How we work

After launch we stay in it. We help you run the system in production, tune it on real usage, watch the metrics that matter, and own the exceptions alongside your team, until you are ready to run it without us. Most vendors leave at handoff. We treat go-live as the start of operating, not the end of the engagement.

A consultancy usually hands over a strategy and leaves the building to someone else, and a lot gets lost in between. With us, the team that shapes the plan also builds it and helps run it, so the thinking carries through to the finished system.

With discovery. We look at your business, your data, and where AI genuinely fits, then tell you what is worth building and what to skip, before any commitment to a build. If the honest answer is that AI is not the right move for a given problem, we will say so.

As long as it helps. Some engagements are a scoped discovery and roadmap. Others run for months as we build and operate alongside your team. There is no lock-in: we hand off cleanly whenever you are ready to own it, with the documentation and access to do so.

A partner. We run the same kinds of AI systems internally that we build for clients, we share the operating burden after launch, and we are honest about what not to build. You are working with a team that operates production AI daily, not handing your problem to a black box.

If you have the time and an engineer who has built these systems before, you may well manage. What we offer is speed and certainty: we have already learned where these projects go wrong, so we get you from a promising demo to something dependable much faster.

You own all of it. Everything we build sits in your own accounts, documented so your engineers can understand and extend it. If we stopped working together the day after launch, the system would keep running.

We agree a specific first version with a clear price, then you decide whether to go further. That keeps the commitment small and costs far less than building an in-house AI team for the same work.

It is realistic for a focused first version: one system handling one piece of work on your real data. A large platform takes longer, and we will tell you so on the first call. We keep to the timeline by starting small and building up.

We assume it sometimes will and design for it. We test the system against real cases before launch, watch it closely once it is live and keep a person in the loop wherever a mistake would be costly.

Yes, and that is where we are most useful. We build around your process, your data and your own rules for what counts as a good result, so the system copes with the details that make your business different.

No. It takes repetitive work off your people so they can focus on the cases that need them. The system handles the routine volume and passes the rest to a person, and someone on your side always owns the exceptions.

We lead the team ourselves and answer for its results. You deal with one accountable person at 38shift, and the engineers stay with your product for the long run, so what they learn about it stays with you.

Kosovo has a young, well-trained engineering community, and our founder has built and led engineering teams here for more than twenty years. Prishtina shares its working day with the UK and Europe.

Yes. When your business is ready, we help you move the team across under terms agreed at the start, so the change is smooth for you and for the engineers.

It depends on the size of the team and the skills it needs. We agree the shape of the team on the first call and give you a realistic start date before anything is signed.

We protect it from the start. Outreach goes out from separate, properly prepared addresses, volumes stay modest and we watch replies closely. If anything starts to slip, we pause it straight away, because a few well-aimed emails are worth more than a damaged reputation.

Niche markets are where this works best. Each campaign is written for one clearly defined type of customer and tied to a real event in their business, such as new funding or a new leader. The narrower the audience, the more specific the message.

No. We build and run the whole system, and you get qualified meetings and customer records that stay current. We need a few hours from you early on to agree who you want to reach and how to describe what you do. After that, the running of it sits with us.

It takes away the manual research and data entry, which is where they lose most of their day. People keep the important parts: approving the messaging, holding the sales conversations and handling anything unusual. Closing the deal deliberately stays with your team.

Usually not. We work with the tools you already have, including HubSpot and Salesforce, and add our system on top. If one of your tools gets in the way, we will tell you and explain why.

First campaigns usually go live within a few weeks. Early volumes stay small on purpose while the system settles in. We do not promise a number of meetings, because that depends on your market and your offer, and anyone who promises one before starting is guessing.

It builds on it. Strong search rankings still matter, because assistants draw on the same web. AI search also rewards clear answers, consistent facts about your business and mentions in trusted places, so the work goes further than keywords.

Nobody can honestly promise that, because the assistants decide what to show. What we can do is remove the reasons you are left out and measure the change, so you can see whether the work is paying off.

Changes to your site can be picked up within weeks. Being recommended more often builds over months, as more trusted sources mention you. We report on progress from the first month.

Yes. Our own site offers a version of every page written for AI assistants and describes exactly what we do, and we track how the assistants describe us in return.

No. We are based in Prishtina and have shipped for local businesses here, and we work remotely with founders and teams across the US, UK, and Europe. Local clients get people on the ground; remote clients get a senior team at strong value with reasonable overlap on European and US East Coast hours.

We are a focused, senior, remote-first team, not a 200-person body shop. You work directly with the people building your system, which is the upside. The tradeoff: we take on a limited number of projects at once, and if you need a fifty-seat outsourced staffing contract, we are not that.

We ship a thin version to production early instead of disappearing for months. Exact timeline depends on the integrations, but the goal is a working system handling real cases in weeks, then tuning. Honest note: voice and CRM builds that touch third-party systems take longer to wire up than a self-contained agent.

