# Proof of Concept, Pilot or MVP: The Right Stage for Your AI Initiative

Author: Marin Niehues, Niehues Consulting · Published: 2026-05-07 · Updated: 2026-11-16 · AI Adoption & Operations
Source: https://www.niehuesconsulting.com/blog/proof-of-concept-pilot-or-mvp-ai/

A proof of concept establishes technical feasibility, a pilot proves value in the real process and an MVP proves operation with real users. Organizations that choose the stage by the open question release larger budgets only once the previous question has been answered. This article sets out what each stage must deliver, how to recognize the right one and which criteria decide whether to continue or stop.

**Each stage answers exactly one question. The choice follows the largest open uncertainty of the initiative, and every stage ends with a decision tied in advance to measurable criteria.**

## Why the choice of stage determines the budget

The choice of stage determines which risk an organization tests with which investment. An initiative that starts at the wrong stage answers a question nobody asked and leaves the decisive one open.

The need for this distinction grows with the spread of AI. According to the official German survey on the use of information and communication technology, around 26 percent of companies in Germany with at least ten employees used AI in 2025, and 57 percent of large enterprises with 250 or more employees. Many now face the second decision: which initiatives go beyond the first tool?

An interview study with 65 experienced data scientists and engineers identifies the most common cause of failed AI projects as sponsors and teams understanding the problem differently, followed by missing data and infrastructure. Both findings argue for clarifying problem and value before investing in integration.

**The starting point in figures**
26 % | Share of companies in Germany with ten or more employees that used AI in 2025
57 % | Share among large enterprises with 250 or more employees in the same year
65 | Practitioners interviewed in a study on the root causes of failed AI projects

## What is the difference between a proof of concept, a pilot and an MVP?

The difference lies in the question each stage answers: the proof of concept tests whether the technology can solve the task. The pilot tests whether the process measurably improves with it. The MVP tests whether the solution works durably in operation with real users.

A **proof of concept** runs in a protected environment with selected or historical data. It answers a technical question, for example whether a language model extracts incoming invoices with sufficient accuracy or whether a forecasting model detects failures early enough from existing sensor data. Users are barely involved and the process remains unchanged. The result is a substantiated statement on feasibility.

A **pilot** runs in the real process with real data and a limited group of future users, such as one team or one site. It measures the metric the business unit defined in advance against its baseline. The result is proven value on which the business case with real figures is built.

A **minimum viable product** is the smallest version of the solution that serves real users in day-to-day operations. For AI in the enterprise the MVP is connected to the business systems and has controlled access rights, monitoring and a support route.

**The three stages compared**
Criterion | Proof of concept | Pilot | MVP | With Niehues Consulting
Guiding question | Can the technology solve the task? | Does the process measurably improve? | Does the solution work durably in operation? | Guiding question fixed in writing before the start
Data | Selected or historical data | Real data in the live process | Real data from connected business systems | Data access clarified before the build
Users | Project team and individual experts | Limited group of future users | All users of the first area | Users involved from the pilot onwards
Success criterion | Accuracy against a threshold | Metric against baseline | Usage rate and cost per transaction | Metric with baseline before the start
Typical duration | Two to six weeks | Six to twelve weeks | Three to six months | Decision date at the end of every stage
Result | Statement on feasibility | Confirmed business case | Solution in day-to-day operations | Handover to a named operational owner

## What is the difference between a pilot project and a proof of concept?

A proof of concept tests the technology outside the process, a pilot project tests the value inside the process. The proof of concept asks whether something works. The pilot asks whether it pays off.

This leads to the most expensive mistake in AI initiatives: a successful proof of concept is read as proof of value. High accuracy on selected test data says little about whether a case handler works faster with real, incomplete cases.

Research shows why only the real process provides this insight. A field experiment with 758 consultants found clear gains in speed and quality for tasks within the capabilities of the model. For tasks outside that frontier, participants using AI were 19 percentage points less likely to reach a correct solution than those without. Where the frontier runs in a specific process only becomes visible with real cases. A field study with more than 5,000 customer service agents also found 14 percent more issues resolved per hour on average and 34 percent for new and less experienced agents. The effect of an application therefore varies with the user group, and this is precisely the information the pilot delivers.

## Which stage suits which AI initiative?

The right stage is the one that removes the largest open uncertainty of the initiative. Two questions lead to the decision.

**Is technical feasibility open?** For tasks new to a model, data of unknown quality or high accuracy requirements, the initiative starts with a proof of concept. Examples are the evaluation of handwritten forms or forecasts from machine data.

**Is the technology proven, but the value in your own process open?** For established application patterns such as suggested replies in customer service or document summarization, feasibility is established. A proof of concept would repeat an answered question. The initiative starts directly with a pilot.

Once feasibility and value are proven, for instance through a pilot with a confirmed metric or a tested solution being transferred to another area, the MVP follows. The effort then shifts to integration and user enablement.

**Checklist for choosing the stage**
- Has a model demonstrably solved this task with comparable data? If not: proof of concept.
- Is the required data available digitally, and does a responsible unit know its quality? If not: proof of concept with a data review.
- Has the business unit named a metric with a baseline? If not: clarify before any stage.
- Is the value proven in your own process? If not: pilot.
- Is a unit named that owns operations from the first day? If not: no MVP.
- Is budget for operations and enablement planned? If not: no MVP.

## How long does each stage take?

A proof of concept is planned for two to six weeks, a pilot for six to twelve weeks and an MVP for three to six months until approval for operations. These planning figures apply when data is available and interfaces are accessible.

