Homeโ€บAI workshopโ€บAI design sprint
Validation sprint ยท Pre-investment AI concept

AI design sprint.

A short, intense validation session for product teams considering AI but still unsure whether the problem really lends itself to it. From problem framing via a use-case shortlist and model choice to a clickable prototype that we test with real users โ€” including technical feasibility, data reality and AI Act classification. The sprint ends with a well-founded go/no-go, not yet another Miro board.

FormatConcentrated sprint
AudienceProduct & management teams
OutputGo/no-go + prototype
ValidationReal users
VendorsIndependent
LocationAt your office + remote

What is an AI design sprint?

An AI design sprint is a focused validation session in which we work with your product and business team to run an AI idea, in a short space of time, through all the questions that are normally only asked after an investment has been made. We work with a hybrid of the classic Google Ventures Design Sprint format (Jake Knapp) and an AI feasibility check: problem framing, user insight, sketching, prototype and test on one clear track, with technical feasibility, model choice and compliance built into every phase.

A regular Design Sprint asks questions like "do users actually want this?" and "does this flow work?". An AI design sprint does that too, but adds three questions that are decisive for AI concepts. One: is the problem suited to AI at all, or are you secretly looking at an if/else rule that would be made smarter by automation rather than by a model? Two: which approach fits: an LLM, classic machine learning, vector search or a combined pattern? And three: is the data reality in order: does the data you need exist, is it accessible, and is it of a quality a model can genuinely work with?

The sprint is therefore the bridge between a broader AI business training (where your team learns what AI is and where it can land) and a full AI implementation project (where we put a validated concept into production). A training opens up thinking, a project executes; a sprint chooses. If you jump from training to project without a sprint, you commit budget to a use case that later gets stuck on a data assumption that turned out to be wrong or a regulatory question nobody had asked.

If you have not yet clarified the broader question of "which AI direction should we take strategically", an AI consultant role is often a more logical first step. The sprint itself assumes you have one concrete candidate use case on the table: not "something with AI somewhere", but a specific problem where you suspect an AI solution fits. Only then is the sprint efficient.

Hybrid
Knapp design sprint plus AI feasibility check in one format
5 phases
Frame, map, decide, prototype, test, each with an AI layer on top
Go/no-go
A well-founded decision point at the end, not yet another Miro board
Pre-investment
Validation before budget, governance and build

When does an AI design sprint fit?

A sprint makes sense when you have a concrete candidate use case but still have doubts about one or more of the typical pitfalls. A few patterns we often see.

01
Problem fit

You doubt whether the problem is suited to AI

You have a process that is slow, error-prone or disproportionately manual, and your team is thinking of AI. Before investing, you want to know whether a model genuinely adds value, or whether what you actually need is a process redesign, a rule-based system or a better search function. Not everything that is tedious is AI work.

02
Approach choice

You don't know whether it should be an LLM, ML or search

An LLM is tempting because it seems able to do everything. Classic machine learning is often cheaper and more stable where patterns recur. Vector search solves a specific class of problems elegantly. During the sprint we identify the right approach based on your data, latency requirements and cost model.

03
Data reality

The data assumptions are unproven

On paper, your organisation has the data you need. In practice, it is spread across five systems, structured inconsistently, or contains exactly not the signal the model needs. The sprint asks this question on day three instead of on day eighty of an implementation.

04
Compliance & AI Act

You don't know which risk profile you fall under

Now that the AI Act is in force, every use case falls into one of the risk categories (prohibited, high, limited, minimal). The classification directly determines how much work comes with it. In a sprint you get a provisional classification on the table, plus a picture of what the Art. 9โ€“15 requirements would mean for your specific case.

05
Stakeholder alignment

Leadership and team talk past each other

Leadership wants "something with AI"; the team doesn't know what outcome is expected; IT wonders where this lands in the architecture. A sprint forces you into the same room to get the same assumptions straight, with a validated prototype as a shared reference.

06
Pre-investment gate

You don't want to release budget without evidence

An AI implementation takes serious investment. A sprint is your gate: you decide based on evidence (users, prototype, technical check) whether to proceed to an implementation project. Both outcomes are valuable.

Three principles that make the sprint different.

An AI design sprint at Appfront looks like a design sprint from the outside: five phases, structured days, a prototype at the end. Under the hood, three choices differ from a classic Knapp sprint.

