GoldenrozBook a call

Enterprise AI solutions: agents that work on your ERP, CRM and Microsoft 365 data under your access rules

Enterprise AI solutions are used in practice to answer employee questions from ERP, CRM and document systems, automate multi-step back-office processes, assist support and sales teams, and draft reports from governed data. Goldenroz builds AI agents and the integration layer beneath them, with single sign-on, per-user permissions, PII masking, audit logs and on-premise deployment where required. Projects start at $25,000 at a fixed price per phase, and larger programs run $100,000 and up, after a free one-hour scoping call.

From $25,000, fixed price per phase

Updated 2026-11-14. By Shai Goldenberg

What usually goes wrong

  • Pilots that never reach production
    A proof of concept on sample data impresses a steering committee, then stalls for months at security review because it has no access model, no logging and no owner.
  • Shadow AI
    Employees paste customer data and source code into personal accounts on public tools. Blocking the sites does not remove the need that sent them there.
  • Permissions do not carry over
    A search assistant indexed with a service account shows a salary file or a board paper to anyone who asks the right question. Existing access rules must apply to every answer.
  • Data is spread across systems of record
    The answer to one business question sits partly in SAP or Oracle, partly in Salesforce, partly in ServiceNow and partly in SharePoint, each with its own owner and API.
  • Procurement outlasts the technology
    Vendor risk questionnaires, legal review and architecture boards take a quarter or more. The model that was evaluated has been superseded by the time the contract is signed.

What is included

  • Use case and control scope
    A written scope naming the process, the systems of record, the data classes involved and the controls your security team requires before go-live.
  • Identity integration
    Sign-in through your identity provider such as Microsoft Entra ID or Okta, with group membership driving what each user and each agent may reach.
  • Governed data access
    An MCP-based gateway between models and your ERP, CRM, SQL and Microsoft 365 systems: read-only by default, approved queries only, PII masked before it reaches the model. See AI integration.
  • Deployment in your environment
    Installation inside your network or tenant with Docker Compose where data may not leave, using the model contracts you already hold.
  • Security review package
    Architecture and data flow diagrams, the threat model including prompt injection, the permission design and test evidence, prepared for your review board.
  • Audit and cost reporting
    A log of every query, tool call and output by user, exportable to your SIEM, and token cost tracked per run and per department.

Enterprise AI agents need an access layer before they need a model

Models are a commodity purchase. What decides whether enterprise AI agents reach production is the layer between the model and your systems of record: who is asking, what they may see, which operations are allowed, and what is written down afterward.

Shai owns an on-premise MCP gateway product built for this purpose: about 27,000 lines of code and 13 connector types. It lets Claude or ChatGPT answer from ERP, CRM, SQL and Microsoft 365 systems with per-user permissions, PII masking, an audit log and prompt-injection defenses. The model runs approved queries and never writes its own SQL.

That design answers the questions a security architect asks first. An employee gets the same data through the agent that they could already open themselves, and nothing more. Text retrieved from documents and tickets is treated as untrusted, so an instruction planted in a file cannot widen the agent's access.

Enterprise automation with agents compared to RPA

Earlier enterprise automation relied on robotic process automation: scripts that click through screens the way a person would. RPA is dependable on stable, structured steps and brittle when a screen changes or an input is free text.

Agents handle the unstructured part, such as reading an email, a contract or a ticket and deciding which path applies. They are less predictable than a script. So we keep the two roles apart: plain code and existing workflows do everything that can be a rule, and the model is called only for the steps that need reading. Each step is logged and testable alone.

Enterprise AI development with us is run by a named engineer, not a rotating team. That suits a first production agent or a gateway deployment, delivered in fixed-price phases. It does not suit a multi-year transformation program that needs dozens of consultants on site. The delivery method is on the AI agent development page, and Shai's background is on his profile.

How the work runs

  1. Scoping call

    Before the project

    We map the process, the systems involved and the cost of a wrong action. You get a fixed price for the first phase.

  2. Prototype

    About 1 to 2 weeks

    A working agent on one narrow task, run against real examples in a test environment. You see where it fails before you pay for production.

  3. Production build

    Roughly 4 to 8 weeks

    Integrations, permissions, fallbacks, approval steps, logging and the evaluation set. The range depends on how many systems the agent connects to.

  4. Launch with limits

    At launch

    The agent starts on a share of the volume, or with every action approved by a person. Limits widen as the logs show it holds.

What it costs

Every phase has a fixed price agreed before work starts. Model usage is paid to the model provider from your own account.

See full pricing

  • Focused agent or agentic workflow
    Fixed price per phase
    From $25,000
  • Larger AI platforms
    Several agents, many integrations or a full product
    $100,000 and up
  • Scoping call
    Ends with a written scope and a price
    $0
  • Model usage
    Billed to your own account
    Paid to the provider

Who it is not for

  • Organization-wide transformation programs that need a large consulting bench on site.
  • Buyers whose procurement requires a vendor with SOC 2 or ISO 27001 certification of its own.
  • Training or fine-tuning foundation models. We build on existing models.

Who does the work

CTO and technology partner

Shai leads the technology side: business software, SaaS and mobile apps, payments and escrow systems, Postgres security, AI agents built with Claude Code and MCP, and IT and infrastructure projects with senior infrastructure engineers.

Questions buyers ask

What do enterprise AI solutions cost?

A focused agent or a gateway deployment starts at $25,000 at a fixed price per phase. Programs with several agents, many integrations or a full internal platform run $100,000 and up. Model usage is billed by your provider under your own contract.

Do you provide enterprise AI consulting before development?

Yes. The free one-hour scoping call covers one use case. A wider assessment across business units, ending in a ranked list with costs, is offered as AI consulting at a fixed price.

Can AI agents for enterprise run fully on-premise?

The integration and permission layer can, and that is how the MCP gateway is installed. The language model itself is either a cloud service you have approved or a model you host. We design for whichever your policy allows.

Does Goldenroz hold SOC 2 or ISO 27001 certification?

No. Goldenroz is a two-partner firm and we do not claim certifications. The software is deployed in your environment under your controls, the code is in your repository, and we supply the documentation your security review needs. We are not auditors and do not issue attestations.

How are your AI solutions for enterprise different from Microsoft Copilot or a platform vendor's agents?

Platform assistants work well inside their own suite. A custom build is for processes that cross suites, need your own approval rules, or must run where the vendor's cloud is not permitted. Many companies run both.

Guides on this topic

Talk to the person who will do the work

One hour, no sales team. You talk to Shai Goldenberg, the CTO, about your systems or your product, and you leave with a clear next step and a price range.

Or send the details

Book a 1-hour consultationWhatsApp us