--- title: "AI Integration Services for Your CRM and ERP | JustAutomate" description: "AI integration services that wire Claude or GPT models into your CRM, ERP and documents, with a permission scope per system and a pilot on real data." url: "https://justautomate.ai/en/services/ai-integration/" lang: en --- AI agents and chatbots # AI integration services that put a language model inside the systems you already run Our AI integration services connect a language model to your CRM, ERP, invoicing, shared drive and data warehouse, so it answers and acts on current company data instead of on whatever someone pastes into a chat window. We decide with you which systems it may read, where it may write, and how every call is logged. The result is an AI agent that works inside your processes, with a pilot on your own data before anyone relies on it. Book a free consultation [See how the workshop works](https://justautomate.ai/en/services/process-workshop/) - 5 to 10 secondsfor an answer to a plain-English question about data held in the connected systems - 4 to 6 weeksfrom the end of the workshop to an AI agent in production, piloted on your own records - EUR 3,500 netper week of the process workshop, where we pick the use case, the systems and the permission scope The problem ## Why does a ChatGPT licence not change how the company works? Plenty of companies already pay for a language model. People draft emails with it, summarise meeting notes, rewrite a proposal. Then someone asks it which invoices from last month are still unpaid, or what stage the three largest deals are at, and the model has nothing to go on. The answer lives in the CRM and the accounting system, and the model cannot see either of them. So the team does the integration by hand. Someone exports a spreadsheet, pastes it into the chat, checks that the figures came through, then copies the answer back into an email. The model saves a few minutes on wording and costs them again on moving data around. It also means company data travels through personal accounts and browser tabs, where nobody can say afterwards who sent what. The rules around this are not optional either. [Regulation (EU) 2024/1689, the AI Act](https://eur-lex.europa.eu/eli/reg/2024/1689/oj), sorts uses of AI by risk and ties obligations to the use, not to the model, and UK GDPR still decides what personal data may go where. An AI integration is the point at which you can actually decide that, per system and per field, rather than hoping people paste the right things. - A model the whole team pays for that never sees a single company record - Exports and copy-paste between the CRM, a spreadsheet and a chat window, several times a day - Client data passing through personal accounts with no log of what went where - An answer that sounds certain but was built from a stale export Why now ## What changed, so that connecting AI to company systems takes weeks? Until recently every link between a model and a business system was a bespoke project. In November 2024 Anthropic published [Model Context Protocol](https://modelcontextprotocol.io/), an open standard for giving models access to data and tools. Each system becomes a separate module with its own permission scope, so adding the invoicing system next month does not mean rebuilding what already connects the CRM. The second change is the systems themselves. HubSpot, Salesforce, Pipedrive, Xero, Shopify, Google Workspace, Microsoft 365 and BigQuery all expose documented APIs. Where a system has no ready connector, we write one, the same way we do in our [API integration services](https://justautomate.ai/en/services/api-integration/), and the model reaches it through that integration. The third is cost. A single query costs the model provider roughly EUR 0.005 to 0.07, which you pay directly to them, so the expensive part of AI integration is no longer the model. It is choosing the right process, the right data and the right limits, and that is where the workshop goes. What changes ## What does an AI integration give you that a chat window does not? - ### Answers built from live data, with the source named The model reads the CRM, the invoicing system or the warehouse at the moment of the question and says which system each number came from. Where data is missing, it says so instead of filling the gap. Every answer points to the system it was assembled from - ### Search across your own documents Contracts, offers, procedures and product sheets are indexed so the model retrieves the relevant passage before it answers. This is retrieval-augmented generation (RAG), and it is what stops the model inventing policy. Answers quote the document and section they rely on - ### Actions in your systems, within agreed limits Adding a note to a contact, drafting an invoice for approval, moving a deal, posting a summary to a team channel. Each system gets read access or read and write, decided per integration, never globally. Write access granted system by system, and revocable in one place - ### AI steps inside automations you already run A workflow in n8n, Make or Zapier gains a step that classifies an incoming email, extracts fields from a PDF or