Principle 01

Feasibility runs alongside, not afterwards

In an ordinary design sprint you mainly test desirability: do users want this? With us, technical feasibility is part of the process from day one. Which model, which data, which latency, what cost per inference. A prototype that is desirable but technically unfeasible is still a no-go, and you would rather hear that on day five than in month three.

Principle 02

Compliance is not a postscript

The AI Act classification is set provisionally on day one and sharpened on day four as the use case takes concrete shape. A high-risk classification immediately changes the scope: different requirements for documentation, logging, human oversight and robustness. We factor that into the go/no-go decision, not as an afterthought but as a variable.

Principle 03

We genuinely build the prototype, not in slides

The sprint prototype is not a clickable Figma mock-up. We build a working prototype with a real AI API behind it (Anthropic Claude, OpenAI or open-source), a UI mock-up in Streamlit or v0, and a minimal flow in which users actually get answers. That changes what you learn in the user test.

How the sprint is structured.

Five phases at the heart of the Knapp format, with an AI layer in each phase. We work in a concentrated block, not scattered working groups, so the whole team tackles the same pieces at the same pace.

01

Problem framing and long-term goal

We open with the classic Knapp question: what is the problem, and what would success mean over a longer horizon? For AI concepts we add two sharper questions. One: who exactly is the user of the AI system: the end customer, the employee, or an agent in another system? Two: which decision or action changes because of the system, and how would you measure whether that decision becomes better, faster or more consistent?

Sprint questionLong-term goalDecision metric
02

Map: use-case shortlist

We map the end-to-end process the AI system has to fit into, and identify exactly where the model plugs in. Then we draw up a short list of possible AI approaches for precisely that point: LLM with or without RAG, classic ML, embedding search, rule-based, hybrid. Often two candidates remain, which are weighed against each other in phase 03.

Process mapUse-case shortlistApproach options
03

Decide: model, data, AI Act

The decide phase combines three considerations. We choose the model pattern (which type of AI and which specific model family), we test the data reality (do we have it, in what quality, in what processing context), and we establish the AI Act classification (prohibited, high, limited or minimal risk). Together, these three give direction to the prototype. Undiscovered data flaws or an unexpected high-risk classification come to the table here, not only during implementation.

Model choiceData checkAI Act classificationCost & latency
04

Prototype: small, working, real

We build a functional prototype that a user can actually use. For LLM use cases, we call Anthropic Claude or OpenAI via the API; for specialised use cases, we work with a suitable model. We build the UI quickly in Streamlit, v0 or a similar layer; integration flows in n8n or Make. The aim is not polish but realism: a real prompt, real data, a real model response and a UI that gives users enough context.

Anthropic / OpenAIStreamlit / v0n8n / MakeWorking flow
05

Test with real users

A handful of real users (your customers, your employees, or the stakeholders who will soon work with the system) go through the prototype in a structured session. We record desirability (do they find it valuable), usability (can they work with it) and trust (do they believe the output) separately. For AI systems, trust is often the real bottleneck: not whether it works, but whether people dare to trust it.

5โ€“6 usersStructured testDesirability ยท usability ยท trust
06

Synthesis: go/no-go and feasibility report

After the tests, we set the findings against the three sprint dimensions (desirable, technically feasible, compliance-fit) and formulate a well-founded go/no-go. For a go: a follow-up proposal with phasing, vendor and architecture direction, and an initial investment estimate. For a no-go: an honest overview of what the blocker is. No obligation to continue after the sprint.

Decision documentFeasibility reportAI Act noteFollow-up outline

The stack we work with during the sprint.

A sprint has to be fast: no waiting six weeks for a dev team to put together a mock-up. The toolstack is optimised for speed, with the option of a clean transition to a production architecture if it's a go.

Model layer

Frontier APIs with fallback

For most sprint prototypes, we work with the Anthropic Claude API or OpenAI. Where the use case calls for it, we consider Google Gemini, Mistral or an open-source route. That way, the concept doesn't depend on a single supplier.

  • Anthropic Claude โ€” strong instruction-following, long context
  • OpenAI โ€” broad tool coverage, fast iteration
  • Open-source โ€” when on-premises or data residency is required
UI layer

Quick mock-up for real tests

A sprint prototype doesn't need to be polished, but it does need to feel real. We build UIs in Streamlit (fast Python), v0 (Next components in minutes) or a lightweight custom app on existing UI libraries. The aim is a mock-up that doesn't get in the way of the user test.

