Questions & answers

No.

You don't need a company-wide strategy to start. Start with one piece of work worth doing better. If it works, you'll have something real to build on.

Start with work that repeats, takes a lot of time, or gets done differently depending on who's handling it.

It doesn't have to be simple. In fact, it gets interesting once the work needs your own rules, data, and experience to get right.

Show us the work you want help with. Ideally with a few real examples, solved the way you'd want them solved.

Prefer to start with a conversation instead? The calendar is open. contact.md →

That's exactly why we ask for real examples instead of long process write-ups.

Show us the work as it's already being done. We'll spend our time understanding the pattern, then come back with the specific gaps we still need answered.

Fastest if the work is already in the catalog. Setup typically takes about 2 weeks. Most of that time isn't spent on the technical side. It's spent learning your rules, exceptions, and gray areas.

Then we'll say so. You'll get a date for a final answer, even if that answer is no.

No. We start with people who are already good at their jobs, and look at what AI can take off their plate.

The point isn't to squeeze more tasks into the same workday. It's to free up capacity and headroom for the work where the human makes the biggest difference.

Then it says so. It's not built to make up the next-best answer just to keep moving. Where a source exists, it shows what the answer is based on.

A chatbot is built for a conversation. What we build is built for a piece of work.

It has one defined job, set up around how you actually get it done.

Whatever systems the work already happens in: email, calendar, CRM, accounting, documents, customer service, and so on. We start with read-only access and only add more once you're ready for it.

No. The setup isn't locked to one AI provider. You can switch model or provider whenever it makes sense.

No. Not unless you've approved it.

The technology is rarely the hard part.

The hard part is getting out how the work actually gets done. Which rules apply? When don't they? Which systems are used? And what does the experienced employee know that was never written down?

That's why we start with real examples of the work — not a two-line description of the process.

No. If everything were already written down perfectly, our job would be a lot easier.

We start from real examples and the people who know the work. Along the way, we pull the rules, exceptions, and workflows out of their heads and make them concrete. So the documentation gets better as part of the work itself.

That's pretty normal.

We work out what data the AI actually needs. Some of it gets cleaned up once. Other parts get more structured over time, because the AI needs the information to live in one place and mean the same thing every time.

So you don't need perfect data to start.

Yes, and that's often where the real work is.

"Always do X" is easy. The interesting part is: "Do X, unless the customer looks like this, the order looks like that, or this has already happened." Those are the rules and gray areas we teach it together with you.

Not just for the AI's sake.

We start with the work as it looks today. If something already works well, the AI learns it as-is. If we find a workflow along the way that only exists because "that's how we've always done it," we can have that conversation when we get there.

Then the setup changes along with it.

New rules, new products, and new exceptions keep showing up. That's why we don't treat a build as something we hand over and forget. It gets better by staying close to the real work.

First of all, people get better at their own work. When the research is done and the next step is already prepared, more of the day goes to what the person is actually good at.

When AI has to learn the work, rules, exceptions, and workflows get made concrete. That makes quality more consistent, onboarding easier, and the company less dependent on knowledge that only a few people hold.

And when the repetitive, context-heavy work takes up less room, there's more headroom for judgment, for customers, for colleagues, and for relationships.

That's one of the first things we go after.

Almost every company has work that's "just how it's done" because one person has done it that way for 5 years. The problem only shows up the day that person is out sick, changes jobs, or has to teach someone else.

When we build AI into the work, that knowledge has to come out of someone's head and get made concrete. The rules, the order of steps, the exceptions, the examples — they all become part of the setup.

So the AI doesn't just get something to work from. Your company gets a better grip on its own knowledge.

Yes. A new hire doesn't have to piece the whole process together from scratch, or get lucky enough to ask the right person.

The work, the rules, and the examples are already made concrete. And the AI can help them do the work the same way the rest of the team does, from day one.

Yes. A generational succession isn't just about ownership. Decades of knowledge can be built into how the company works.

When workflows, rules, and exceptions get made concrete, less of the company depends on specific people remembering how things are usually done.

Yes. And you should.

How long did the task take before? How long does it take now? How much gets produced? How often does something need fixing? Where does the work tend to stall?

As the work gets more structured, it also gets easier to measure. So the value doesn't have to end up as a gut feeling that "AI probably saves some time."

Yes. Once work can be measured, the human effort behind it gets clearer too.

If someone with the AI behind them can handle more customers, deliver more, or raise the quality, you can actually see the difference. That makes it easier to talk specifically about the value a person creates, instead of just hours at a desk.

It shouldn't be.

For the company, it can mean more capacity, better quality, and more customers, without the workload growing at the same rate.

For the employee, it means less time on repetition and more time on what they're actually good at. And once the work is measurable, their contribution is easier to see too.

If you can handle twice the work in the same hours, or deliver noticeably better results than a year ago, the company shouldn't be the only one who sees the value in that. It's also a pretty good place to start a conversation about a raise.

Usually more about your own work than you'd expect.

To teach the AI a task, we have to work out what a good result looks like, which rules apply, where the data comes from, and what happens when the normal case breaks down.

That's often the moment you realize how much of a process was never actually written down.

Your knowledge, methods, and experience stay with you. The AI model itself runs on the provider you use, like Anthropic or OpenAI.

What we run for you, we run from Denmark. The model itself may run on a provider outside Europe.

How data can be used and sent is something we agree on as part of setup.

You can. It starts with read-only access, and you decide when and where it gets more.

Yes. You can always pull the plug. And what we've set up for you, you keep.

It depends on the work.

A new piece of AI work gets scoped as a Project: an agreed outcome, a timeline, and a fixed price. You get the number before we start, not after.

If you want us involved on an ongoing basis afterward, that's Department. It's all set out under Pricing. pricing.md →

No. There's no subscription to cancel.

A Project is scoped: an agreed outcome, a timeline, and a fixed price. Nothing follows on automatically once it's done.

In a Project: building what we agreed, putting it into operation, and staying on through an agreed stabilization period.

If you want us to maintain it afterward and help when something breaks, you can add a Care agreement.

Then we build it. That's how most of this starts — the examples are work we've solved before, not a menu to pick from.

Big or small, it gets scoped as a Project, and the price follows the scope.

No. You need to know the work. We handle the technical side.

What matters most is that you can show us how the task gets done right, including the exceptions and gray areas.

No. It runs inside the AI you already use, like ChatGPT or Claude, and connects to the systems the work already happens in.

First, we build it so it doesn't just guess its way forward when it's unsure.

Where approval is required, the work goes through a person. And if something still goes wrong, there's someone on our side to call.

Ellypsis was founded and is run by Sewar Sidou, in Odense, Denmark. Want the rest of the story? It's in the letter. letter.txt →