Podcast episode

S2E7: Chalice's Ali Manning on AI with Receipts

by Ken Rona ·June 23, 2026 ·Leadership In

ai-in-adtech dsp measurement programmatic walled-gardens

TL;DR

Ali Manning, COO and co-founder of Chalice — an AI-driven media optimization startup founded in 2020 — discusses building custom AI models that sit inside DSPs (demand-side platforms, software advertisers use to buy digital ads) to replace platform-native algorithmic decisioning. The episode is primarily a leadership/culture conversation; ad-tech strategy content is present but thin. Listeners seeking deep product or market analysis will be disappointed; listeners interested in founder leadership at an emerging AI-for-advertising company will find it useful.

What was covered

Chalice's core product: Founded in 2020 by Manning and her husband Adam, Chalice deploys custom AI models directly inside platforms advertisers already use (e.g., The Trade Desk, Meta) to control bidding, targeting, and context decisioning — replacing the platform's own optimization logic with brand-specific models trained on proprietary data sets.
Product-market fit and AI adoption curve: Manning says Chalice now has product-market fit but characterizes the overall industry as still in "early adopter phase" for AI in advertising, with widespread "AI washing" — established vendors relabeling old algorithmic tools as AI — creating market confusion.
Agentic AI (AI systems that take autonomous, multi-step actions) as next frontier: Chalice is actively building agentic capabilities into its products, not just talking about them. Manning's explicit positioning: the industry talks about what AI will do; Chalice shows what it has done.
Attribution reckoning thesis: Manning argues major platforms (implicitly Google and Meta) systematically over-claim credit for sales — citing a scenario where a marketing team reports 32% sales lift while total company sales rose only 5%. She expects a CFO/CEO-level reckoning that drives demand for independent, brand-owned optimization tech.
Client mix and go-to-market: Largest clients often bypassed formal sales pitches, coming in with specific outcome goals — Manning names Hershey's (candy sell-through optimization) and an unnamed financial-services client (incremental customer lifetime value) as illustrative examples. Chalice is now scaling beyond top-tier "bleeding edge" clients to a broader base.
Political vertical: Chalice's first-ever client was the Biden 2020 campaign; second was Jon Ossoff's Senate campaign, both sourced by Manning's sister. Manning recently hired her sister part-time to run political for the 2026 election cycle, framing it as a trust/speed decision in a narrow-window vertical.
Company headcount: ~50+ employees as of recording.

Notable claims & predictions

Manning on platform attribution inflation: "Marketing drove sales up 32% last year… but total sales went up 5%. How can marketing say they went up 32? These platforms are taking more credit than due for the sales they're driving." — direct challenge to walled-garden measurement credibility.
Manning on budget concentration risk: "If I give all of my money to two platforms… are they getting 80% or more of advertiser budgets? A lot of times, yes." — framing over-reliance on Google/Meta as a structural problem Chalice is positioned to solve.
Manning on AI washing: "There's a lot of AI washing in the industry, which is just slapping AI onto a reskinned old product… customers can kind of tell." — suggests brand-side sophistication is growing faster than vendor claims.
Manning on agentic AI: "Everyone's going to be talking about agentic. People are mostly going to be talking about what it will do or what it could do. I'm more interested in what it has done." — a pointed critique of vaporware positioning across ad-tech.
Manning on the market shift: "We think more and more brands are going to wake up to having been kind of played by the big platforms… that is already at a reckoning in some organizations and will continue to be." — Chalice's core growth thesis staked on advertiser disillusionment with walled gardens.

Why this matters for ad-tech operators

Platform attribution is under pressure from within the buy side. Manning's anecdote about the 32%-vs.-5% sales discrepancy reflects a real tension emerging in CFO suites. If brands increasingly demand third-party or brand-owned optimization layers rather than trusting platform-reported ROAS (return on ad spend), it structurally benefits independent measurement, clean-room, and custom-AI vendors — and creates headwinds for walled-garden self-reported metrics.
The "custom AI model inside the DSP" architecture is a meaningful category signal. Chalice's model — deploy bespoke AI decisioning within an existing DSP rather than rebuilding the stack — is a low-friction wedge for sophisticated advertisers. If this pattern scales, DSPs face pressure to open APIs further or risk becoming commodity pipes; agencies face pressure to prove they can deliver equivalent optimization.
AI washing is now a buyer-side concern, not just an industry talking point. Manning's observation that clients "can kind of tell" when AI claims are hollow suggests brand-side procurement and innovation teams are becoming more discerning. Vendors relying on rebadged algorithmic tools as AI differentiators may find the window for that positioning closing.
Direct impact on most operators is low from this episode. No financial data, no partnership announcements, no regulatory content, no M&A signal. This is primarily a founder leadership profile useful for competitive intelligence on Chalice's positioning and thesis, not an operational briefing.

Full analysis

Decision Council: AI With Receipts

Step 1 — Frame

This is a founder-profile podcast, but underneath it sits a real question for ad-tech operators: Is the "custom AI model inside someone else's DSP, plus independent attribution" pitch becoming a durable category — or is it Cannes-season noise that fades when budgets tighten?