The duration of the pilot follows the metric. If fewer than a few hundred cases arise in the pilot period, extend the duration or enlarge the pilot group. Users work more slowly with a new tool in the first weeks, so measurement starts only after familiarization.
## What does an AI proof of concept cost compared with a pilot?

A proof of concept costs a fraction of a pilot because integration and user support are not required. The cost of each stage results from person-days and licenses plus the time of the business unit.

For external support, the fee study of the German association of management consultancies cites daily rates of around 1,000 to 2,500 euros, depending on seniority and specialization. Estimate the effort per stage with a simple calculation: people involved times weeks times days per week times daily rate, plus licenses and computing capacity.

**Calculation with assumed values**
- Proof of concept: two people, four weeks, three days per week each make 24 person-days.
- Pilot: three people, ten weeks, three days per week each make 90 person-days, plus the time of pilot users for training and feedback.
- MVP: integration into the business systems, hardening and enablement of all users in the first area, typically a multiple of the pilot.

The largest cost lever is the sequence: a proof of concept stopped after four weeks is cheap, an MVP that fails after six months on a feasibility question is not. How to set value against these costs is covered in [AI business case: calculating benefits and costs robustly](/blog/ai-business-case-benefits-and-costs/).

## Set stop criteria before the start

Every stage ends with one of three decisions: continue, adjust or stop. The criteria for this are fixed before the start and are measurable.

For the proof of concept this is a threshold, for instance a minimum accuracy in recognizing invoice fields, measured against a test set provided by the business unit. For the pilot it is a change in the metric against the baseline, for instance a handling time at least 20 percent shorter at the same error rate. For the MVP it is the usage rate and cost per transaction after 90 days.

Without criteria fixed in advance, the permanent pilot emerges: an initiative that is neither transferred nor stopped because nobody defined when it had proven enough. This pattern is described in [Why AI pilots fail to reach operations](/blog/why-ai-pilots-fail-to-reach-operations/). A decision date with a named decision-maker at the end of every stage prevents it.

## The transition from pilot to MVP is the critical stretch

The transition from pilot to MVP succeeds when it is planned as a phase of its own with budget and operational ownership. What the project team compensates for by hand in the pilot has to exist as a regular process in the MVP.

Two commitments precede this transition. A unit takes over monitoring and support. A product owner in the business unit decides on expansion and priorities. Added to this is training all users in the first area on their own process. How this transition is built is covered in [Delivery of AI solutions](/blog/ai-solution-delivery-prototype-to-product/). The route to a portfolio of several applications is set out in the article [From pilot to scale](/blog/from-pilot-to-scale-ai-operations/).

## Niehues Consulting as a partner for the sequence of stages

[Niehues Consulting](/) takes AI initiatives from the first proof to day-to-day operations and works according to the Niehues Delivery Model with the phases Understand, Build and Implement. Each phase ends with fixed review questions that decide on the next stage.

Three of them matter for the sequence of stages. Before every stage: what is the baseline, and how do we measure success? At the end of the pilot: does the prototype prove its value in the real process with real data? Before the MVP: who owns operations from the first day? An initiative that leaves one of these questions open does not move to the next stage.

Founder Marin Niehues is a management consultant who works from the delivery perspective. He builds organizations that deliver and brings AI into operation, as delivery manager, project manager or product owner. His projects range from VC-funded scale-ups to DAX 40 corporations, in 20+ countries on six continents, with his home base in the DACH region. In a free [30-minute initial consultation](/#kontakt) you can establish with him which stage your initiative has reached and which question needs answering next.

**The stage follows the open question**
A proof of concept pays off when feasibility is open. A pilot pays off when the value in your own process is open. An MVP follows once both are proven and operational ownership is settled. Organizations that close every stage with criteria fixed in advance and a decision date invest in rollout only once the value is proven.

**Sources**
- Federal Statistical Office of Germany (Destatis): Use of information and communication technologies in enterprises, 2025
- Ryseff, De Bruhl, Newberry: The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed, RAND Corporation, 2024
- Dell'Acqua et al.: Navigating the Jagged Technological Frontier, Harvard Business School Working Paper, 2023
- Brynjolfsson, Li, Raymond: Generative AI at Work, NBER Working Paper, 2023
- Bundesverband Deutscher Unternehmensberater (BDU): Fees in management consulting, 2022

## Frequently asked questions

### What is a proof of concept in AI?

A proof of concept in AI is a short, bounded test that establishes whether a model solves a specific task accurately enough with the available data. It runs outside the real process, usually with selected or historical data. Its result is a statement on feasibility, not proof of value.

### What is an MVP in AI projects?

An MVP is the smallest version of an AI solution that serves real users in day-to-day operations. It is connected to the business systems and has monitoring and a support route. It is extended with every expansion stage.

### Does every AI initiative need a proof of concept?

A proof of concept is only needed when technical feasibility is open. For proven patterns such as suggested replies or document summaries, start directly with a pilot. A proof of concept for an answered question costs time without producing insight.

### How many users does an AI pilot need?

An AI pilot needs enough users for the pilot period to generate sufficient cases for a robust measurement. In practice this is one team or one site with a few hundred cases. The group mixes experienced and less experienced staff, because the effect differs strongly between them.

### When should an AI pilot be stopped?

An AI pilot is stopped when the metric fixed in advance misses its threshold after familiarization and no adjustment is in sight that would change this. Stopping on clear criteria is a usable result. What becomes expensive is a pilot that continues without a decision.

### Is a prototype the same as a proof of concept?

A prototype is the artefact that has been built, a proof of concept is the purpose for which it is used. The same prototype can establish feasibility in a proof of concept and prove value in the real process during a pilot. What matters is which question the stage is meant to answer.
