Rolling Out Langdock: The Practical Guide for Companies

Rolling out Langdock means giving your team a centrally managed, GDPR-compliant access point to AI – with clear roles, a signed data processing agreement (DPA), training built around real tasks, the first custom AI agents, and support until usage sticks in daily work. This guide walks through the rollout in six steps, with realistic costs and the classic mistakes.

Some of your people are already using AI. Privately, on personal accounts, without IT ever seeing it. The question is no longer whether AI enters your company – it's whether it comes in under control or through the back door. According to Salesforce's 2026 AI index for German SMEs (KI-Index Mittelstand), 51.2% of small and mid-sized businesses in Germany are already using or testing AI. Roll out a clean platform now, and you pull the topic out of shadow IT – the tools people bring in without IT knowing – and make it productive.

Quick context first: Langdock is an AI platform from Berlin. It gives your entire team centrally managed access to AI models from several providers in one interface – GDPR-compliant, ISO 27001-certified (the international standard for information security), with hosting in the EU. Langdock names Merck and Personio among its references. What the platform can do in general is explained in What is Langdock?; what we do around rollout, training and ongoing support is on our Langdock page. This article is about the practice: the rollout.

Transparency: This is a referral (affiliate) link. If you use it, we may earn a commission – the price you pay stays the same.

Step 1: Define roles and permissions – before the first login

The most common opening mistake: give everyone access immediately and sort out the rest "later". Later never comes. Nail down three things before the first login:

• Who owns the platform? One person – plus a deputy – takes care of users, permissions, and settings. Without this role, the system manages itself. Badly.
• Who may do what? Not every team needs the same access. Sales, accounting and management work with data of different sensitivity – the permissions should reflect that. Better to start with a small group and expand than to claw permissions back afterwards.
• What's allowed? A short written rule is enough to start: which data may go into the platform, which may not? No legalese – one document everyone understands instantly and everyone knows exists.

Step 2: What do you have to sort out on data protection before real data flows?

Langdock covers the foundations: GDPR compliance, ISO 27001 certification, EU hosting, and your data isn't used to train AI models. The DPA lists the sub-processors – the AI providers and hosting companies involved – and where they sit. Read that list. Because "the platform is compliant" doesn't automatically mean "your use of it is too". Two things belong on your list before real customer data flows:

• Sign the DPA. The data processing agreement governs how Langdock handles data on your behalf – mandatory under GDPR as soon as personal data is involved. It's part of Langdock's standard, but it still has to be signed. Before the start, not after.
• Bring in your data protection officer. If your company has one, they belong at the table early, not after the rollout. A short conversation up front saves long discussions afterwards.

Article 28 GDPR sets out what a DPA has to contain. It pays to read these points once instead of just signing:

• Subject, duration and purpose of the processing – what Langdock may process your data for at all.
• Type of data and categories of data subjects – customer data, employee data, prospects.
• Processing on instructions only – Langdock processes on your instruction, not for its own purposes.
• Technical and organizational measures – encryption, access control, logging; this is where the ISO 27001 certification pays off.
• Sub-processors – which service providers Langdock itself uses, for hosting and the AI models for instance, and where they're located.
• Deletion and return of the data at the end of the contract.

Important: this is not legal advice. Read the DPA with your data protection officer or your lawyer – especially the list of sub-processors, because that list decides which models process which of your data, and where.

For how you communicate this to the team: data protection isn't the brake here – it's the argument. It's exactly why you're introducing a platform instead of tolerating private accounts.

Step 3: Start small – one team, one use case

One team, one use case: the rollout starts with a single step – and grows from there.

The reflex to roll Langdock out to the whole company at once is understandable – and almost always a mistake. Big chunks fail; small wins carry. The better route: start with one department and one concrete use case.

Good first use cases share three traits: they happen often, they follow a recognizable pattern, and their output is easy to check. Typical candidates:

• Drafting emails and proposals – the daily writing work everyone knows and nobody loves.
• Summarizing documents – contracts, meeting notes, long email threads boiled down to the point.
• Preparing research – collecting and structuring information before a human decides.

And then: measure. Estimate roughly how many hours the use case costs per week today – that's your baseline – and compare after the pilot phase. Then you have numbers instead of opinions – and a story that convinces the rest of the company without you having to evangelize. For the pilot, the hours number is the right comparison. Whether the platform lands in daily work is measured later by two other numbers – see Step 6.

Step 4: Train the team – this is where the rollout is decided

The classic software training – slides, feature tour, Q&A – doesn't work for AI tools. The reason: the hurdle isn't the interface. Langdock runs in the browser and largely explains itself. The hurdle is the question of what to actually use the thing for in your own workday.

