Donnerstag · Technology & AI

Build with AI.
Run with Trust.

Our principles for creating useful enterprise AI: ambitious in what it enables, disciplined in how it operates, and designed to keep experts in control.

Talk to an expert

Explore the platform

Our view of enterprise AI.

Nine principles that govern how we build, and what we will not automate.

At Donnerstag, we believe AI creates the greatest value when it helps experts build better systems of work, not when it is added indiscriminately to every task. Our aim is to turn domain knowledge into durable, governed operations that continue to improve outcomes long after an individual prompt or conversation ends.

We use AI where intelligence is genuinely useful: to accelerate engineering, adapt software to the specific reality of each customer, and handle information or situations that cannot be resolved reliably through fixed logic alone. Wherever work can be made predictable, reproducible and auditable, we prefer trusted software and approved rules. Intelligence belongs where the problem is uncertain. Predictability belongs where the outcome must be trusted.

This philosophy also shapes how we treat data and human responsibility. We minimise the information exposed to AI, apply models selectively and keep clear boundaries between an AI suggestion, a rule-based result and a human decision. Automation should remove repetitive preparation and manual coordination, not accountability or expert judgment.

Our long-term ambition is to bring technology closer to the people who understand the work. Finance professionals should be able to shape the systems that support them without translating every change through consultants, engineering backlogs or fragile spreadsheets. We build with AI so that our customers can run with greater trust, ownership and control.

We do not measure progress by how much AI we use. We measure it by how much meaningful work the technology improves.

The principles in one view

Value before novelty

Beyond the chatbot

AI builds. Trusted systems run.

Intelligence where the problem is uncertain. Predictability where the outcome must be trusted.

Data restraint is a feature

Human judgment is not a failure of automation

Software should adapt to the customer’s reality

Technology should move closer to the expert

Our engineering commitment

AI is not the outcome. Better work is.

Artificial intelligence can interpret language, write software, identify patterns and work with information that traditional systems could never handle reliably. But technical possibility is not the same as customer value.

At Donnerstag, we do not use AI simply because a model can perform a task. We use it when it helps solve a real operational problem, improves an important decision or makes previously uneconomic work possible.

Technology matters when it changes what people and organisations can achieve. The hardest engineering challenge is not building something impressive. It is building something that remains useful, reliable and valuable in the reality of a customer’s work.

That principle guides how we build our company, our product and our relationship with AI.

Value before novelty

We are engineers because difficult problems are interesting. We build a company because solving the right problems matters. Both motivations belong together.

Technical ambition without a meaningful outcome becomes experimentation for its own sake. Commercial ambition without technical depth produces shallow tools that cannot address the complexity underneath the problem.

We begin with the work itself: what the customer is trying to accomplish, why the current process fails, where time and money are lost, which decisions require expertise and which parts should become repeatable. Only then do we decide where AI belongs.

We do not measure technological progress by how much AI we use. We measure it by how much meaningful work the technology improves.

Beyond the chatbot

Chatbots are useful. They make information easier to access and allow people to interact with software in more natural ways. But a conversational interface is only one expression of AI.

The deeper opportunity is to describe the shape of an operational problem and use AI to help build a lasting way of solving it. A finance professional should not have to ask the same question every time a new invoice arrives. A team should not rebuild the same spreadsheet or restart the same technology project whenever a process changes.

The shift is from answering one question to improving the process that produces the answer; from resolving one exception to creating a reliable way to handle future exceptions; from generating one result to building a system that keeps producing results.

Do not use AI only to solve the next case. Use AI to build the system that handles the next thousand.

AI builds. Trusted systems run.

We use AI in distinct ways. It increases the reach of our engineers. It helps adapt our product to the systems, data structures, rules and requirements of each customer. And it can assist within selected processes where information is genuinely unstructured, ambiguous or too variable for fixed logic alone.

These are different jobs and should remain visible as different jobs. AI can help create a document reader; the reader can then run as software. AI can help translate a policy into executable logic; reviewed and approved logic can then run repeatedly. AI can interpret an unusual message; its interpretation should remain distinguishable from a deterministic result or a human decision.

We do not believe every transaction needs to be reconsidered by a probabilistic model every time it moves through a process. If work can be expressed as stable, inspectable and reproducible software, it should be.

AI should help build the system. It should not be forced into every execution of the system.

Intelligence where the problem is uncertain. Predictability where the outcome must be trusted.

Some work benefits from interpretation. Language varies. Documents change. People make typing errors. Information arrives in unexpected formats. A supplier may describe the same issue differently in every email. AI is valuable in these situations because it can work with ambiguity.

Other work should not be ambiguous. When a system checks whether a payment was received, compares an invoice rate with an approved rate or applies a defined tolerance, the same inputs should produce the same result. The logic should be understandable and the outcome reproducible.