Reversibility: Type 2 for most operators. Watching a thesis develop costs nothing; you can react when the signal firms up.
What's actually being decided: Whether the walled-garden attribution backlash and the "agentic execution layer" land-grab are real enough to change roadmap priorities — open your APIs, build independent measurement, or call the bluff.
Forcing function: Cannes 2026 just happened. The three saved AdExchanger/Digiday pieces show agentic AI and "show me the bill" skepticism dominating the same week. That's the live context.

The episode's own summary is blunt: direct operational impact is low. No deals, no numbers, no regulation. So this council weighs the thesis, not the company. (Per the brief, Chalice is the messenger, not the protagonist — the reader could be at any of fifty firms.)

Step 2 — The Council

The Skeptic. The headline claim — "marketing reported 32% lift while company sales rose 5%" — is rhetorically great and analytically empty. Total sales move on pricing, distribution, competition, and macro; a 32% incremental lift on the ad-touched segment and a 5% total company number aren't contradictory at all. The anecdote proves measurement is hard, not that platforms are lying. The "brands woke up and feel played" reckoning has been predicted every year since 2018. Walled gardens still take the budget because they work well enough and nobody got fired for buying Meta. In plain terms: the villain story is satisfying, but the math in the anecdote doesn't actually indict anyone.

The Operator. The "custom model inside The Trade Desk / Meta" wedge is clever precisely because it's low-friction — you don't rip out the stack. But the offboarding meeting notes tell the real story: AI tooling is powerful and expensive. Agentic workflows "burn tokens fast," 20-30k vs 500k tokens is the difference between a margin business and a science project. Anyone shipping agentic execution at scale hits the same wall: token cost, context management, and silent failure when an autonomous bidder does something dumb at 2am. The receipts cut both ways — receipts also mean accountability when it breaks. In plain terms: autonomous ad-buying robots are cheap to demo and brutally expensive to run reliably.

The Customer / End User. Here the thesis has legs. CFOs genuinely don't trust self-graded platform homework — that distrust is real and growing. The most telling line in the summary: big clients "bypassed the sales pitch" and arrived with outcome goals (Hershey's sell-through, financial-services lifetime value). That's a buyer that already knows what it wants and just needs an independent execution layer. And buyers "can kind of tell" when AI is rebadged — procurement is getting sharper. In plain terms: the brands paying the bills increasingly want a referee they hired, not a scoreboard the player controls.

The Long-Term Thinker. Three years out, two structural shifts look durable regardless of any single vendor. First, "independent optimization and measurement that the brand owns" becomes a standard line item — the demand is real even if today's vendors aren't the winners. Second, DSPs face a fork: open the APIs and become valuable infrastructure, or stay closed and risk becoming commodity pipes that smart buyers route around. The QP/DSP roadmap notes — curation, data layer, integrations with Amazon and Google DSPs — show platform builders already wrestling with exactly this openness question. In plain terms: the long-run winner is whoever owns the neutral measurement layer, and the platforms know it.

Step 3 — The Tensions

Is the attribution reckoning real or perennial? The Customer sees genuine CFO distrust; the Skeptic notes it's been the same prophecy for eight years while budgets kept flowing to Google and Meta. Both can't be fully right.
Does "AI inside the DSP" empower buyers or just add a fragile, costly layer? The Operator's token-cost and silent-failure warnings collide with the Customer's appetite for a low-friction independent layer. The economics decide who's right.
Receipts as marketing vs. receipts as liability. "What it has done, not what it will do" is a sharp Cannes differentiator — but the Operator notes shipping autonomous systems means owning the failures too. Vaporware never crashes a campaign.

Step 4 — Synthesis

What this actually hinges on:

Belief 1: CFO distrust of self-reported platform metrics is converting into budget moving to independent layers. Directional evidence is real (the Cannes "show me the bill" mood, buyers arriving with their own outcome goals), but conversion to dollars is unproven.
Belief 2: Agentic execution can run profitably at scale. Unproven, and the Operator's token-economics concern is the binding constraint nobody on the Croisette is discussing.

The council leans like this: the demand-side thesis (independent, brand-owned measurement and optimization) is real and worth roadmap attention. The supply-side claim (agentic execution is here and working) is mostly Cannes positioning until the unit economics are shown. The strongest operator move is unglamorous: API openness is a competitive variable now — DSPs that stay closed invite smart buyers to route around them.

What to verify before reacting: ask any "agentic" vendor for the cost-per-managed-dollar and the failure-mode runbook, not the demo. And watch whether attribution skepticism shows up in actual budget reallocation in Q3/Q4 prints from Meta and Alphabet — not in conference quotes.

Step 5 — The Prediction

The episode's own summary flags direct impact as low, and the load-bearing claim — a wholesale walled-garden attribution reckoning shifting budgets — is exactly the perennial prophecy the Skeptic flagged. I won't dress a hunch as a call.

No high-conviction prediction this week.

The durable signal here — agentic execution layers and independent measurement as a real category — is genuine but slow-moving, and nothing in this leadership-profile episode is grounded enough to date a falsifiable claim. The honest read: watch Q3 platform earnings for any sign attribution distrust is moving dollars, and that's a different episode's prediction.

Listen to the episode →