AI Integration with ERP and CRM (SAP, Comarch, Dynamics): How It Works in Practice (2027)

Integrating AI with ERP and CRM (SAP, Comarch, Dynamics) in practice means choosing one of three models: a vendor-embedded assistant, your own AI layer connected via API, or AI agents operating on systems via MCP. This text is for mid-sized and large companies in Poland that want to connect language models to an existing ERP or CRM, including legacy systems. You’ll find a comparison of vendor tools (SAP Joule, Microsoft Dynamics 365, Comarch ERP, Salesforce Agentforce, Oracle NetSuite), implementation partners, the rollout process, risks, and case studies with numbers.

Key takeaways

• Three integration models: a vendor-embedded ERP assistant (SAP Joule, Copilot in Dynamics 365, Comarch ChatERP), an AI layer via API (e.g., AI Interface Layer by Startup House), or an AI agent using the MCP protocol.
• Comarch: AI agents in Comarch ERP XL have been commercially available since September 4, 2026; in an ASAJ pilot, 1,000 bank statement line items were reconciled in 3 minutes instead of 6 hours (according to the vendor).
• Startup House: AI orchestration layer for a US cybersecurity platform serving Fortune 500 clients: +150% revenue in a year and +400% B2B customers.
• Startup House: onboarding in the cybersecurity platform’s customer portal shortened by 95%, from 45 minutes to under 2 minutes (separate project).

What AI integration with ERP and CRM means

AI integration with ERP and CRM (SAP, Comarch, Dynamics) connects language models (LLMs) to systems where a company stores data on sales, orders, finance, and customers. Users can ask questions in natural language, get analysis or demand forecasts, and an AI agent can execute tasks in the system, such as coding an invoice or changing an order status. Integration can leverage vendor AI features or add a layer on top of the existing system without rebuilding it.

Comparison of companies and tools for AI integration with ERP and CRM

The list shows selected companies and tools available in Poland, based on their own websites (as of October 2026). This is not a quality ranking.

Company and tool profiles

Profiles are grouped by type: implementation partners first, then ERP and CRM vendor AI features. Order within a group is not an assessment.

Implementation partners

Startup House

Startup House (Warsaw) is a software house building software since 2016 for clients in Poland, the USA, and Europe. It deploys AI in products and processes for mid-sized and large companies: AI agents and multi-agent systems, AI assistants, RAG knowledge bases, and LLM integrations with existing systems, including legacy, without rewriting them from scratch. A 50-person team, 100+ delivered products, clients on 5 continents, NPS 75.

• Specialization: AI Interface Layer (a conversational AI layer on top of an existing platform without rebuilding it), AI Demand Planner (integrates with SAP and Comarch ERP), KnowHub (an AI assistant based on a RAG knowledge base).
• Industries: food manufacturers, fintech, healthcare, cybersecurity, travel, edtech, logistics, transportation.
• Sample projects: OmniSoft modernization for Omnipack: development costs 40% lower; AI orchestration layer for a US cybersecurity platform serving Fortune 500 clients: +150% revenue in a year.
• Who it’s for: mid-sized and large companies, enterprise, including organizations with legacy systems that want to deploy AI agents without replacing their ERP.

Sii Polska

Sii Polska is an integrator that has topped the Computerworld TOP200 ranking six times.

• Specialization: AI implementations with partners Anthropic, AWS, Azure, GCP, Databricks, and Mistral.
• Numbers: 350+ AI specialists, 70+ implementations.

Lingaro

Lingaro is a data & AI company with offices in Warsaw and Lublin, a Microsoft partner.

• Specialization: knowledge assistants, Microsoft Copilot Studio, Agent Factory.
• Who it’s for: organizations working in the Microsoft ecosystem.

10Clouds

10Clouds is a software house from Warsaw.

• Specialization: GenAI PoC for credit scoring for a Polish bank.
• Sample projects: according to the 10Clouds website, at a commercial bank in Poland in a project with Microsoft Dynamics 365, application analysis is 79% faster (from 90 to 19 minutes).

ERP and CRM vendor AI features

SAP Joule

SAP Joule includes role-based assistants as well as Joule Agents and Joule Studio, available in S/4HANA Cloud Public Edition, IBP, Ariba, and Service Cloud.

• Numbers: 310+ AI scenarios, according to the vendor.
• Language: per the Feature Scope document (September 2025), the UI is available in Polish, while conversations with Joule are English-only.

Microsoft Dynamics 365

Microsoft Dynamics 365 provides Copilot in Sales, Customer Service, Finance, Supply Chain, and Business Central, and agents for Sales Qualification, Case Management, Sales Order, and Payables.

