Studio
A practice, not an agency.
There is no agency behind this page. That is the point, not an omission.
I’m Yusuf Algan. Werk Studios is my product and AI practice — the company I use to take on advisory and product work, and to build products of my own.
I spent more than a decade on product and design at Shopify, eBay, Tajawal and DaWanda before moving into product and AI work in the Gulf. These days the work runs from the first research through the design and, more and more, the build itself. That last part matters: if I can build a rough working version, I can find out whether an idea holds up instead of writing a deck arguing that it will.
Most of the time I am called in by organisations with an ambition, a pile of data, and a product nobody is using. The useful question is almost never which features are missing. It is what someone is supposed to do differently — and whether they actually will.
How the work runs
Watch the work. Find the decision. Build enough to prove it.
- 1
Watch how the work really happens
I sit with the people doing the job and watch what they actually do — the spreadsheets, the WhatsApp threads, the workarounds. What gets described in a meeting is rarely what happens at the desk.
- 2
Find the decision that needs to get better
Nearly every request is a symptom of something else. I work out which decision is actually going wrong, before anyone argues about what to build.
- 3
Show something real, early
A map, a sketch, a working slice. People cannot disagree usefully with a description — they can disagree with something in front of them.
- 4
Test it properly
Does anyone understand it, trust it, and change what they do because of it? What happens when it is wrong, and who is allowed to overrule it?
- 5
Ship it, then keep watching
Launch is not the finish line. Who uses it, what it saves them, and where people still quietly go back to the old way of doing things.
What I do
Four ways I get brought in.
- 01
Working out where AI actually helps
I watch how the work really gets done, map the decisions people are making, and separate the ideas worth building from the ones that only sound good.
Before asking where AI fits, ask whether the process should exist at all.
- 02
Deciding what to build
Who it is for, which decision it improves, what the AI does and what it should not touch, and how someone checks or overrules it when it gets things wrong.
Executives do not need more dashboards. They need better decisions.
- 03
Building enough to test it
A working version rather than a slide deck — enough of the real thing to put in front of the people who will have to use it every day.
I build far enough to test the hypothesis instead of handing over a deck.
- 04
Getting it used
Keeping everyone pointed the same way, translating between the business, the engineers and the people doing the work, and making sure research keeps happening after launch.
A technically successful product that people avoid is still a failed product.
Track record
Three numbers I can stand behind.
The client work on this site is under NDA, so its numbers stay with the clients. These three are from earlier roles and are already public.
Shopify · 2019–2022
+7.5%
add-to-cart conversion in the SFN delivery-promise pilot
Led design for the delivery-promise work that became Shop Promise.
Tajawal, Dubai · 2018
+30%
conversion; Lighthouse 64 → 90; shown at Chrome Dev Summit 2018
Owned product and design for the mobile-web PWA team.
DaWanda, Berlin · 2015–2017
2.6 → 4.8
App Store rating; checkout conversion roughly +20%
Native iOS and Android apps — onboarding, discovery, review prompts.
The rest of the record — roles, dates, references — is on yusufalgan.com.
Two fixed shapes
When you want a start date and an end date before we talk.
Most engagements are shaped around the problem. These two are shaped in advance, so the first conversation can be about whether the fit is right rather than what the work would be. Quoted per engagement, like everything else.
- 01
Two-week readiness assessment
I sit with the people doing the work, find the decision that is actually going wrong, and tell you what is worth building and what is not.
- Length
- Two weeks
- You get
- A written read of the real problem, the decision it comes down to, and a ranked list of what to build, fix or stop.
- Feasibility
- Whether the data, the security posture and the systems it would have to plug into are ready for it. I do that first pass myself and bring in a specialist partner where it needs one.
- Right when
- Before committing a build budget. When there is a mandate and a pile of ideas but no agreement on the first one.
- 02
Six-week working prototype
One decision, one working version of the product that improves it, tested with the people who would use it every day.
- Length
- Six weeks
- You get
- A working prototype, the test evidence, and a plain account of what it would take to ship it properly.
- Right when
- When the decision is clear and the open question is whether people will trust and use the answer.
When the job is bigger than one person
It usually is. Here is how that works.
By default I lead and your own team builds and owns the result. That is the right outcome anyway: the product has to live with the people who will run it, and the fastest way to make sure it does is to build it with them.
When there is no team to hand to, I bring in independents I have worked with before — engineers, researchers, a designer when the surface is large — contracted for that engagement only. You still deal with one person, and that person is still accountable for the whole.
There is no bench here, and so no pressure to find work that keeps one busy.
On confidentiality
Most of the work cannot be shown.
Client work here is mostly under NDA — government, sovereign and enterprise projects where the screens, the data and often the client's name are not mine to publish.
Rather than fill the work page with invented dashboards, each case describes the shape of the problem: what everyone assumed, what turned out to be true, and what changed. Every diagram is labelled as an illustration, because a portfolio that quietly fakes its evidence is worth less than one that admits the limit.
Principles
Eight rules I hold AI products to.
Short enough to argue with. Each one has its own link, so you can send the one you disagree with. Break the mark and take one.