Konciv brings conversational AI to offshore operations with Katalan

The Stavanger-based software firm has moved its AI capability from experiment to production, betting that the bottleneck in operational software is no longer recording work — it is retrieving what has already been recorded.

https://video.wixstatic.com/video/e2cbfa_6bcda29a95b94054ac939ea9dc105530/1080p/mp4/file.mp4

Konciv AS has released Katalan, a natively integrated artificial intelligence layer that allows users of its operations platform to query their own data in plain language and receive verified answers in seconds.

For Konciv, which supplies operations software to offshore oil and gas service companies, the release marks a deliberate escalation of its investment in artificial intelligence. Where much of the sector has approached AI as an adjacent feature — a summarisation tool here, a search box there — the company has positioned Katalan as a layer over the entire platform, and has been explicit about the ambition attached to it.

"The goal is that Katalan becomes the main interface for insight in Konciv," says Henrik Heggland, CEO and founder of Konciv. "Not a feature people click into occasionally, but the first place they go when they need to know something."

A retrieval problem, not a recording problem

The premise behind the product is a familiar frustration in operationally intensive industries. Systems of record capture data thoroughly. Extracting an answer from that data remains a manual, skill-dependent exercise.

According to Heggland, the constraint is structural rather than a shortcoming of any particular module. Workflow screens are designed for the people doing the work — planning a mobilisation, updating a certificate, managing equipment — and they perform that function well. They were not designed to serve as a query interface, and no vendor builds a screen for every possible question.

The practical consequence, the company argues, is a two-tier organisation. Users fluent in the system can locate an answer in minutes. Occasional users — project managers, base coordinators, personnel covering a rotation — typically cannot, and instead route their question to a colleague and wait. Questions spanning multiple areas of the platform, such as those combining personnel, assignments and absences, frequently have no single screen at all and are resolved through spreadsheet exports.

"What customers notice first is not the technology, it is who can suddenly get answers," Heggland says. "The power users always knew where to click. It is everyone else where the change shows up. Time-to-answer used to depend on who was asking."

Usage in practice

In operation, Katalan accepts questions in English or Norwegian through the Konciv web application or Microsoft Teams, and responds in the language of the question. Typical queries cited by the company include which components are ready for a given project, which personnel are booked into a particular office in a given week, what has arrived at a base over a period, and how work assignments are distributed across positions.

The system determines what to retrieve; the resulting figures are computed programmatically and checked against the underlying data before a response is returned, an approach intended to address the reliability concerns that have limited enterprise adoption of language models for numerical work. Where a breakdown aids interpretation, results are charted inline. Every record referenced in an answer links back into Konciv, allowing users to verify a figure at source rather than accept it.

Each response also exposes the steps taken to produce it, a transparency measure the company regards as necessary for operational decision-making.

Agnostic core, learned context

Architecturally, Konciv has taken a position that distinguishes Katalan from both generic assistants and bespoke implementations. The core system contains nothing customer-specific: it serves equipment and rental, personnel and competency, and projects and logistics from the same engine, and understands only Konciv's data model, permissions, approvals and audit functions.

Customer-specific knowledge is layered on top and can be taught to the system — terminology, internal codes, the business rules and processes particular to an individual operator. That context, once provided, applies across the company and does not extend beyond it.

"Generic AI is of little use in this industry, and bespoke AI does not scale," Heggland says. "So the core is agnostic and the context is taught. A customer injects their own operational reality, and from that point the system understands their operation rather than a generic version of it."

The approach also has commercial implications the company is candid about: onboarding a new customer or a new operational domain becomes a configuration exercise rather than a software development project.

Beyond querying, the platform supports the construction of workflows — recurring operational routines such as pre-mobilisation checks or period-end reviews — that Katalan executes on the user's behalf, with human approval required at any point where data would change.

Controls

Konciv has built the product around several constraints that reflect the governance expectations of its customer base.

Access rights mirror the user's existing Konciv permissions and are evaluated on every request, covering company, projects and locations. The company states that no input can widen that scope and that queries cannot cross company boundaries.

The system does not write data autonomously. Requested changes are staged as proposals showing precisely what would be written; a human approves or rejects, and approved changes execute under the approver's own identity and appear in the audit trail attributed to that person.

Queries are served from a dedicated copy of the customer's data rather than from the production system, an arrangement intended to ensure that query volume cannot affect day-to-day platform performance. Usage operates under per-customer plans with enforced ceilings on requests and consumption, and individual users can view their remaining allowance. AI processing takes place within an EU data zone.

Outlook

Konciv reports that Katalan's full stack is running in production under continuous monitoring, with write capability enabled on a customer-by-customer basis. Releases are gated against a battery of test questions with verified answers.