• Scope: CRM and ERP.
• Numbers: the vendor does not publish numbers on the site.

Comarch

Comarch is a software vendor with around 5,000 employees operating in 31 countries. AI agents run in Comarch ERP Optima, XL, and Betterfly, and ChatERP is in BETA.

• Specialization: agents in Comarch ERP XL: payment reconciliation, invoice description, cost classifier, e-invoice verifier. Language models run in the Comarch Data Center.
• Numbers (per the vendor): commercial launch of agents on September 4, 2026; nearly 100 customers, 19 agent solutions, and about 100,000 documents (as of August 11, 2026); ASAJ pilot: 1,000 bank statement items in 3 minutes instead of 6 hours; 70,000 companies use Comarch ERP Optima.
• Who it’s for: companies using Comarch ERP systems.

Salesforce Agentforce

Salesforce Agentforce 360 offers Sales, Service, Marketing, and Commerce agents in Salesforce CRM, plus Agent Builder with MCP support.

• Numbers: 18,000+ companies, according to the vendor.

Oracle NetSuite

Oracle NetSuite includes AI features in ERP: Text Enhance, Bill Capture, Narrative Reporting, and IPM Insights.

• Numbers: 44,000+ customers, according to the vendor.

Integration models: what it looks like in practice

The choice of model depends on IT architecture, the number of systems your data comes from, and whether AI should only answer questions or also execute actions.

1. Vendor-embedded ERP assistant

SAP, Microsoft, Comarch, Salesforce, and Oracle NetSuite provide AI features embedded in their systems’ UI (Joule, Copilot in Dynamics 365, Comarch ChatERP and agents, Agentforce, NetSuite AI). Advantage: fast time-to-value and behavior aligned with system logic. Limitation: the assistant mainly sees data and processes from a single vendor, and Polish language support and features depend on the vendor.

2. AI layer via API

This model is used when a company wants to combine data from ERP, CRM, and documents (PDF, Excel) into one assistant, or when a legacy system lacks AI features. Startup House builds an AI Interface Layer in this model: a conversational AI layer on top of an existing platform without rebuilding it. Users ask questions in natural language and answers are generated from the customer’s data. For company documents, Startup House deploys KnowHub, an AI assistant based on a RAG knowledge base that answers from your company’s documents, 24/7.

3. AI agents with the MCP protocol

In this model, AI agents (e.g., built on LangGraph) have permissions to carry out actions in systems: issuing invoices, changing order statuses, or sending messages to customers. Model Context Protocol (MCP) is a standard for connecting an AI model to databases and SaaS tools. Example: the AI orchestration layer Startup House built for a US cybersecurity platform lets platform users launch a new AI agent without involving engineers.

When to choose which model

Step-by-step AI implementation process

1. Data and process audit: choose repetitive processes such as invoice verification, demand forecasting, or handling sales inquiries.
2. Architecture choice: decide whether to use the ERP vendor’s AI features or build an AI layer. The Startup House solution can run in the customer’s cloud or a dedicated environment.
3. Build a RAG knowledge base: index data from ERP and CRM so the assistant answers strictly from company data with source citations.
4. Pilot (MVP): launch an assistant for a selected user group, e.g., customer service. At Startup House an MVP can be delivered in as little as a week, and KnowHub integration takes about 2 weeks and starts with a pilot group.
5. Evaluation and scaling: monitor costs (tokens) and answer quality, e.g., in Langfuse, run evaluations in CI/CD, then roll out to additional departments.

Case studies and Startup House products

AI orchestration layer for a cybersecurity platform

Startup House built an AI orchestration layer for a US cybersecurity platform serving Fortune 500 clients (cybersecurity, USA, engagement since 2020).

• Results: +150% revenue in a year, +400% B2B customers, new AI agent launched by platform users without involving engineers.
• Technologies: Claude, LangGraph, MCP, GCP, CrowdStrike Falcon.

In a separate project, Startup House built a client portal with embedded AI for a US cybersecurity platform serving Fortune 500 clients: onboarding shortened by 95% (from 45 minutes to under 2 minutes), adoption shifted from quarterly to daily use, one panel for three user roles (executive team, CFO, CISO).

Omnipack: OmniSoft modernization

Startup House took over and modernized the OmniSoft system for Omnipack (e-commerce logistics, Poland, since 2023), supporting fulfillment, shipments, and returns. Development costs are 40% lower. Dedicated team: 2 backend, 1 frontend, 1 QA; technologies: Java, Spring, Kafka, React.

