AI for your business
AI automation for businesses
Not yet another chat window on the side, but AI at the point in your process where someone now has to read, sort or write. Built into the software you already use, with a review step included.
AI automation here means that a model takes over work that requires judgement: reading a document, classifying a message, writing a draft. Darvis builds that step into the software you already use, with a review step before or after it and a recorded trail of what happened. We start with one well-defined step that runs alongside the existing work, so you see how often it is right before it takes anything over.
Does this sound familiar?
AI only becomes interesting once you can point to the work it takes over. This is the kind of work we mean.
“Someone reads through the inbox every morning”
Enquiries, invoices and questions arrive all mixed up. Someone has to read and sort them before anything happens.
“We retype data from PDFs”
Supplier invoices, packing slips or forms arrive as documents. The data in them has to be entered into the system by hand.
“Every quote starts with the same text”
The content differs per customer, but the structure and the standard passages are always the same. Yet someone writes them from scratch every time.
“The answers are spread over twenty documents”
Your staff search through manuals, folders and old emails for something that exists but cannot be found.
“ChatGPT works, but not in our system”
Colleagues paste data into a chat window and copy the result back out. That helps, but it is not a process and nobody can see what is happening.
Where it works
Four applications that pay for themselves
Well-defined steps with a clear input and output. That is where the technology is reliable enough to take something off your hands.
Reading documents
An incoming invoice, packing slip or form is read and the data enters your system as fields. A person checks the exceptions instead of everything.
Sorting mail and enquiries
Incoming messages are classified and assigned to the right person or file, so they can be acted on straight away.
Writing drafts
A first version of a quote, a reply or a report, built from your own data and in your own tone. An employee reviews it and sends it.
Searching your own documents
A search function across your own manuals, contracts and files that answers with the source attached, so a colleague can check it.
Not sure yet where to begin? Then AI consultancy is the first step: mapping out what there is to gain, without building anything.
Scope
When this fits and when it does not
AI suits work in which someone has to interpret something and in which the occasional mistake can be caught: a message filed in the wrong place, a draft that gets polished. There does need to be a person in the middle or at the end.
It does not suit steps that must always turn out the same, such as calculating a price, posting an invoice or granting a discount. Those are rules, and rules belong in software as rules, where you can read them back and where they do the same thing every time. That is why we rarely start with AI: what can be done with a rule, we do with a rule. That is the subject of business process automation.
Nor does it suit a decision you must be able to justify to a customer, an accountant or a regulator without a person having looked at it. In that case we build it as a tool that suggests, not as a step that decides.
How it works
Measure first, then take over
You do not have to rely on a promise. You first see how often it is right.
Finding where judgement is needed
We go through your process and look for the steps where someone has to read, assess or phrase something. Those are the places where AI adds value.
Starting small and measuring
We take one step and first run it alongside the existing work. That way you see how often it is right before it takes anything over.
Building it into your system
What works, we build into the software you already use, with a review step included. Not a separate chat window next to your process.
Keeping track of what it does
Every action is logged: what came in, what the model made of it and who approved it. Without that trail you cannot explain anything afterwards.
Have you already built something yourself with Lovable, Bolt, v0 or ChatGPT that will not go live? That is a different question, and we cover it under taking an AI app live safely.
Frequently asked questions about AI automation
What happens to our data when we use AI?
We agree on that in advance and put it in writing. We work with business services under which your input is not used to train models, and we only send the data needed for that one step. Anything that does not need to go, does not go.
What if the model gets something wrong?
That is why we almost always build in a review step and start alongside the existing work rather than on top of it. You first see how often it is right. For steps with financial or legal consequences, a person keeps the final say.
Can you connect AI to the system we already use?
Yes, and that is usually the core of the assignment. Standalone AI tools often work but sit beside your process. We build the step into the software your people already work in, so nothing needs to be copied and pasted.
Is AI automation only for large companies?
No. The applications that pay off fastest are actually small: reading documents, sorting mail, drafting a text. Those are well-defined steps you can have built one at a time.
Where do we begin if we are not using AI at all yet?
With one step that currently involves reading or typing work. We first look over your shoulder and map out what there is to gain, so you start not with a system but with a result.
What does AI automation cost?
That depends on the step and on where the data comes from. We start with an assessment and a quote per component, so you decide piece by piece. You will receive a response to your enquiry within 24 hours.
Among other things, we connect with Microsoft 365, Azure and the AI services of Anthropic, OpenAI and Google. See the integrations we provide.
Which reading work takes up most of your time?
Tell us what gets read, sorted or retyped every day. Within 24 hours you will hear whether AI has a role to play there.