# Keith Peiris

Keith Peiris builds Lightfield, an AI-native CRM that organizes customer and company data so AI agents can use it for tasks such as prospecting, sales follow-up, and deal preparation.

- Profile: https://b2binfluencer.marketing/creators/keith-peiris/
- LinkedIn: https://www.linkedin.com/in/keithpeiris
- LinkedIn followers: 15,819
- Topics: artificial-intelligence, sales, ai-agents, go-to-market, b2b-saas
- Based on 12 dated posts, latest 2026-10-06

## Summary

Keith Peiris is useful for technical content on structuring CRM data for AI agents and context-led prospecting.

- Best for: A B2B data platform or CRM migration vendor wanting practitioner-led content on fixing CRM data.
- Not for: A broad business sponsored post package needing rate card and audience data.
- Check first: Confirm whether they offer sponsorship formats, rates, deliverables, and audience data.

### Takeaways

- The bottleneck is usually the data, not the model.
- Agents need a programming door into business software, not a dashboard.
- Outbound should start from your own customers, not a bought list.

## Main ideas

- **The bottleneck is usually the data, not the model**: Peiris argues that when a company connects its CRM, call transcripts, and product data to a language model and gets weak answers back, the usual suspects are the wrong ones: it is rarely the connector layer or the quality of the model. Access is not understanding. A model can reach the systems but cannot interpret relational schemas, stitch together fragmented transcripts, or read event streams that have no ordering. Their proposal is to rebuild the data underneath the model before expecting anything from the model itself.
- **Agents need a programming door into business software, not a dashboard**: Most CRM platforms, Peiris notes, were designed for humans clicking through dashboards. That is now a mismatch because the thing increasingly needing CRM data is not a person. Agents cannot log into a graphical interface. They need to query accounts, pull contact history, and check meeting notes programmatically, quickly, and without a pile of glue code. Their stated design goal is a CRM as useful to a company's AI tools as it is to its people.
- **The CRM should be a model of the business, not a database of records**: The category argument is that record-keeping software is being replaced by a model of how a business actually works — one that updates itself from every interaction and can be trusted by both people and agents. In this account, Salesforce was built more than twenty years ago for humans to update records and store data in fields; layering agent processes on top produces low-quality output because the underlying data is incomplete and lacks the structure agents need.
- **Outbound should start from your own customers, not a bought list**: Peiris argues that most outbound tools start blind: they reach into external databases, scrape for signals, and generate personalized copy from prompt templates. They say this is why many startups do outbound badly and their cold email does not land. Instead, outbound should start with what a company already knows about its customers, since the CRM holds that knowledge.
- **Reach early-stage founders through platforms they already pay for**: Where the other posts argue about capability, this one is about distribution. Peiris's stated goal is to make it easier and cheaper for early-stage founders to start building on an AI go-to-market platform, through a bundle inside software those founders already use.

## Work to read

- [Why most CRM-plus-LLM projects disappoint—and what to fix first](https://www.linkedin.com/posts/keithpeiris_most-people-connect-their-crm-call-transcripts-activity-7442998842276532224-PMey)
- [The case for giving agents a command line into the CRM](https://www.linkedin.com/posts/keithpeiris_most-crm-platforms-were-designed-for-humans-activity-7452768699528200192-b0fG)
- [Turning the CRM data model into an outbound prospecting product](https://www.linkedin.com/posts/keithpeiris_im-excited-to-share-that-agentic-pipeline-activity-7459656223886401537-y_7_)
- [Announcing a $47M Series A—and the case against the Salesforce-era database](https://www.linkedin.com/posts/keithpeiris_today-were-announcing-lightfields-47m-activity-7503482705519583232-P3yo)
- [A distribution move: Lightfield bundled into Ramp rewards](https://www.linkedin.com/posts/keithpeiris_excited-to-announce-that-lightfield-is-the-activity-7445695257624637440-d3bZ)

