The last 10%: production readiness challenges

Listen on Spotify

Dave Birckhead is a marketing technology and growth systems leader with over 20 years of experience at the intersection of engineering, product, and GTM. He served as the first Global Head of Marketing Technology at Spotify for six and a half years, where he built and scaled the MarTech stack across Spotify’s three-sided marketplace of consumers, artists, and advertisers during a period of rapid company growth. After Spotify, Dave joined ezCater, a corporate catering company valued >$1B, as Director of Marketing Technology and Operations, leading a rebuild of their growth stage data and systems foundation. Spurred by a growing sense that the shift to agentic AI systems was happening rapidly, in September 2025 Dave stepped away to take what he calls an AI sabbatical. Since then, he’s been deep diving full-time into building AI-native GTM systems, coding prototypes, and sharing what he learns in public through his Substack, Full-Stack Growth.

In this podcast, we discuss:

* Why Spotify built their own messaging stack instead of buying off the shelf and how they used machine learning to optimize message volume at an individual user level

* How Spotify automated global creative asset production from a three to four month manual process down to days

* What prompted Dave to leave a leadership role and go on his AI sabbatical

* How AI has fundamentally changed the velocity of learning and why jumping into building is easier than ever

* Why the last 10% of building AI systems takes more time and expertise than the first 90%

* How GTM engineering roles will converge across marketing, sales, and CS under unified leadership and why companies need to build dedicated IC career tracks for GTM engineers

Episode highlights:

* At Spotify, Dave’s team tackled a critical user protection problem around messaging. Multiple teams across the business were sending messages to users without centralized visibility, leading to high opt-out rates. The engineering team built machine learning models that optimized message volume at the individual user level rather than applying blanket cutoffs, and unsubscribe rates dropped steadily as a result.

* Global marketing campaigns at Spotify required thousands of ad variants across channels, formats, countries, and languages, and as a result took three to four months of manual work to accomplish. Dave and his team automated down the slow and manual creative production work to just days through a combination of process rationalization, vendor consolidation, and a creative production platform. Now on his AI sabbatical, Dave’s rebuilt a prototype version of this same workflow using Claude Code and Figma’s MCP server, where a single creative asset can be used to quickly generate variants across every channel for any country.

* During his sabbatical, Dave has built over 15 prototype systems and two production-grade AI applications. One of his most compelling builds is a cross-functional AI system where marketing, sales, and CS agents share memory and context. Sales agents could reference marketing interactions to inform their recommendations, and CS agents could see what the sales team committed to during the deal cycle, creating a seamless customer experience across the entire lifecycle.

* One of Dave’s learnings from building during his sabbatical is that building an AI prototype to 90% can happen in a day, but completing the final 10% is far more challenging. For LLM-based systems, the work of setting up evals, monitoring, tracing, and security is significantly more time-consuming than building the initial demo. As a result, there’s often a perception gap when stakeholders see someone build a working prototype and wonder why it can’t ship immediately.

* One of the things that gave Dave confidence in taking his sabbatical was how much easier AI has made it to learn. He uses AI-assisted coding to handle syntax he hasn’t brushed up on in years, and he keeps a ChatGPT window open as a tutor, asking it to explain unfamiliar concepts as he builds. As a result, Dave’s able to learn faster than ever, and never feels like a lack of knowledge blocks him from continuing to build.

* Dave believes roles in marketing tech, rev tech, and CS tech will converge under unified leadership responsible for shared customer data foundations and cross-functional prioritization. He also sees a parallel shift in career pathing, where companies will need to establish IC tracks for GTM engineering similar to what exists in software engineering. Rather than measuring contribution by team size or managerial scope, companies will need to identify and reward individual contributors who are having an outsized impact on business outcomes and build compensation and performance structures around that.

Where to find Dave:

* LinkedIn

* Full-Stack Growth Substack

Transcript details:

(00:00) Intro

(03:17) Joining Spotify as their first Global Head of Marketing Technology and the state of the stack in 2017

(06:33) Spotify’s three-sided marketplace and the hybrid build vs. buy approach to their MarTech architecture

(10:20) Automating global creative production at Spotify from months to days

(11:56) Building Spotify’s messaging stack and using machine learning for individual-level message optimization

(13:51) Why Dave chose ezCater and the challenge of rebuilding data foundations at a growth stage company

(16:18) The decision to take an AI sabbatical and why evenings and weekends were not enough

(24:31) Structuring the sabbatical around building, writing, and community & favorite projects

(33:43) What building teaches you about the future of super ICs and 10x GTM contributors

(35:34) The emergence of GTM engineer roles at companies like Notion and Ramp and what IC career pathing needs to look like

(37:54) Why you should build bespoke AI systems on top of your platforms instead of vibe coding your CRM

(39:26) Non intuitive places that AI can positively impact system builds and the 90% to 100% fidelity challenge

(42:20) Evals and understanding what production-ready means

(43:53) Dave’s summarized thoughts about taking an AI sabbatical, advice for people considering a similar path and why jumping back into building is easier than you think

(46:39) Why GTM roles across marketing, sales, and CS will converge under unified leadership

(50:08) Behind the scenes of Spotify Wrapped and the systems that make it work at scale

(54:15) The Tao of MarTech

(57:49) Favorite underrated tools, growth hack, and wrap up

For inquiries about sponsoring the podcast and to recommend any guests, email noah@thegtmengineer.ai

This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit thegtmengineer.substack.com