You don't need more AI Pilots. You need a system for deciding which ones to run. Enter the AI Lab

Which kind of AI Lab this is
"AI Lab" means different things depending on where you work.
At frontier AI companies such as OpenAI and Google DeepMind, a lab is a research unit. It answers the question what can AI become? It trains models, advances architectures, and publishes research that everyone else later builds on.
A corporate innovation lab, such as Lufthansa Innovation Hub or Bosch Business Innovations, answers a different question: what new businesses should we build that don't exist yet? These are separate units with long timelines, deliberately kept apart from the parent company's day-to-day work.
Both are legitimate. Neither answers the question most organizations are stuck on right now:
Which AI opportunities are worth pursuing, and how do we decide that before we commit serious resources to finding out the hard way?
That is a decision-making problem. It lives in the middle of the enterprise, between the AI strategy most organizations already have and the engineering teams already trying to build.
The enterprise AI Lab in this article is built for that question. It uses the AI capabilities that already exist and asks a sharper set of questions about them: should we build this, for whom, measured how, constrained by what?

The problem: activity that doesn't add up
Most organizations that have been investing in AI for one to two years end up in a similar place. They've bought tools and run AI foundations training. People can prompt. Individuals, and a few teams, have found real shortcuts that make their work better. A handful of pilots shipped.
And it still doesn't add up to anything the business can measure, repeat, or build on.
There's activity and energy, but no way to connect the dots. The wins don't stack.
In most cases the technology is fine. What's missing is a clear way to decide what to do with it, and what to stop doing: a consistent method to pick the few use cases worth real investment, test them fast, and kill weak ideas before they turn into six-month pilots that could have been resolved in days.
Why more pilots won't fix it
The usual response is to run more pilots. That makes the problem worse, because a pilot starts after the most important decision has already been made.
A pilot assumes you already know what to build. The conversation is about resourcing: engineering time, vendor spend, and organizational attention to prove that a chosen solution works. Teams skip pressure-testing the problem and the design. It feels faster, but pilots hit walls quickly, first with systems and data integration, then with legal and compliance.
The AI Lab runs in the opposite direction. It earns the decision before the spend. A structured session with the right people costs a fraction of a pilot and answers the first question: is this worth building at all?
In the Lab, you test the problem and the solution against the barriers you'd otherwise meet later: data access, technical feasibility, legal and compliance constraints, and how users actually think and work. You design something that can survive the real constraints from the start.
The Lab doesn't replace engineering teams, AI tools, or data infrastructure. It sits upstream of them and decides what's worth building before anyone starts building.

How the AI Lab works
An AI Lab has three parts: the people who make the decisions, the person who runs the process, and a sequence of workshops that turns an AI idea into a Scale, Iterate, or Stop decision.
1. The AI Discovery Pod: the people who decide
Each Lab runs with an AI Discovery Pod: a temporary, cross-functional team of six to eight people assembled around one specific challenge. It typically includes:
- The Decider, the business owner whose outcome is at stake and who has the authority to act on the result.
- Someone who does the work, or a domain expert who knows exactly where it breaks.
- A technical lead who can assess AI and data feasibility.
- Legal or compliance, involved from day one.
- A designer who shapes how people will interact with the solution.
- An AI Champion, where the organization has one: the person already using AI in that area.
The pod forms around the challenge, works through the Lab, and disbands once the decision is made.
.png)
2. The AI Facilitator: the person who runs it
The AI Facilitator runs the workshops, keeps to the playbook, and repeats the cycle on a steady cadence. They make sure the right pod is in the room, the right data is available, and the Decider takes part in every workshop.
3. The workshops: from use case to decision
Step 1: Define the use case with AI Problem Framing. Every Lab starts with AI Problem Framing, a one-day workshop that turns a vague leadership mandate into a short list of specific use cases. The pod moves from a mandate like "use AI in procurement" to a precise definition: which workflow, for which user, solving which problem, with which data and governance limits, measured how. The output is a set of prioritized AI Use Case Cards.

