The Profitable AI Organization

Every company can build AI prototypes. 

Very few can actually operate AI solutions in production. The reason why 90% of AI projects never scale isn’t because of the technology, but because the organization isn't set up to sustain it.

Who owns this long-term? Which budget pays for it? How does it fit into existing operations? Who maintains it when the original team moves on?

For this edition, I teamed up with Patrick Tammer, who leads Strategy & Operations at Google AI and previously managed a $125M+ AI investment portfolio at Scale AI. He also worked as a management consultant with the Boston Consulting Group, advising 50+ executives on large-scale digital & AI transformations throughout his career. 

Together, we'll show you the 3 proven organizational models for scaling AI, which one to start with, and the exact measurement framework that separates profitable AI solutions from expensive science projects.

Let's dive in!

Organizational AI Readiness

Successfully scaling AI requires three things working together:

1. The right pilots - Use cases with clear business impact and technical feasibility

2. Talent - People who can build and operate AI systems at scale

3. Organizational readiness - Structure and governance that supports shipping AND operating products

A lot of words have been written about picking the right pilots and the talent war in AI. Today's focus is on the organizational piece. Because even the best use case will die if your org isn't ready to sustain it beyond the prototype phase.

The 3 Operating Models for AI at Scale

There are three proven ways to organize for digital and AI transformation. Each works at different scales and requires different levels of organizational change.

The first model, is the Digital/AI Factory approach:

Then, there’s the Product & Platform model:

And finally, an enterprise-wide agile structure:

For most companies Model 1 (Digital/AI Factory) is the right starting point. Here's why:

Why Start With the Digital/AI Factory Model

Fast time-to-value with "lighthouse" wins: A factory model lets small, cross-functional pods ship real products quickly and scale what works. Instead of attempting a risky big-bang transformation, you get tangible wins that prove the model.

Contained scope, lower change risk: You're piloting new ways of working in a focused unit without rewiring the entire enterprise. This reduces the failure modes common in broad transformations—and makes it easier to secure executive sponsorship.

Talent concentration and repeatable tooling: Centralizing your scarce digital and AI talent enables shared standards, reusable code, and knowledge transfer. You build a center of excellence that can scale practices across the organization.

One important note: it's often best to shield the factory from your traditional IT organization initially, but allow for talent transfers. Your ambitious engineers should see the factory as a career accelerator, not a dead end.

Making the Factory Model Actually Work

Having a factory sounds good in theory. But the execution details matter – a lot. Here's what separates functional factories from expensive innovation theaters:

Hub + Line Structure

The factory operates as a hub that provides expertise and standards, while line managers hold responsibility and budget. Factory squads build products with sponsorship from a specific business unit. That BU owns the product and its long-term outcomes – tied to their P&L and OKRs. This structure is critical. Without clear line ownership, you get "everybody's project, nobody's responsibility" syndrome.

Budgets

Initial hub funding typically comes from a central budget. This covers the cost of standing up the factory, hiring core talent, and running initial pilots. But, and this is crucial, business units should take over funding once a pilot passes the stage gate to scale. At that point, ownership transfers to the BU, and the hub becomes an enabler rather than the owner. This funding transition forces real accountability. If a BU won't fund the scaled version of a pilot they sponsored, that tells you something important about the actual business value.

Money follows results. That's how you avoid the pilot graveyard.

The 3-Layer KPI Framework

Once you've identified pilot use cases, robust impact measurement from Day 1 is non-negotiable.

Most companies only measure technical metrics without connecting to business outcomes, or they only measure strategic metrics without understanding what drives them

The solution is the 3-layer S-O-T monitoring framework that connects technical performance to business results to financial impact:

• Layer 1: Technical KPIs (System Health)

• Layer 2: Operational KPIs (Usage & Model Performance)

• Layer 3: Strategic KPIs (Value & Economics)

All three layers must connect. If your system is reliable (Layer 1) but users aren't adopting it (Layer 2), you have a product problem. If adoption is high (Layer 2) but ROI isn't materializing (Layer 3), you're measuring the wrong business outcomes.

Case Study: AI Demand Forecasting That Actually Worked

Let’s see this framework in action.

A major CPG company created an AI factory in Canada. The strategy was simple: test use cases in Canada, scale to the global network if successful.

They worked with business units to prioritize pilot use cases. Demand forecasting emerged as high-impact (product on shelf is a key business driver) and high-feasible (rich POS data, relatively standardized systems).

Here's how they implemented the 3-layer measurement framework from Day 1:

Layer 1 - Technical (System Health):

• Uptime targets: 99.5% availability

• Latency: Forecasts generated within 30 minutes of data refresh

• Error rate monitoring for data pipeline failures

• GPU utilization tracking for cost optimization

Layer 2 - Operational (Usage & Model Performance):

• User activity: Daily forecast requests by region and category

• Model accuracy: Lift in forecast accuracy measured via RMSE and MAPE

• Feedback rates: Category managers rating forecast reliability

• Adoption tracking: Percentage of SKUs using AI vs. legacy forecasts

Layer 3 - Strategic (Value & Economics):

• Value threshold: Reduction in stockout events (target: 15% decrease)

• ROI calculation: Increased same-store sales from better availability minus system costs

• Cost trends: Infrastructure costs vs. forecast volume capacity

• Secondary benefits: Reduction in SLA penalties and improved inventory turns

The results validated the model across all three layers. System reliability stayed high, category managers actively used the forecasts, and the business case held up. They scaled it to additional markets.

What made this work wasn't just the AI model, but it was the organizational structure and measurement discipline. The factory built it, the business unit owned it, and everyone knew exactly what success looked like at every layer.

Your Next Steps

If your organization isn't ready to sustain AI products beyond the prototype phase, here's what to do:

1. Start with the Digital/AI Factory model
Don't try to transform your entire organization at once. Build a focused unit that can ship products quickly and prove the model works.

2. Set up the 3-layer KPI framework before you build anything
Define technical, operational, and strategic metrics upfront. If you can't connect all three layers, you're not ready to build.

3. Get clear on ownership and funding
Who sponsors this pilot? Which business unit will own the scaled product? When does funding transfer from the hub to the line? Answer these questions before the first sprint.

The companies that scale AI successfully don't have better technology. They have better organizational models.

Pick your model. Measure what matters. Ship solutions that create profit!

See you next Friday!
Patrick & Tobias

PS: Sign up to Patricks newsletter to get key insights from a hand-curated list of the top 15+ AI news, tools, and thought pieces – straight to your inbox.