Using an AI agent does not automatically make a process more advanced. In some cases it makes execution more expensive, less predictable and harder to audit.

If a process can be made reproducible, auditable and trustworthy, we build it that way first. We use AI where a predictable process cannot solve the problem well enough.

Data restraint is a feature

Enterprise AI is often approached by beginning with the largest possible dataset: collect everything, upload everything and search for value afterwards. We take the opposite approach.

We begin by asking what the process needs, what the model needs and whether the use of AI creates a defined benefit. In many cases, AI can help configure a process without examining every transaction once that process is running. In other cases, a model may need access to a specific document, message or field. That access should be purposeful and limited to the task.

This creates a more disciplined set of design questions: What information is required? Why is it required? Where will it be processed? What result will follow? Can the same value be achieved with less exposure? Can recurring execution run without a model?

We do not begin by asking how much data we can give to AI. We begin by asking how little AI needs in order to create real value.

Human judgment is not a failure of automation

Not every human decision is inefficiency. Finance includes policies, commercial relationships, risk choices, materiality judgments and responsibilities that cannot simply be delegated because a model produced a likely answer.

AI and software should remove the work that prevents experts from exercising judgment well. They should collect information, structure evidence, perform repeatable checks, identify issues and prepare the next step. Where accountability matters, people must remain able to understand the situation and make the decision.

An AI recommendation is an AI recommendation. A rule-based result is a rule-based result. A human approval is a human approval. Systems should not blur those distinctions.

The goal is not to remove the expert from the process. It is to remove everything that prevents the expert from applying their expertise.

Software should adapt to the customer’s reality

Enterprise processes look similar from a distance. Up close, the differences matter. Documents contain different information. Systems identify the same object in different ways. Contracts use different structures. Teams apply different tolerances and approvals.

Traditional software often responds with a large configuration catalogue. The customer receives a general template and becomes responsible for adapting it to the real work. Traditional custom development can reflect that reality, but it is expensive, slow and difficult to maintain.

AI creates a better possibility: a stable product foundation can be adapted more deeply to the specific systems, information, rules and workflows of each organisation. The customer can receive software shaped around the operation without inheriting a bespoke engineering project.

Built for you, not configured by you.

Technology should move closer to the expert

For decades, the people who understood a business problem and the people who could change the software were often separated. A finance professional identified the need. Consultants, project teams and engineers translated it through multiple handovers. Every handover created delay and lost context.

AI can reduce that distance. The long-term opportunity is for the person who understands the process to participate directly in shaping the system that supports it: explaining a new document, a changed rule, a required approval or an unfamiliar exception, while governed technology translates that knowledge into a reliable process.

This does not mean unrestricted control over production software. Empowerment still requires review, permissions, testing and clear boundaries. It means domain expertise should no longer have to pass through unnecessary layers before it can improve the system.

The future of enterprise technology belongs to the people who understand the work, not only to the people who know how to code.

Our engineering commitment

We use AI extensively to build Donnerstag. That does not reduce our responsibility for what we build. It increases it.

AI allows experienced engineers to explore more possibilities, move faster and take ownership of larger areas of the product. But speed without understanding is not engineering discipline.

Our standard is trust and verify. We review code, test behaviour, examine how data moves and design for security, privacy and auditability. We expect engineers to understand the technology they are responsible for, even when AI helped produce it.

We believe AI should increase the leverage of experienced people, not provide an excuse to abandon experience.

Every acceleration creates a corresponding responsibility to review, test and understand.

What we refuse.

We will not use AI where a simpler and more reliable system is better.

We will not treat enterprise data as undifferentiated input for a model.

We will not hide consequential decisions inside a black box.

We will not confuse automation with the removal of accountability.

We will not present a probabilistic suggestion as a verified fact.

We will not ask customers to trust an outcome they cannot understand.

We will not force AI into every transaction merely to make a product sound more advanced.

We will not call a conversational interface a complete transformation of enterprise work.

We will not pursue technical novelty without a clear connection to customer value.

The future we are building

We are building a future in which finance professionals spend less time repeatedly solving the same operational problem and more time improving the system that solves it.

A future where expertise can be translated into software without disappearing into a development queue. Where recurring work becomes reliable, while unusual cases receive more attention. Where AI can work with ambiguity without turning every process into a black box.

A future where organisations benefit from advanced technology without losing control of their data, decisions or responsibilities. People do not become less important because systems are more capable. They become more effective because systems carry more of the repetitive burden.

That is the opportunity we believe in: not AI replacing expertise, but AI extending it. Not one more answer, but a better system for producing the answer every time.

Our pledge

We will build technology with a purpose and use AI ambitiously, but never indiscriminately.

We will design for security, transparency, reproducibility and control, and keep human judgment where human accountability matters.

We will bring technology closer to the people who understand the work, and use AI not just to solve today’s problem, but to build a trusted system that keeps solving it tomorrow.