It will sometimes, so we design for it. Every system has escalation paths that hand edge cases to a human, plus logging so you can see exactly what happened. For the restaurant, complex requests go to staff rather than the model improvising. We tune against real traffic in the first weeks, not in a lab.

Yes. We are in Prishtina, we have shipped for a Prishtina business, and we track the local landscape, including the digital-maturity gap across Kosovo banking. That context matters when the system has to fit how business actually runs here in the Balkans, not a generic template dropped in from elsewhere.

They hate a voicemail and a busy signal more. This is a real conversation, not press-1-for-menu: callers speak naturally and the agent answers in the moment. In the Prishtina restaurant we built, guests booked without complaints and confirmed bookings went up 30%. If someone insists on a person, the agent routes them to your staff.

We do not pretend the agent handles everything. We agree up front on what goes to a human: large parties, private events, allergy questions, and complaints, and those calls reach your staff with the details already gathered. The honest tradeoff: the agent handles the high-volume, repetitive calls well, and we deliberately hand the judgment calls to your team.

That is the whole point. We wire the agent directly into the booking system you already run, so a confirmed call becomes a confirmed reservation with no re-typing. During scoping we check that your system lets us read availability and write bookings. If it does not, we tell you before we start, not after.

No to both. The assistant runs on your existing number and needs no new staff. In the Prishtina restaurant it lightened the load on the front-of-house team, who now look after the guests in front of them during the dinner rush.

General-purpose assistants are made for quiet conversations and struggle with interruptions, accents and a full dining room in the background. We build on a platform made for phone calls and tune it on your real calls, so it sounds like your restaurant.

It depends on how clean your booking integration is and how many call types we cover at launch. We usually start narrow, ship your highest-volume call reasons first, then expand. After go-live we run a tuning pass on real calls, because diners always find phrasing the test scripts missed.

That is why every image is approved by a person. We tune the system against your best existing photos so new images match your lighting, framing and styling. Some will miss, and your team simply rejects those.

No. It takes the repetitive, high-volume work off them: the tenth background, the seasonal variation, the extra angle needed at the last minute. Hero shoots and the creative direction behind your look stay with people.

This is the honest limit. Image generation is strong on scenes, backgrounds and lighting, and weaker on exact packaging text, precise shades and fine texture. For those products we start from real photography and restyle around it, and where color must be exact, the product stays on a real shoot.

Full control. Nothing reaches your library until a person on your team approves it, and you can approve in batches or send images back. You also set the brand guidelines the system follows.

We only use image tools whose commercial terms we have checked, and we avoid any with unclear licensing. The products and brand assets going in are yours. For high-stakes campaigns we recommend your own lawyer reviews the final terms, and we document exactly which tools are used to make that easy.

Your product catalogue, your brand guidelines and a set of images that show the look you want. The first version runs on part of your range so you can judge the quality before we extend it. Clean, well-organised product information is what speeds things up most.

Possibly. We can set n8n up for you and get your first workflow running. We will also tell you when a job does not need it, because some tasks are simpler as a small piece of code.

We write code for that step and connect it to the rest of the workflow. n8n is excellent for connecting tools, and less suited to heavy data processing or steps where an automatic retry could cause damage. We point those out before building, so you always know which parts are which.

Each AI step returns its results in a fixed format that is checked automatically, so a bad answer is caught early. Anything a customer will see is reviewed by a person first, so the mistakes AI still makes stay away from your customers.

There is no lock-in. Everything runs in your own n8n account, documented, and we walk your team through it. Most of it can be followed visually, and the few custom code steps need someone comfortable reading code.

We scope a fixed first build so you see something running before you commit to more. How long it takes depends on how many systems the workflow touches and how clean your data is. The usual thing that slows a build down is messy or undocumented source systems, so we look at those early and tell you what we find.

Most AI sales tools focus on writing and sending more emails. Ours focuses on the decisions: who is worth contacting, why now and what each reply means. It works alongside the tools you already have.

A message is only written once the system has confirmed the company is a good fit and found a real reason to get in touch this week. Every email has a specific purpose, which is what makes it worth reading.

The system scores companies, finds the reason to reach out, drafts the message and sorts the replies. People approve the messaging before anything goes out at scale, hold the sales conversations and handle anything the system is unsure about. Closing the deal stays with your team.

A well-written email to the wrong company is still wasted, and it can hurt your email reputation. The real gains come from contacting the right companies and reading replies correctly, so your team spends its time where it matters.

Niche markets are where scoring helps most, because job titles alone return mostly noise. We agree your criteria together at the start. In one live engagement the system runs 12 campaigns at once across 14 industries, each aimed at a different type of customer.

Usually not. We work with the tools you already use, including HubSpot and Salesforce, and add the decision-making on top. If one of your tools gets in the way, we will tell you and explain why.