drafts a reply for a person to approve. The rest of the workflow stays as it is. One model step added to an existing workflow, not a new platform - ### A log of every query and every call Who asked, what the model read, what it changed. The permissions of the agent follow the permissions of the person asking, so nobody sees through the model what they cannot see in the CRM. Full call log available for audit, data not used to train models - ### The model can be swapped without rebuilding The integrations sit between your systems and the model, not inside the model. We build on Claude and GPT models, and when a better or cheaper model appears, the connections and permissions stay where they are. Model choice kept in configuration, separate from the integrations Proof ## Have you integrated AI into your own systems? Yes, and that is the proof we can publish without asking a client for permission. The AI agent inside our founders' agency reaches its data through the same kind of integrations described on this page. Below is what passes through them, and what the number does NOT show. Built for ourselves. These are not client deployments but implementations inside the founders' own companies. We say so plainly, because proof you cannot verify is not proof. - ![JustIdea Agency](https://justautomate.ai/klienci/justidea.png)![JustBot](https://justautomate.ai/agenci/justbot-80.webp)JustBottheir agent JustIdea Agency, a performance marketing agency of around 30 people 23 people from the company work with the agent 38 systems wired in permanently 112 processes run without a human 475 thousand lines of the agent’s code A language model connected to the CRM, the task system, Google Ads, Meta, Analytics, Search Console, the SEO tool and the data warehouse. Work started on 1 March 2026, and 23 people use the agent today. 8 of 38 systems it works in - Google Ads - Meta Ads - HubSpot - ActiveCollab - Ahrefs - GA4 - Search Console - BigQuery Before Answering a question about a client meant opening several panels, pulling figures into a spreadsheet and asking whoever knew where to look. Cross-system summaries were rare, because each one was a manual job. After People ask the agent in a sentence and it reads the connected systems itself. A large share of the work now runs on schedules and events, and people step in where a decision is needed. 64,903 operations through the connected systems in 30 days, 29,854 of them with no human involved Operations run by the agent, month by month Monthly operations run by the agency agent in 2026, from 83 in March to 62,985 in August, while the number of people using it grew from 4 to 23\. A rollout that nobody uses would show a flat line here. Measurement: Agent call log in the data warehouse, March to August 2026, counted per tool call. And in your company ## Will this work with our systems and our data rules? These are the four doubts we hear most often before an AI integration, and how we deal with each. - We cannot send client data to an AI provider. Then we decide field by field what reaches the model and what does not, and we write that into the scope before the pilot. Company data is not used to train models, every call is logged, and personal data handling goes into the processing agreement under UK GDPR or the GDPR. Some questions can be answered from aggregated figures, with no personal data involved at all. - Our data is a mess. That is the usual starting point. The workshop picks one use case where the data is good enough, and the pilot shows quickly where it is not. Often the first useful result of an AI integration is a list of fields nobody fills in. - The model will make things up. A model with no access to your data has to guess. Once it reads the source system or the document at the moment of the question, it answers from that, names the source and says when data is missing. We test this on your own records in the pilot, before anyone relies on it. - We have no in-house developers. You do not need them. We build and run the integrations, and hand over documentation of every connection, permission and prompt. If you later hire a developer or switch suppliers, they start from that document. Language models Claude and GPT models, with the choice kept in configuration CRM HubSpot, Salesforce, Pipedrive, Zoho Finance and ERP Xero, sevDesk, on-premise ERP systems Documents and team tools Google Workspace, Microsoft 365, Slack, Notion Data BigQuery, Google Sheets, GA4, SQL databases Automation layer n8n, Make, Zapier, custom API integrations ## Four ways to integrate AI into company systems Most companies need one or two of these, not all four. The workshop decides which, based on the questions your team actually asks and where the answers are kept. | Approach | What the model does | When it fits | What it needs | | --------------------------- | ---------------------------------------------------------------- | ------------------------------------------------------- | ------------------------------------------------------------------- | | Chat window, no integration | answers from general knowledge and whatever is pasted in | drafting, summarising, rewording | nothing beyond a