## Brand relationships

- **Lightfield** (owned_business)
- **Andreessen Horowitz** (investment)
- **Maverick Ventures** (investment)
- **Coatue Management** (investment)
- **Greylock Partners** (investment)
- **Lightspeed** (investment)
- **Audacious Ventures** (investment)
- **Alumni Ventures** (investment)
- **Mistral** (investment)
- **Ramp** (disclosed_partnership)

## Partnerships

Keith Peiris's posts contain no published partnership or sponsorship offering. They announce products and company news for Lightfield, the AI-native CRM they build, but don't describe creator sponsorship formats, rates, deliverables, or a booking process. Lightfield announced a $47M Series A led by Andreessen Horowitz in September 2026, with Maverick Ventures, Coatue, Greylock, Lightspeed, Audacious Ventures, Alumni Ventures, and Mistral named as participants. Lightfield is also the only AI-native CRM bundled in Ramp rewards, which Peiris describes as a way to make the product cheaper for early-stage founders. Those are business and distribution arrangements, not sponsorship packages a brand could buy. Buyers interested in working with Peiris should contact them directly; the posts provide no inquiry route.

## Alternatives

- Matt Serna: Choose Matt Serna when the campaign is about getting a GTM team to move, not how the data layer works. They offer the operator-level version of the same decision. One saved post describes a customer rebuilding its go-to-market on the platform: creating accounts and running research automatically when an app is installed, using trust-center visits as an intent signal to trigger outbound workflows, pulling product usage from PostHog into the CRM, running roughly thirty automations, and rebuilding a sign-up flow that had been a fifty-block Zapier tree in less than a day. The other post names the obstacle directly: at a Pavilion GTM talk attended by about 100 GTM leaders on Salesforce or HubSpot, nobody raised a hand to say they loved their CRM. In follow-up conversations, the dominant feeling was fear of recommending a new tool, retraining a team, or rebuilding a stack they had invested in for years. For a campaign about internal adoption, migration anxiety, or convincing a revenue leader, that is closer to the buyer's audience than a founder explaining schema design.
- Edwin Trebels: Choose Edwin Trebels when the AI-readiness problem sits in documents rather than CRM rows. Their material tackles institutional knowledge locked in regulatory filings, clinical reports, product manuals and compliance records, including cases where the answer to a compliance question existed in six document formats across three languages and no system could connect them. The proposed fix is a semantic layer rather than another model: knowledge graphs that turn unstructured text into queryable, governed, AI-ready data, deployed on-premises, on AWS or on Azure, kept model-agnostic, with traceable context and auditability for compliance teams. That approach targets regulated buyers in financial services, pharma, manufacturing, insurance, energy and government, where getting AI wrong is a board-level risk. The same author also brings buyer-side adoption observations worth using in a campaign: in the hundred largest US companies, the share with the term in their sustainability report title fell from roughly 40% in 2023 to about 25% in 2024 and about 6% on a partial count last year; the demand that replaced it was for lower waste and protected margins. Pilots also fail when the operators who run the plant switch the optimiser off, and founders who never agree in writing who owns the work end up unable to commercialise it.
- Ivelina Dineva: Choose Ivelina Dineva only when the campaign is about reaching founders and investors, or when the thing worth copying is a proprietary structured data asset rather than an argument about architecture. The two saved posts demonstrate that asset clearly: a tracking database covering 55 new VC fund announcements in September totalling $16.8 billion, broken out by fund size, sector, stage and geography, with a median disclosed fund size of $70.4 million and 33 of 54 disclosed funds under $100 million; and a weekly roundup of $2.86 billion across ten US rounds, each entry naming the stage, valuation, sector and investors. Both posts direct readers to the database and a newsletter. That is the same motion a CRM-data vendor could borrow: own the dataset the buyer already searches for and distribute it repeatedly.

---
AI-assisted summary of public posts. Audience demographics are not available. Engagement appears only where we checked recent posts. Method: https://b2binfluencer.marketing/methodology/