A good Langdock training therefore starts from real tasks: everyone brings an actual task from their week – an email, a proposal, a report – and completes it with Langdock during the session. The difference is enormous: after a slide training, people know what the platform can do. After a hands-on training, they've already worked with it – and come back the next day on their own.

Two more things that make the difference:

• Build champions. Every team has people who are curious on their own. Give them early access and a stage – they'll convince the rest more credibly than any announcement from above.
• Plan for follow-up. The real questions only surface after the training, in the middle of the work. A short recurring session or an internal channel for questions catches them.

And whoever asks "Will this replace me?" deserves a straight answer: not this platform, and not on its own. It takes over the grunt work – judgment, customer relationships and responsibility stay with the humans.

Step 5: Build the first agents

Chat is the entry point – the real level above it is your own AI agents: fixed helpers for a recurring task that you describe properly once, instead of explaining it from scratch with every request. An agent knows its job, its tone and its limits – your team calls it up instead of retyping the same context over and over.

The good news: agents can be created in Langdock without code – the real work isn't the technology, it's describing the task precisely. And if an agent needs to reach deeper into your systems than the interface allows: Langdock has an API – the technical interface through which other programs talk to the platform. Whether and how your existing systems can be connected through it is something we check case by case; we build those connections as custom software.

Start small here too: one agent for one task that costs the pilot team hours every week. Once it runs reliably, the next one follows. What agents can realistically take over today – and what they can't – is covered in detail on our AI agents page.

Step 6: Stay on it – AI adoption takes months, not a meeting

Go-live isn't the goal, it's the starting line. Whether an AI rollout succeeds is decided in the months that follow – by whether your team actually uses the platform, not by whether it's set up. That's why we stay on board after the kick-off sprint for several months: with workshops on new use cases, fixed check-ins for the questions that only surface in daily work, and the sharpening rounds every agent needs.

Success isn't measured in licenses but in real usage – two numbers we track together:

• Weekly active users. How many employees actually work with the platform every week – not how many have a login.
• Agent share vs. chat. How much of the usage runs through the agents you built rather than the open chat. When that share grows, AI has moved from playground to process.

As long as both numbers rise, the rollout is working. If they stall, that's not a reason to panic but the pointer to where the next training session or the next agent should start. How long this phase takes depends on the business – we deliberately don't name a number, we name the metric.

What does rolling out Langdock realistically cost?

Four items you should look at separately – because they go to different places and run for different lengths of time:

Item	Goes to	What it costs	When
License	Langdock	See Langdock's current plans – we don't print a number that ages	Ongoing
Kick-off sprint	Us (SCALE² sprint)	Fixed price, most sprints between €2,500 and €13,000 depending on scope, three weeks of support included	One-off
Ongoing support	Us	Monthly fee, agreed up front – depends on team size and workshop rhythm	Ongoing, over several months
Internal time	Your team	The platform owner's hours plus training and onboarding per employee	From day one, not just the first weeks

• The kick-off sprint. Setup, permissions, hands-on training and the first agents run with us as a SCALE² sprint – three weeks, five phases, fixed price, usually between €2,500 and €13,000 depending on scope: number of teams, number and depth of agents, whether existing systems get connected. The DPA itself you sign with Langdock and review with your own lawyer or data protection officer – that part isn't ours; we only make sure it is on the checklist before go-live. Three weeks of post-go-live support are included.
• The ongoing support. After the sprint, the support from Step 6 simply continues. Billed as a monthly fee agreed up front; the amount depends on team size and rhythm. Measured by weekly active users and the share of agent usage. How both phases run is on our Langdock page.
• The internal time. The honestly underestimated item: the platform owner needs regular time throughout the rollout, and every employee needs training plus onboarding. How much exactly is what we work out in the analysis phase – if you don't plan this time at all, you get a rollout that's finished on paper and never happens in daily work.

Against that stands the calculation that counts. A worked example, not a promise: if a single well-built agent gives a team back five hours a week, that's roughly 250 hours over a working year (about 50 weeks). What's realistic for you only shows in the baseline from the pilot. Whether the rollout pays off is therefore decided not by the license, but by whether the first use case is well chosen.