licence, and no company data reaches it on its own | | Document search (RAG) | finds the relevant passage in your documents and answers from it | contracts, procedures, offers, product knowledge | an index of the documents and rules on who may see which folder | | AI agent with system access | reads and, where allowed, writes in the CRM, ERP or warehouse | questions that span several systems, routine operations | an integration per system, a permission scope and a call log | | AI step in an automation | classifies, extracts or drafts inside an existing workflow | incoming email, PDF invoices, form submissions | a workflow in n8n, Make or Zapier and a human approval step | The approaches combine. A common first step is an AI step inside one workflow, followed by an agent with read access once the integrations are in place. How we work ## How an AI integration project runs, from use case to production Each stage ends with something you can assess before you commit to the next. 1. ### Workshop: use case and data map We list the questions and tasks that cost your team the most time, then check which systems hold the answers, how good that data is and what may not leave a given system. The output is one use case to start with and a written permission scope. 2. ### Connections and permissions We connect the chosen systems, through ready connectors where they exist and custom integrations where they do not. Each one gets read or read and write access, and the agent inherits the permissions of the person asking. 3. ### Pilot with 5 to 10 people A small group from different teams uses it on real work. We read the call log for wrong sources, missing data and questions nobody expected, and fix them before the wider rollout. 4. ### Production with logging The integration goes to the whole team with monitoring of failed calls and a full query log. The old way of getting the answer stays available until people stop using it. 5. ### Review at 30 and 90 days We look at what people actually ask, which systems the agent reads most and where it falls short. That decides the next system to connect, or whether to stop there. Before and after ## How a working week looks once AI is integrated ### Before - Export from the CRM, paste into a chat window, copy the answer into an email - Questions spanning two systems wait for the one person who knows both - Client data passes through personal accounts with no record - Nobody can say whether the AI licence changed anything ### After - A question in a sentence, an answer from current data with its source named - Anyone on the team can ask, within their own permissions - Every query and every change sits in one log - Usage and gaps are visible in the log, so the next step is decided on numbers The pilot group uses it in the first weeks after the workshop. Whether it has become part of daily work shows after 30 days, whether it is worth extending after 90. For the person signing this off ## What an operations or finance director gets out of it You get a clear answer to the question the board keeps asking: what is the company doing with AI, and is it paying off. The call log shows how many people use it, for what, and how often, so the answer is a number rather than an impression. You also get control over a risk that already exists. Staff are pasting company data into AI tools whether there is a policy or not. An integration moves that into one place with defined access, a log and a processing agreement, which is a far easier position to defend in an audit. - Permission scope written down per system, revocable in one place - A full log of queries and changes, available for audit - A pilot on real data before the whole team relies on it - A review at 30 and 90 days, so extending is decided on usage figures What it costs ## The workshop has a published price, the integration is quoted after it The process workshop costs **EUR 3,500 net per week** and runs from one to seven weeks. For a first AI integration one or two weeks is usually enough. We quote the integration after the workshop, because the cost depends on things nobody knows before it: how many systems are involved, whether they have documented APIs, and how much of the data is usable as it stands. Model usage and any automation platform you pay directly to the provider, with no margin from us. At roughly EUR 0.005 to 0.07 per query, model fees are rarely the deciding cost of an AI integration. - ### Process workshop EUR 3,500 net per week, one or two weeks for a first AI integration A self-contained piece of work. It ends with a document your own team or another supplier could build from. - The questions and tasks that cost the most time, ranked - A map of systems and data that hold the answers, with data quality noted - Permission scope per system and the fields that must not reach the model - One use case to start with and the metrics to check at 30 days Start with the workshop - Most popular ### Workshop and AI integration Quoted after the workshop scope follows from the number of systems and the state of their data The option most companies choose. An AI agent in production 4 to 6 weeks after the workshop, piloted on your