  • Streamlit โ€” Python prototype UI in hours
  • v0 / Vercel โ€” polished front-end mock-up where it helps
  • Native mock-up โ€” when the use case needs to land on mobile or in-app
Flow layer

Integrations without a full dev team

Many AI use cases depend on what happens on the side: fetching data, putting output somewhere, sending a notification. We build those flows quickly in n8n or Make, so we can test the full pattern without a production integration. If it's a go, we translate the flow into the right architecture.

  • n8n โ€” open-source flows, can run on-premises
  • Make โ€” SaaS integrations stitched together quickly
  • Custom Python โ€” when specific logic needs more flexibility

What sets an Appfront sprint apart.

For AI concepts, it is crucial that the sprint is led by people who actually build AI systems, not by agency strategists who only know the model from slides.

What sets us apart

Builders instead of facilitators

The sprint is led by people who have worked on a production AI system this very week. The prototype gets the same code style and architecture choices as a real project: not throwaway work, but a focused first version.

What sets us apart

Vendor-independent

No reseller deal with Anthropic, OpenAI, Microsoft or Google. We choose the model that fits your use case, and we'll tell you honestly when open source is a better choice than a frontier API.

What sets us apart

Compliance built in, not bolted on

The AI Act runs as an overlay across the entire sprint, not as a separate paragraph in the report. For clients with sector-specific regulation (healthcare, financial, public sector), we bring in a compliance specialist where needed.

What sets us apart

Custom from the first idea

No standard curriculum. We build the plan from a short pre-sprint call: the nature of the problem, industry, data and systems, stakeholders. The five phases stay; how each phase is filled in differs per project.

What do you get at the end?

The sprint delivers a set of tangible outcomes that work as a package for the decision point: for your team, your management team, your board.

  • Validated concept with a substantiated go/no-goA clear verdict, based on desirability, technical feasibility and compliance, on whether this AI idea deserves the next investment.
  • Clickable working prototypeA functional prototype with a real AI API behind it and a UI mock-up. Available in a shared environment so you can keep using it yourself to explain things to stakeholders.
  • Technical feasibility reportRecommended approach (LLM, ML, vector search, hybrid), model choice with a fallback option, an indication of latency and cost model, and critical architectural points of attention.
  • AI Act classification noteProvisional classification (prohibited, high, limited, minimal risk), the Art. 9โ€“15 implications for high-risk systems, and links to GDPR, the need for a DPIA and DORA where relevant.
  • Data requirements overviewWhich data the system needs in production, which sources hold that data, and which quality and structural adjustments are needed before implementation can start.
  • User insights from the testsWhat worked, where users stumbled, and which additional trust and transparency requirements apply to your target group.
  • Follow-up plan (optional)If the decision is a go, a plan for the implementation track or a broader enterprise AI implementation. If it's a no-go: clear conditions under which it would be feasible.

How we factor the AI Act into the sprint.

The EU AI Act changes what an AI pilot is allowed to do, and what it must do in production. A sprint that ignores the Act delivers a prototype that is legally unusable. We take the Act into account from phase one, not as a hurdle but as a design constraint.

In phase 01 we draw up a provisional risk classification based on the problem framing. In phase 03, when the model pattern and data become concrete, we refine that classification, because the exact shape of a use case can tip between limited and high risk. With a high-risk classification the scope shifts: Articles 9 to 15 require a risk management system, data governance, technical documentation, logging, transparency towards users, human oversight and robustness, and we factor those requirements directly into the feasibility assessment.

We also work out the link with GDPR, the need for a DPIA and, for financial institutions, DORA. For sector-specific regimes (NEN 7510 in healthcare, the assessment framework for public organisations) we bring in the relevant references. The report includes a section you can hand straight to your legal or compliance officer; for deeper elaboration it ties in with our AI development approach.

What the sprint does not do is deliver a formal DPIA or audit report: that belongs to implementation, not validation. What it does do is stop you building on an assumption that the law already ruled out. Much of the waste sits in exactly that gap between "we heard it was allowed" and "it should have been done the way the Act prescribes".

How a sprint typically runs: three scenarios.

Scenario ยท Document AI

Smart case-file research for a professional

A team wants an internal assistant that helps lawyers or consultants work through large case files. Sprint question: does an LLM with RAG solve this, or is targeted hybrid retrieval a better route? Output: a prototype on one sample case file, tested with three professionals, with clear conclusions on hallucination risk and source attribution.