The classic rollout mistakes

For the record, straight from practice:

1. Buying licenses, forgetting the rollout. Access alone changes nothing. Without training and a first use case, the platform stays an icon on the desktop.
2. Everyone at once. A company-wide rollout without a pilot phase produces many half-convinced users instead of a few enthusiastic ones. The enthusiasts convince the rest – the half-convinced convince no one.
3. No owner. Without a platform owner and an internal point of contact, the platform fizzles out. It's not a full-time job, but it is a real role.
4. Postponing data protection. The DPA has to be signed before the first real customer data goes into the platform. Sign it late and you'll have spent that time working with a processor without the written agreement Article 28 GDPR requires – an avoidable risk that can cost you the whole rollout. (Again: not legal advice – that call belongs to your data protection officer or your lawyer.)
5. Training as a checkbox. Slides produce attendees, not users. Training has to start from people's actual tasks, or it stays theory.
6. No baseline. Without a baseline, you can't prove success afterwards. Then the rollout becomes a matter of faith. With a number, it's arithmetic.
7. Letting go after go-live. The platform is set up, the training is done – and a few months later hardly anyone uses it. AI adoption isn't an event, it's a process: without ongoing support, new use cases and a regular look at the usage numbers, even the best rollout quietly dies.

Our experience from our own rollout

We don't just walk this path with clients: we run more than 40 of our own AI agents in Langdock – among them our bookkeeping agent, with which our bookkeeping runs around 90% automated; the others range from research to copywriting. On top of that come further agents and AI workflows elsewhere in the business. We've walked these steps ourselves. Three things that surprised us:

The first agent rarely lands on the first try. Most of our agents needed more than one round of sharpening before they ran reliably: a more precise task description, better knowledge sources, clearer limits. That's not a flaw of the platform; it's normal. If you know that in advance, you plan the rounds in instead of giving up disappointed after the first one.

Describing the task is the real work. With our bookkeeping agent, the effort sat in describing the task – which cases it handles on its own and when it hands back to a human. Take that description seriously and you get an agent the team trusts; cut it short and you get one the team works around.

The second agent went faster than the first. In our own build-out, the second agent took noticeably less time than the first – the learning sits in the team, not in the platform: by then, your people know how to describe a task so that an agent understands it. That's why it pays to build the first agent carefully – it's the school for all the ones that follow.

How do you know the rollout is working?

Not by the number of licenses. But by your team moving tasks into the platform on its own. By meeting-room sentences like "I had the assistant prep this." By the first department coming to you with its own agent idea. In numbers: weekly active users go up, and a growing share of the usage runs through agents instead of chat. From that point on, the topic carries itself – and you can tackle the bigger processes.

Frequently Asked Questions

Is Langdock GDPR-compliant?

Yes – for the platform: it's ISO 27001-certified, hosts in the EU, and doesn't use your data to train AI models. A data processing agreement (DPA) is part of the standard. The second half matters: that makes the platform compliant, not automatically your use of it. For that you need the signed DPA, defined permissions, and a written data rule. This isn't legal advice – the assessment of your case belongs to your data protection officer.

What's in Langdock's DPA?

The contents are prescribed by Article 28 GDPR – Langdock supplies the document as part of its standard. The work is on your side: check the sub-processor list, because it names which AI providers and hosting companies touch your data and where they sit, and check that the data types match what your teams will put into the platform. Sign it before the first real customer data flows. Not legal advice – read it with your data protection officer.

Does my team need technical skills?

No. Langdock runs in the browser and works like a chat – anyone who can write an email can work with it. The difference between dabbling and productive use isn't prior knowledge; it's training built around real tasks from people's own workday. What you do need is one person who owns the platform internally – not a technician, but someone who maintains users and permissions and answers questions.

How long does a Langdock rollout take?

In two phases. The kick-off sprint takes three weeks in five phases – analysis, concept, development, testing, go-live – followed by three weeks of support. Then ongoing support begins, over several months: workshops, check-ins for everyday questions, agent sharpening. How long it runs depends on the business; the metric stays the same: weekly active users and the share of agent usage. What you can plan reliably is the sprint, not the day it clicks for everyone.

What if the team doesn't use the platform after the rollout?

In the cases we've seen, it usually comes down to one of the classic rollout mistakes: no owner, no hands-on training, no clear use case or no baseline. The good news: all of them can be fixed after the fact – most effectively with training built around real tasks, one clearly defined pilot use case, and ongoing support that keeps an eye on usage for months instead of leaving the team alone after go-live.

The next step

You can implement this guide on your own – it works. Or you walk the path with people who know these pitfalls from their own rollouts. We set up Langdock, train your team on its real tasks and build the first agents – as a SCALE² kick-off sprint: 3 weeks, 5 phases, fixed price, with 3 weeks of post-go-live support included. After that we stay on board for several months until usage sticks in daily work. Details on the Langdock page.

The simplest first step remains the Time Potential Analysis: 45 minutes, free, no sales pitch. Together we look at where the most time gets lost in your business – and whether Langdock is the right lever for it. Book your Time Potential Analysis →