Questions and answers

Yes. Through an MCP server or plugin, the assistant your company already uses gets access to selected data and permitted actions. You define the permissions, and every action is logged.

Not necessarily. Employees get new tools inside the assistant the company already pays for. Where a dedicated application or automation without a language model is a better fit, we suggest that.

The assistant reads the data it has access to, performs only actions within the permissions granted, and logs who requested what and what was done. You set the permissions and rules.

No. We build the solution as a layer on top of your existing systems. They stay in place, and we connect to them.

Yes. The system recognises speech, extracts customers, locations and quantities, matches them to your master data and prepares a draft order that the user confirms.

Speech recognition works in Slovenian. We link informal names of customers and locations to your master data through synonyms. We check accuracy in a trial on your data.

Yes. An agent guides the conversation with the sales rep, checks that all required fields are filled in and hands marketing a structured request. No more back-and-forth emails.

Yes, as a draft. It takes items, stock and prices from the system and applies discounts according to your rules. That is business logic, not a decision made by the model. A person reviews the quote before it is sent.

Yes, with a human in the loop. We score contacts against your ideal customer profile, and the salesperson approves messages before they are sent. LinkedIn restricts automation, so we do not do mass messaging.

Every contact has a source and a score for how well it matches the customer profile. We also check contacts by cross-referencing data from public registers and by checking that email domains are valid. A person always reviews them before use.

Yes, with a decision layer on top of your existing system: demand forecasting, receiving checks, timely markdowns and redistribution between stores. Each part has its own measurable indicator.

First we define the cost objects, then the reporting model, then automatic alerts, for example when the margin falls below a threshold. Management gets an up-to-date view, not once a month.

Yes. It cleans supplier data, groups it by product category, shows trends and anomalies, such as duplicate invoices, and exports the result to Excel or PowerPoint.

Yes. It prepares a monthly plan and allows weekly adjustments by the hour, taking into account the Slovenian Employment Relationships Act (ZDR-1) and your internal rules.

Copilot works well inside Excel and Microsoft 365. When you need your own logic, several sources or automatic workflows, a custom solution is needed. Where Copilot is enough, we tell you.

Yes. Bookings are synchronised both ways with Google Calendar, so there are no double bookings.

The big providers offer general-purpose assistants. The value lies in the connection to your data, rules and processes, and that part stays yours regardless of which assistant uses it.

Handling updates is part of monthly operation. We track model versions and, before an old version is retired, we test the system in good time and adapt the instructions to the new version so your process keeps running smoothly.

Yes. The assistant searches your internal knowledge base for answers and names the document the answer comes from. We set it up to say clearly when the documents contain no answer, and we check accuracy on a sample of questions before rollout.

Yes. Modern models communicate fluently in most languages. Your system and employees can work in Slovenian, and the assistant presents the data to a customer in Germany or Italy in their language.

Ready-made solutions require you to adapt your processes to their tool. We first analyse your process and then build AI in so that it adapts to your existing systems and rules.

Yes. Multimodal language models extract numbers, dates and amounts from a scanned document, technical drawing or photo and pass them to your ERP.

Yes. The system reads every incoming message, recognises what the sender wants and sorts the message or assigns it to the right person. Where replies are routine, for example shipment status, it checks the data itself and prepares a draft reply.

Yes, we can build the assistant into your B2B portal. Customers can then ask technical questions about products, check open invoices or place an order in the chat at any time, which takes load off your sales team.

Yes. Management can write, for example, “show sales by region over the last year”. The system finds the right data, displays a chart and allows export to a spreadsheet.

When describing the process, we list the exceptions and build them into the business logic. If the system encounters a case the rules do not cover, it does not guess; it stops the process and asks the responsible employee to decide.

With one process that is repeated often, takes a lot of time and has clear rules. We first describe it as a decision (criterion, rules, exceptions) and only then choose the tool.

Those with a lot of repetition and clear rules, where an error is not critical and a person spots it quickly: data entry, draft quotes, sorting email, reports. Processes with rare, very varied cases come later.

Company size is not decisive. What matters is whether you have a process that is repeated often enough for the savings to show. We assess this during the assessment.

Most often because they start with the tool instead of the process, because there is no measurable baseline, or because the solution stays on the side of everyday work. That is why we start with a description of the process and indicators.

From €490 to €1,490, depending on the size of the company and the number of processes. The introductory call is free of charge.

From €690 to €890, for one process using your data.

From €2,900 to €3,900, and it takes one month. If the scope requires more time, we prepare a new estimate.

From €90 to €149 per month. It includes hosting, monitoring and updates.

The starting point is how much the solution saves you each month. The total cost must be substantially lower than the savings.

For small and predictable usage, the cost is included in the monthly subscription. For systems with heavy data processing, the cost is billed by actual usage directly with the provider, without our markup, so it is the lowest the chosen provider offers. We prepare an estimate during the assessment.

Yes. A pilot for larger or regulated projects costs from €5,000 to €12,000, depending on the scope and the regulation.

We plan the rollout so that the initial investment pays off in less than a year. After the rollout, monthly costs with measurable results are lower than before, or the agreed non-financial goals are reached, for example faster response or fewer errors. We prepare the calculation during the assessment.

Before we start, we agree in writing on the baseline and the measurement method. We measure the effect using the same method, so there are no disputes when invoicing.

A description of the current process, sample documents or data, a list of the systems you use, the number of users and the desired result. The fastest way is a 30-minute call where you show us the process on an example.