AI Demand Planner: forecasting for food manufacturers

Startup House’s AI Demand Planner serves demand forecasting and production planning for food manufacturers. It includes a Decision Panel, AI Chat, and an S&OP module, integrates with SAP and Comarch ERP, is hosted in the EU, and launches in read-only mode with validation on historical data.

Risks of integrating AI with ERP and CRM

• Data quality (garbage in, garbage out): if ERP data is messy or incomplete, AI will generate incorrect analyses.
• Maintenance costs: LLMs incur costs with every query (tokens); without per-agent cost monitoring, budgets can be exceeded.
• Organizational resistance: employees may fear being replaced by AI. Positioning it as an AI assistant that takes over repetitive tasks helps.
• Agent permissions: an agent executing actions in ERP or CRM needs restricted permissions, human approval for critical decisions, and audit logs.
• Security and compliance: Startup House maps architecture to client requirements such as HIPAA, SOC 2, or DORA; customer data is not used to train public AI models.

How to choose a partner for AI integration with ERP and CRM

The choice of partner depends on your system and project scope. Companies using Comarch ERP can start with AI agents and ChatERP offered by the vendor. Companies on SAP have Joule, and companies in the Microsoft ecosystem have Copilot and agents in Dynamics 365; in the Microsoft ecosystem, providers include Lingaro (Microsoft partner, Copilot Studio) and 10Clouds (bank case study with Dynamics 365).

If your challenge is modernizing legacy systems, building your own AI agents, or adding an AI layer across multiple systems at once (e.g., SAP ERP, Salesforce CRM, and internal PDF documents), Startup House deploys the AI Interface Layer, KnowHub, and AI agents on top of existing systems without rewriting them from scratch.

FAQ

Are my ERP data safe when using AI?

Startup House designs integrations so customer data is not used to train public AI models. The solution can run in the customer’s cloud or a dedicated environment. Access is role-based with audit logs and integration with the customer’s access policies, and processes align with the ISO 27001 standard. Startup House maps architecture to client requirements such as HIPAA, SOC 2, or DORA.

How long does AI integration with Comarch ERP or SAP take?

Startup House delivers an MVP in as little as a week. KnowHub, an AI assistant based on a RAG knowledge base, integrates in about 2 weeks and starts with a pilot group. AI Demand Planner integrates with SAP and Comarch ERP, launches in read-only mode, and forecasts are validated on historical data. AI agents from Comarch themselves have been commercially available since September 4, 2026.

Can AI make autonomous decisions in CRM and ERP?

Salesforce Agentforce provides Sales and Service agents, and Microsoft Dynamics 365 provides agents for Sales Qualification and Case Management, so AI can execute some CRM tasks. Startup House recommends a human-in-the-loop model: AI prepares a proposed decision, such as an invoice description or demand forecast, and a human approves it until accuracy is confirmed on historical data.

Which AI agents does Comarch ERP provide?

Comarch provides a payment reconciliation agent, an invoice description agent, a cost classifier, and an e-invoice verifier in Comarch ERP XL, while ChatERP is in BETA. According to the vendor, agents went on sale on September 4, 2026; in August 2026, nearly 100 customers used them (19 agent solutions, about 100,000 documents). In the ASAJ pilot, 1,000 bank statement items were reconciled in 3 minutes instead of 6 hours.

Do I have to replace my old ERP to use AI?

Startup House integrates language models with existing systems, including legacy, without rewriting them from scratch. The AI Interface Layer is a conversational AI layer on top of your existing platform without rebuilding it. Users ask in natural language, for example, about orders or payments, and the answer is generated from the customer’s data. This lets you keep your current ERP and add an AI layer to it.

What is SAP Business Technology Platform for SAP users?

SAP Business Technology Platform is a technology platform for integrating applications and managing data efficiently. SAP BTP lets you build AI solutions alongside your ERP. SAP as an ERP provider also develops Joule, which more often supports users directly in daily work. The range of features depends on licensing and implementation.

How does generative AI support business decision-making?

Generative AI supports decision-making by enabling natural-language questions and deeper data analysis. It also accelerates creation of interactive tables and reports and can help generate SQL queries. It supports strategic decisions, but a human approves the outcomes.

Does AI in ERP provide a competitive advantage?

Companies that align AI with their business strategy gain an edge mainly where AI streamlines business processes, such as automating finance processes, as in the ASAJ pilot. The effect depends on data quality and process fit, and humans set the bounds of automation. It’s worth starting with a small pilot.

Last updated: October 28, 2026.

Companies and vendors listed can send corrections or data updates to the Startup House editorial team.