Step 2: Choose the path. Prioritized use cases don't all point the same way, and AI Problem Framing shows which way each one should go.

Some use cases are about how employees work: an internal claims process, a document review workflow, a data extraction task that currently takes a team two days. These are workflow problems, and they go to an AI Workflow Sprint.
Others are about building something new for customers: a new service, a product feature, an AI-assisted interaction that didn't exist before. These are product problems, and they go to an AI Design Sprint.
The pod chooses the sprint only after the use case is defined. Defining the problem and designing the solution are different jobs, and mixing them up is where many AI initiatives go wrong.
Step 3: Shape, build, and test in a sprint. Both sprints follow the same logic.
The AI Workflow Sprint runs over four days. The full pod spends Days 1 and 2 on discovery and solution design: mapping the current workflow, identifying where AI creates real value, and agreeing on a solution concept. On Day 3, a lean Build Trio (an AI engineer, a designer, and a subject-matter expert) builds a working prototype. On Day 4, structured sessions with employees who do the work put it to the test.
The AI Design Sprint follows the same structure: the full pod on discovery and solution design, a small team building the prototype, then testing, this time with customers.
Step 4: Make the decision. The pod and the Decider come back together to review what testing showed. The Decider then makes the call:
- Scale: the evidence supports moving toward a production build.
- Iterate: the concept is promising, but specific issues need to be redesigned and tested again.
- Stop: the evidence doesn't support further investment.
The decision rests on what employees or customers actually did and said during testing.
What makes this work is the sequence. The full pod shapes the problem and the solution together. The Build Trio moves fast because the problem is already defined. Testing produces real evidence because the right users were identified in advance.
A healthy Lab kills more ideas than it advances
Many of the use cases that enter the Lab should not survive it. That is the system working. Every use case stopped in the Lab is a six-month pilot you didn't need.
This makes the kill rate the number to watch. A high kill rate shows the organization is making faster, cheaper, better decisions. (The FAQ below covers what a healthy rate looks like.)
That is what the Lab produces: organizational clarity, and a repeatable way to answer the question every executive keeps asking. What should we build with AI?
Why the Lab compounds
The AI Lab is an operating model: a repeatable loop that runs alongside the business without hijacking it.
You can run several pods at once. Different business units can run their own Labs using the same methods and standards. Each Lab ends with a decision, and each decision informs the next. Over time, you build a portfolio of use cases (some in production, some in build, many stopped early) and a clear view of where AI is creating value.
The internal facilitators running the Labs get better with every session. They learn the organization's data landscape, which stakeholders slow decisions down and which ones unblock them, and where the constraints are before anyone has to ask. That knowledge stays inside the organization after a consulting project ends.
Frontier AI labs improve what models can do. The enterprise AI Lab improves the organization's ability to decide what to do with them.
As the models get better, the Lab's job stays the same. It points at what's newly possible, independent of any vendor or tool, and keeps running.
Where to start
Start with one real challenge: a painful problem with a business owner who cares about the outcome and a cross-functional team that can be assembled around it.
Skip the biggest AI bet in the company and the use case that looks best on a board slide. Pick something concrete, where a clear Scale, Iterate, or Stop decision would be worth having.
That first Lab does two things.
It produces an outcome: a validated use case you can build, or a stopped one you don't waste months on. Both are valuable.
And it shows people what a well-run Lab feels like: the right people in the room, the real constraints on the table before anyone falls in love with a solution, and a decision after about a week of workshops instead of months of pilot work.
That's what changes the leadership conversation. A session that produced a decision is more persuasive than another AI strategy deck.
From there, the Lab scales on its own cadence.
Ready to run your first Lab?
Design Sprint Academy, the team behind this article, helps organizations set up AI Labs. We work with a small number of organizations at a time: we start with a first facilitated Lab, train internal AI Facilitators, and coach them through their first independent sessions until the capability is fully yours.


