companies look like Salesforce deals at 5,000-person companies. Listen
Keith said investors writing a $50M check had to believe Lightfield could reach a $100B outcome without moving enterprises like Costco off Salesforce. Switching systems of record is easier than before but still hard and annoying. The economic thesis was much larger deal sizes in SMB and mid-market, driven by agentic work and consolidation, and he says that only works if customers are happy enough with the agentic work to pay consumption pricing.
“look at our deal size for a company that's 200 people or 300 people or 400 people. This actually looks like a deal size for Salesforce with like, you know, 5,000 people because of all of the work that we're doing.”
Lightfield is deliberately not optimizing margins at this stage, choosing to take market. Listen
Asked whether investors cared about margins, Keith said they definitely asked. He said Lightfield is trying to take the market and his head of finance 'glares at me every day when we close deals,' but he believes there is a path to margins.
“we're not in a margin -optimizing state of the company right now. We're trying to take the market”
CEOs and revenue leaders have an uncapped budget for business intelligence and want the best models for scenario planning, so Lightfield meters it as consumption. Listen
Keith said that running top models on the business world model to plan what to build next quarter, how to change the sales process, or which deals are at risk is something leaders will pay for without a meter. He said he hasn't met a single CEO willing to do this on cheaper models, because they want 'token maxing' to squeeze out a little extra alpha.
“most revenue leaders, CEOs, have an uncapped budget for business intelligence”
One host argued that pricing consumption toward the CEO works because the CEO will overrule the CFO on spending for questions they care about. Listen
Responding to Keith's point about uncapped BI budgets, the host said that if token consumption can be pushed to the CEO, it reaches a person who will override the CFO. They added that embedding it in the product makes this effective.
“if you can push token consumption to the CEO, that is a person that will overrule the CFO and say, this is important to me.”
Companies now ship far more product per $1M of ARR than a decade ago, and today's Series A products feel like Series C products used to. Listen
Asad gave this as his argument for why more functionality ends up inside each platform. He said if you charted product shipped per stage or per $1M ARR over the last ten years, it would be dramatically higher now, and he expects the trend to continue. He did not give numbers.
“Now I go into a series a product and it feels like what a series C product used to feel like”
AJ Bruno credits part of Gong's resurgence to tying everything together in a single 'ask anything' AI experience across accounts. Listen
AJ said Gong's landing page now leads with 'Gong AI, ask anything' and pointed to the newly released Gong Bridge. He said this lets users find all their information across accounts quickly, rather than going to separate forecasting or call tools. He then asked Keith how Lightfield ties its own capabilities together.
“It's not about just the forecasting or the calls or whatever. It brings everything together.”