Scenario ยท Customer interaction

AI-triaged incoming customer queries

A service organisation is considering first-line triage by an AI agent. Sprint question: is this a chatbot problem, a classification problem, or a human-in-the-loop problem? Output: a prototype that routes 50 real queries, with an evaluation of precision, frustration risk and escalation path, plus an AI Act classification for this use-case category.

Scenario ยท Operational decision

AI-supported pricing or planning advice

An operator is considering a model that supports daily decisions: planning, pricing or allocation. Sprint question: is this an ML model or an LLM application? Output: a prototype on historical data, tested with the people who currently make the decision, and a clear distinction between "advice" and "automation".

Fabian van Dijk Business developer ยท Appfront

Frequently asked questions.

What exactly is the difference from an AI workshop?
An AI workshop or corporate training is broader and focused on building knowledge: what AI is, where it can land, which models, what compliance involves. An AI design sprint is narrower and focused on one concrete candidate use case: does this specific idea deserve an investment or not. A workshop leaves all the doors open; a sprint picks one door and checks whether there is really anything behind it.
And the difference from an implementation programme?
An AI implementation programme is execution: taking a validated concept into production, with governance, change management and handover. A sprint is validation: should the concept be implemented at all. Many clients see the sprint as a gate before the programme, and only direct budget towards implementation once the sprint has delivered a go. If you already have a validated concept, skipping the sprint makes sense.
When is a sprint not useful?
If you do not yet have a concrete candidate use case (then broader training or AI consultant advice makes more sense). If the idea has already been validated. If the real question is a strategy question at portfolio level. And if the organisation does not yet have enough data access to feed a prototype.
How do you handle the AI Act during the sprint?
On day one we do a preliminary risk classification, and on day three or four a refined version. For high-risk applications we work out the implications of Art. 9 to 15, not as legal advice, but as a design constraint. We link this to the GDPR and the need for a DPIA and, for financial institutions, to DORA. The sprint does not replace a formal DPIA or audit; that belongs to implementation. But it stops you building on an assumption that is already ruled out by law.
Do you work with our existing data and systems?
Preferably, yes: a prototype on a fictional dataset yields limited insight. For most sprints we work under NDA with a representative sample of your own data or a controlled extract. We work in your cloud tenant where the data classification requires it; for sprint purposes it can also be a ring-fenced environment on our side, depending on what the GDPR and your security team allow.
Which models and vendors do you use?
For most sprint prototypes we work with Anthropic Claude or OpenAI. Where the use case calls for it, we also weigh up Google Gemini, Mistral or an open-source route (Llama, Qwen). We build UI mocks in Streamlit or v0, and integration flows in n8n or Make. We are vendor-independent: no reseller deals, no mandatory integrations.
What if the sprint results in a no-go?
That is a legitimate outcome, not a failure. You have gained a clear answer that this specific idea is not the right next step. The final report states exactly where the blocker lay (data, model fit, user signal or compliance) and which changes would make it feasible in a possible second sprint. Both kinds of clarity are worth more than a half-hearted go-ahead.
How does the transition from sprint to implementation work?
If the decision is a go, we immediately deliver a follow-up plan: phasing, scope, model and technical recommendations, governance points to watch and a first indication of the investment. This fits seamlessly with our AI implementation programme or, for larger organisations, an enterprise AI implementation. You are under no obligation to run that programme with us. We will also give honest advice if a different type of partner is a better fit.
Is this suited to enterprises or to scale-ups?
Both. For scale-ups, it is often the first pre-investment gate for an AI feature on the roadmap. For enterprises, it is a structured way to bring clarity to one of several parallel AI initiatives before it lands in the broader governance layer. It is a workstream we also build in systematically for larger enterprise AI implementations.
What does a sprint cost?
That depends on the nature of the use case, the complexity of your data context, the number of stakeholders and how much pre-sprint preparation is needed. We work with a fixed sprint price that we provide after the intake, transparently broken down per phase. A sprint costs less than a misdirected implementation; that is the whole economic logic of this format.

Talk to us about your AI design sprint.

A half-hour introduction in which we go through the candidate use case you have in mind, the uncertainties around problem fit, approach, data or compliance, and how a sprint can give the sharpest answer to them. No obligation. Afterwards, we will send you a plan with phasing and transparent pricing.

Reply within 1 working day
No-obligation conversation
Westerdoksdijk 599, Amsterdam
Share this article: LinkedIn Email

Edit content