A free introductory call, a paid assessment, a trial on one process, a production rollout within one month, and then monthly operation.

You keep full ownership of your data and business rules. The intellectual property of the core code and the integration layer stays with OI, and you have the right to use it under the contract.

When the collaboration ends, we export all your data in a standard format (CSV or JSON). After the notice period, access to the system and the right to use OI software end.

We keep the user experience as simple as possible. Where we can, we build the assistant into tools employees already use (Teams, Slack, ERP), and during the production rollout we run hands-on training on their own cases.

Only basic use and knowledge of the process. We take care of the infrastructure, maintaining integrations and model updates.

No. Our goal is to take time-consuming retyping, copying and searching for data off employees' hands. That lets them focus on work where people create the most value: customer relationships, judgement and handling exceptions.

A trial on one process shows the result on your data, and the production rollout takes one month.

No. Employees use the solution in tools they already know, for example in the assistant the company already has or in a form in the existing system.

We involve the people who run the process from the moment we describe it. We offer training on using AI, and during the rollout hands-on training on real cases from their work.

With indicators we define before the start, for example time per task, error rate and the share of cases without manual correction, and with a baseline measured before the rollout.

A description of the process, access to the data or the system where the process runs, a person who knows the process and checks the results, and a management decision on who confirms the AI's proposals.

Yes. A trial on one process with your data costs from €690 to €890 and shows whether the rollout pays off before you decide on more.

Yes. With the API and business plans you can put a data processing agreement in place (Article 28 GDPR) and control data retention. We do not send personal data unless it is necessary, and we pseudonymise it where needed.

The solutions are hosted in the EU. We route language model calls to EU endpoints where the provider offers them.

You are a deployer. Obligations for the model itself lie with the model provider. What matters most for you are the transparency rules, for example in contact with customers. You must also take measures to support your employees' AI literacy; basic training, which we offer, helps with this.

For data that goes on documents, such as a tax number or address, we do not take it from the model but from official sources, for example VIES or the Slovenian business register AJPES.

No. AI prepares a proposal, and a person decides. This is also what the AI Act requires as human oversight for high-risk systems.

Software that supports clinical decision-making can be a medical device under the MDR and a high-risk system under the AI Act. That is why we build in traceability, human oversight and documentation from the start.

No. We only use business plans and APIs for which the model providers (OpenAI, Anthropic, Google) do not use data to train their models. Data retention is limited according to the provider's terms. Where a provider offers zero data retention, we use it.

The AI model does not calculate or set prices. When the assistant prepares a quote, it sends a request through MCP to your system, which returns the exact price and the approved discount from your price list. The AI only puts this data into text.

Yes. The system checks the user's permissions. When a worker asks about stock, the assistant returns the warehouse level. If the same worker asks about margin, the system rejects the request and says they are not authorised for that data.

By not sending personal data to the model provider unless necessary, and by having a legal basis, a data processing agreement and EU hosting in place for every processing activity. We connect the assistant to your systems through an MCP server that determines which data it can access and logs every action.

Yes, but not with a consumer account. Use a business plan or the API with a data processing agreement and retention settings, and set up access to company data through a connection that defines permissions and logs actions.

Where a mistake would have consequences, the AI prepares a proposal and a person confirms it. Every action is logged, so a mistake can be found and corrected.

The company that uses the AI. That is why we define who confirms a proposal and write it into the work procedure.

Yes. The method depends on the version: Business Central has a REST/OData API, older NAV versions have web services, otherwise we work with export and import. We check the version during the assessment.

In the trial, yes. For production with several users we recommend forms and a database, because Excel is not reliable for that.

With one line of code that you add to your site. The assistant runs on our infrastructure, and updates take effect immediately.

A booking only becomes valid once the user confirms it by email. The number of attempts is limited.

Yes. We can expose the solution's REST API as FHIR resources (for example Patient, Encounter, Observation). We determine the specific resources based on your hospital system.

No. Our architecture is not tied to any single model. If a faster, cheaper or better model appears, or if you want to switch to a local open-source model, we switch the back end without replacing the whole system.

The system has built-in safeguards. If the primary model is temporarily unavailable, the call can be automatically routed to a backup model. The user is informed of the status, and no data is lost in the meantime.

It depends on how demanding the task is. For simple queries the response is almost immediate; larger documents or tables are processed in the background, and the system notifies you when the result is ready.

MCP is an open standard through which an assistant securely accesses your systems, such as an ERP or documents. The assistant only gets the data it needs for an answer, without copying it into a separate database. Data goes to the model for processing, not for training, and you decide on access.

Yes. For organisations with the highest security requirements, we can install the whole solution, including an open-source language model, on your own servers. That way no data leaves your network.

No. We do not train models on your data, because that would be expensive and would become outdated with every change, for example to the price list. For every question, the assistant reads the latest data directly from your system.

Yes. For most local and regional systems (Pantheon, Vasco, SAOP, e-invoices) we can build custom connectors. We connect through the existing API, web services or directly to the database, if we have the necessary permissions.

In such cases we use robotic process automation (RPA). The system captures and enters data through the user interface, just as a person would click and type.

Only the integration layer needs to be replaced; the business rules, processes and user experience stay the same. Employees do not need to know that the back-end system has changed.

Yes. We size the solution for the expected load, including peaks, and load-test it before rollout. If volume exceeds the plan, we increase capacity.

For most business processes they are close in capability. What matters more than the model are the terms of the business plan, where data is processed and which systems it connects to. That is why we build solutions that work with the assistant the company already uses.