own records. - Everything in the workshop - Connections to the chosen systems, ready or custom - Document search where the use case needs it - Pilot with 5 to 10 people, then production with logging - Documentation of connections, permissions and prompts Talk about an AI integration - ### AI integration and BOND Quoted after the workshop for companies that want one agent across all their systems Everything in the integration, with BOND, our business AI agent, as the single place the whole team asks about company data. - Everything in the AI integration - BOND connected to each system as a separate module - Answers in 5 to 10 seconds, in English or Polish - Further systems added after the 30 and 90 day review Ask about BOND The workshop ends with a document that stands up without us: the use case, the data map and the permission scope. If it shows that a ready tool already covers your case, or that the data is not good enough yet, we say so plainly and the engagement ends there. ## Frequently asked questions What are AI integration services? AI integration services connect a language model to the systems a company already uses, such as the CRM, ERP, invoicing, document storage or a data warehouse, so the model can answer and act on current company data. The work covers the connections themselves, the permission scope for each system, document search where it is needed, logging of every call and a pilot on real data. It differs from buying an AI licence in that the model gets controlled access to your data instead of relying on what people paste in. What are examples of AI integration? An AI agent that answers which invoices are overdue by reading the accounting system. Document search that finds the relevant clause in a contract. A step in a workflow that reads incoming emails and routes them to the right person. A weekly summary assembled from the CRM and the ad platforms without anyone exporting a spreadsheet. Each of these needs a different integration, which is why the workshop starts from the task, not the tool. Which AI model is best for integration? The one that handles your language, your data volume and your security requirements at an acceptable cost per query. We build on Claude and GPT models and keep the choice in configuration, so the integrations do not depend on one vendor. For most business use cases the quality of the data and the permissions matter more than which of the leading models sits behind them. How much do AI integration services cost? The only fixed price is the process workshop: EUR 3,500 net per week, and one or two weeks is usually enough for a first AI integration. The integration is quoted after the workshop, because the cost depends on the number of systems, the state of their APIs and the quality of the data. Model and platform fees you pay directly to the provider, with no margin from us. How long does an AI integration take? One or two weeks of workshop, then 4 to 6 weeks to an AI agent in production, including a pilot with 5 to 10 people. The timeline slips mainly when a system has no documented API or when access from the vendor takes weeks to arrive. We check both in the workshop, before giving you a date. Is our data used to train the model? No. Company data is not used to train models, and we agree before the pilot which fields may reach the model at all. Every query and every change is logged, and the agent can only see what the person asking is allowed to see in the source system. How is this different from your API integration and AI consulting services? API integration connects business systems to each other so data moves without retyping. AI consulting decides which processes are worth automating and in what order. AI integration is the step that puts a language model on top of those connections, so people can ask questions and delegate tasks against live company data. ## Find out which of your systems AI should reach first The free consultation takes 30 minutes. Tell us which questions your team spends time answering and which systems hold the data. We will tell you whether an AI integration makes sense, which use case to start with, and what we would leave alone. You do not need to share any data at this stage. We run at most two rollouts at a time, so that someone is genuinely sitting on each of them. [Book a free consultation](https://justautomate.ai/en/contact/) **PS.** If your team pays for an AI model and still exports spreadsheets to feed it, that is the gap an AI integration closes, and the workshop document stays with you whoever builds it. ## Related - [API integration services](https://justautomate.ai/en/services/api-integration/) - [BOND, business AI agent](https://justautomate.ai/en/services/bond-business-ai-agent/) - [AI consulting for companies](https://justautomate.ai/en/services/ai-consulting/) - [AI data analysis](https://justautomate.ai/en/services/ai-data-analysis/) - [Model Context Protocol explained](https://justautomate.ai/en/guides/mcp-model-context-protocol/) - [Process workshop](https://justautomate.ai/en/services/process-workshop/) --- **JustAutomate** · contact: contact@justautomate.ai · form: https://justautomate.ai/en/contact/