Lately, I’ve embarked on a project to sit down with the companies driving real change across the AI landscape in digital advertising—specifically those building solutions for publishers and ad tech providers.
I’ll be completely transparent: half the reason I’m doing this is for my own benefit, so I have a sharper take on who is doing genuinely interesting work in the space. But I also know I’m not alone. Plenty of operators across our industry see a logo pop up on the Lumascape and have no idea what the company actually does—or, at best, only know what they were doing five years ago.
For this discussion, I sat down with Vivek Vaidya, Co-Founder and CTO of Kana.
(Full disclosure upfront: I have worked with Kana in an advisory capacity.)
Vivek and I first crossed paths 20 years ago when we worked on the same project at Rapt (later acquired by Microsoft). From there, Vivek co-founded Krux with Tom Chavez (acquired by Salesforce, where Vivek served as CTO of Salesforce Marketing Cloud), helped launch Habu out of their startup studio, super{set} (acquired by LiveRamp), and is now building Kana.
If there is anyone who understands the structural data bottlenecks of media operations, it’s Vivek. Here is what we covered, what Kana is solving, and what it signals for the future of publisher yield and operations.
1. The Core Problem: Moving from “What’s So” to “So What”
Whether you operate on the publisher side or the brand side, the fundamental operational headache of ad tech and MarTech hasn’t changed in two decades: a fragmented plethora of tools that do not talk to each other cleanly.
Publishers juggle ad servers, order management systems (OMS), inventory forecasting engines, and CRMs. Because these platforms were historically sold as standard SaaS with heavy implementation layers, every media company had to configure and bend the software to fit their idiosyncratic business rules.
The result? Operators live in a chronic cycle of:
- Exporting CSV files from three or four disconnected systems.
- Dropping them into Excel or Google Sheets to run VLOOKUPs and manual joins.
- Building reporting dashboards that tell leadership “what’s so.”
As Vivek pointed out, the industry is drowning in dashboards that show current status, but almost none of them connect the dots to tell the operator “so what”—let alone execute the fix. That operational drag is the exact problem Kana was built to eliminate.
2. Why Foundation Models Aren’t Enough (Kana’s Differentiators)
Today, anyone with an internet connection can access foundation models from OpenAI, Anthropic, or Google. Building a few prompts on top of an LLM is not a defensible product. When I asked Vivek how Kana differentiates from the dozen other startups spotting the same opportunities, he pointed to two structural pillars:
A. The Canonical Data Model + Semantic Layer
Domain expertise isn’t just knowing that an ad server delivers ads. It’s understanding that every publisher configures their ad server, rate cards, and OMS differently based on bespoke business rules.
Kana built a canonical data model reflecting 25 years of institutional knowledge across ad tech and MarTech. Layered on top of that is an AI-driven semantic layer that inspects a customer’s existing systems and reverse-engineers their unique setup.
In the pre-agentic world, reconciling those systems required bringing in a systems integrator for an expensive 6-to-24-month ETL consulting engagement. Kana uses purpose-built agents to map those workflows automatically, sparing the customer the tax of manually rebuilding their operations to fit a new platform.
B. “Build With” Instead of “Build vs. Buy”
For 20 years, enterprise tech forced a painful binary:
- Buy SaaS: Fast to stand up, but rigid and never quite fits your custom edge cases.
- Build In-House: Tailored to your exact workflow, but saddled with massive, permanent engineering maintenance.
Kana introduces a third model: “Build With.” Through conversational AI and agentic software, operators can configure custom extensions, specialized reports, and proprietary workflows without needing Kana (or their own internal dev team) to staff a custom software project. You get the agility of custom software with the maintenance-free scalability of SaaS.
3. Real-World Execution: What Kana is Doing with Time Inc.
Theory is fine, but in yield optimization, I care about production wins. Vivek walked through two specific applications currently deployed with Time Inc. across their pre-sale and post-sale ad operations:
- Media Proposal Generator (Pre-Sale): Helps sales teams take unstructured RFP briefs from agencies and brands and rapidly generate viable, rate-card-compliant proposals tailored to inventory constraints and advertiser guidelines.
- Campaign Orchestrator (Post-Sale): Once campaigns go live, this agentic layer tracks multi-system delivery in real time, automatically flagging pacing issues, optimizing delivery, and eliminating costly make-goods and under-delivery.
To put the architectural ambition into perspective: Vivek noted that the entire product footprint of Krux represents the equivalent of just two agents and applications inside the Kana platform today. Kana currently has seven modular applications deployed across both buy-side and sell-side organizations.
4. Don’t “Pave Over the Cart Path”: Governance & Change Management
Tom Chavez has a phrase I’ve always liked: “We’re not going to pave over the cart path.”
In other words, the goal of modern AI cannot just be making archaic, broken 2005 workflows run slightly faster. It should be eliminating redundant middle steps entirely—moving directly from Step A to Step C because intelligent agents handle the connective tissue.
However, moving operators toward that reality requires addressing the elephant in the room: trust and governance.
Media companies cannot afford black-box AI hallucinating inventory commitments or misallocating spend. When I spoke on stage in Austin recently, my core message to operators was to insist on strict data governance and detailed audit logs.
Kana has built governance directly into its core chassis:
- Every single agent run leaves a verifiable audit log.
- The system tracks the exact role permissions, data sources accessed, and tools executed.
- It provides regulatory-compliant trails that prove why an agent made a specific operational recommendation or adjustment.
5. Augmented Intelligence, Not “AI Slop”
When I asked Vivek for a contrarian perspective on the current state of technology, his answer resonated with my own consulting philosophy: full autonomy is the wrong goal.
At Kana, the “A” in AI stands for Augmented Intelligence.
In an era saturated with “AI slop” and automated meat-proxying—where people blindly run prompts through an LLM and pass the raw output along without thought—the competitive advantage still belongs to human judgment. The software should handle the heavy lifting, data reconciliation, and synthesis, but experienced human operators must remain in the loop to make critical decisions.
The Takeaway: How to Evaluate Your AI Partners
As we wrapped up, Vivek offered a clean, three-part framework that every publisher, network, and media executive should use when vetting new AI platforms:
Can the platform Sense, Decide, and Act?
- Sense: Can it digest disparate data across your tech stack and understand what is actually happening?
- Decide: Can it apply business rules and domain logic to tell you what needs to be done?
- Act: Does it have the programmatic hooks to make those decisions effective inside the downstream systems you rely on?
If an AI tool stops at “Sense,” it’s just another dashboard. The real yield unlock happens when a platform can connect all three.