Most companies start with Claude the same way: a few people create accounts and use them to write emails and summaries, and a few weeks later everyone uses it differently and nobody knows what the company is getting out of it. Implementing Claude in your company starts when Claude stops being each employee's private tool and starts working inside the company's processes, on the company's knowledge and by the company's rules. In this post we show what changes in your team's work at that point and where to start, so the rollout doesn't end with buying accounts.
A Claude implementation is not buying accounts
Team plan accounts are a prerequisite, but they are not an implementation. You can tell the difference by where knowledge lives and who remembers how things are done.
| Claude accounts without an implementation | Implementing Claude in your company | |
|---|---|---|
| Client knowledge | everyone pastes the context again in every conversation | one company knowledge base in files, used by both people and agents |
| Way of working | everyone has their own prompts, often in a notepad | processes described step by step as skills, the same for the whole team |
| Corrections | you fix the same mistake every week | a correction is saved as a rule and doesn't come back |
| Control | nobody knows what went to the client or on what basis | a person approves every step that can't be undone |
| Result | "AI helps with something" | you know which processes the system has taken over and how often its output goes through without corrections |
So the implementation is about building around Claude what the model doesn't have on its own: the company's memory, documented processes, rules, and points where a person approves the output. We call that setup an agentic system. We explain what such a system is and what it isn't in a separate post.
What changes in your team's work
It's easiest to see through three people who have the least time in a small B2B service company.
The owner. Instead of going through the inbox, calendar and notes, they start the day with a review prepared by an agent: what is waiting for a reply, which meetings are scheduled, which tasks are due. Instead of writing proposals, they read drafts and decide on scope and price.
The salesperson. Before a meeting, they get a note on the company and the person: what we already know, what we agreed earlier, what is worth asking. After the meeting, they hand over their notes or a transcript and get back the agreements, the tasks and a draft follow-up. They read it, fix it and send it. They still run the conversations and own the relationship, but they no longer spend evenings on prep work.
A new team member. From day one, they have access to the same knowledge base as the agents: how the company writes proposals, what it promised to which clients, how the process works. They don't have to guess or ask colleagues about everything.
One thing doesn't change: nothing goes out to a client without a person. The agent prepares, the person decides and sends.
What your company needs to get started
Before you start, check four things.
- A Claude Team plan for everyone who will work in the system. Both the agents and the team work on it. Current prices are on the Claude pricing page.
- One process to begin with. Not "we're rolling out AI across the company" but "an agent prepares us for meetings with new clients". A good first process repeats every week, takes time from the person whose time is most expensive, and has an output that is easy to judge.
- Knowledge that can be gathered. Price list, proposal templates, service descriptions, client notes. It doesn't have to be organized: organizing it is part of the implementation. It just has to exist outside people's heads, or be possible to get out of them in a meeting.
- A person who will develop the system. The whole team uses the system, but someone has to develop it: add rules, add processes, keep an eye on quality. That takes dedicated time on the company's side. You can also hand this part over to us.
Claude implementation step by step
1. Meeting and choosing the first process
About an hour with the team. We go through the company's processes and choose the one we'll start with. At this stage the key question is: where do people lose the most time on work that is neither a conversation with a client nor a decision?
2. Knowledge base
We gather the knowledge the first process needs and arrange it in files so the agent reads it in a fixed order: state first, then the map, details last. This matters for more than quality. An agent that loads only what it needs uses less of the subscription limits, so the team can get more done with it during the day.
3. Skills, rules and gates
We describe the process step by step as a skill: where the agent gets its data, what it checks, what the output looks like. We add rules, meaning the things the company never does or always does. Wherever the output leaves the company, we put a gate: a step where a person or a script checks the output before it moves on. Anything a script can check is checked by a script, because it is cheaper than the model and gives the same result every time.
4. Team training
The team learns two things: how to assign work to agents and how to correct their output so that the correction turns into a rule. The second skill decides whether the system fits the company better and better over time or stays at the level of its first week.
5. The first weeks and measurement
We measure one number: what share of the agent's proposals a person approves without corrections. It goes up when corrections turn into rules. It stays flat when people fix things quietly and record nothing. This is a way of measuring, not a promise of a specific result: in every company this number starts from a different place.
6. More processes
Once the first process works, we add more on the same knowledge base: proposals, follow-ups, the daily review, onboarding new people. Each one is easier, because the knowledge base and the rules already exist. Integrations with the company's tools, such as a CRM, are added when a process really needs them.
The most common mistakes when implementing Claude
- Everyone works alone in their own chat. Knowledge stays in conversations nobody but the author can see, and the company pays every time for pasting the same context.
- Too many processes at once. Five processes half implemented give you less than one implemented fully, because nobody knows which output can be trusted.
- No approval before sending. An automation that sends on a person's behalf only has to get it wrong once, and the team stops trusting it.
- Corrections without a record. If a correction doesn't become a rule, the same mistake comes back next week.
- Nobody develops the system. A system without an owner stops matching how the company works after a few weeks.
What about company data?
The agents work on the company's files and on the company's Claude account, not in a new external dashboard. Before the implementation, it's worth agreeing which data goes into the knowledge base, who has access to it, and which data processing terms apply on the chosen Claude plan. The plan terms are described by Anthropic. This post is not legal advice: the company decides on the scope of data, and we help translate that decision into configuration.
Frequently asked questions
Where do you start implementing Claude in your company?
With one process that repeats every week and takes time from the most expensive person. In service companies, this is most often preparing for meetings or follow-ups after meetings.
Do you need a developer to implement Claude?
Not to use the system. The team works in Claude in plain language, including by voice. Building and developing the system takes a person who will dedicate time to it. That can be someone in the company or us.
Will Claude connect to our CRM?
Claude connects to the company's tools through connectors. Whether and how to connect a specific CRM is decided once we understand the process, and integrations are priced separately.
How much does implementing Claude in your company cost?
An agentic system implementation starts from PLN 9,000 net, depending on scope. On the company's side, there is a Claude Team subscription for each person working in the system. Details are on the agentic system page, and we break down what that price is made of in the post on AI agent implementation cost.
Summary
Implementing Claude in your company starts where buying accounts ends: with a shared knowledge base, documented processes, rules built from corrections, and a person's approval before every step that can't be undone. It's best to start with one process and add more once the first one works.
Want to know which process to start with in your company? Book a meeting at a time that suits you. It takes about an hour and ends with choosing the first process. Prefer to write first? hello@200iqlabs.io.