Odysı
See if we fit
Case Study Product leadership

A modern, AI-forward product function at Libnova.

Libnova builds digital preservation software trusted by national archives, universities and museums to keep content intact for decades. After a search-fund acquisition, we took the fractional CPO role: building a real product organization, AI-native from the start, and pointing it toward a scalable platform.

Client Libnova · digital preservation
Our role Fractional Chief Product Officer
Type Product & operating-model build
Status Foundation set · winding down
Read
The problem

A company that grew on engineering, without a product function.

Libnova grew the way strong technical companies grow: organically, driven by engineering and deep domain expertise, without a formal product function between the customers and the code. That works until it doesn't. By acquisition, two problems had built up.

01

Structural

Product decisions had no clear owner, no shared vision, no repeatable way of working. No product managers in the modern sense, no competency model, no goals framework. Priorities were set case by case.

02

Technical & commercial

The software had been shaped client by client. Fine with a handful of bespoke customers, a real constraint when the plan is to scale as SaaS, where every new client adds cost instead of leverage.

FIG. I · Bespoke to platformdivergent
PER CLIENT ONE PLATFORM Repeatable deployment, without regressing the integrity guarantees.
What we built

Two things at once: a product organization, and an AI-forward way of running it.

The organization

The scaffolding a scaling SaaS company needs

  • Hired product managers and built a competency model for what good looks like
  • Set a clear product vision and structured the teams and their ways of working
  • Put a goals framework and product OKRs in place, aimed at outcomes not activity
The AI-forward layer

Make the company's knowledge legible to AI, then remove manual work

  • Structured knowledge from code, customers and support so people and AI can use it
  • A pipeline that generates documentation from the codebase and Jira, per release
  • Internal reporting and product dashboards stood up with Claude Code, assembled automatically
How it works

Documentation as a by-product of work already happening.

The docs-as-code pipeline takes what the team already produces, merged code and the Jira tickets tied to each release, and drafts documentation in two registers. It lands in Confluence as clean Markdown. A person reviews and approves before anything external ships.

FIG. II · The docs-as-code pipelineidle
MERGED CODE JIRA TICKETS docs-as-code INTERNAL EXTERNAL HUMAN REVIEW CONFLUENCE
Internal docs land directly. External docs pass a human review gate before they reach a customer. AI-forward, not AI-unsupervised.

Reporting works the same way

Rather than a PM pulling numbers into a status update each cycle, the Claude Code dashboards assemble reporting and insights from the underlying sources, so the team reads and decides instead of collecting and formatting.

One deliberate constraint

It is easy to point AI at a company and generate a mountain of docs no one reads. We built for the opposite: anchored to what teams actually reference, judged by whether it gets consumed rather than by how much exists.

How we de-risked it

Integrity is the whole promise. It could not regress.

Platformisation carries the real risk. You cannot move a client onto a more scalable deployment at the cost of the guarantees that make the product worth buying. So repeatable deployment had to match the reliability of the bespoke setups it replaced, never trade reliability for speed.

The same discipline applied to the AI tooling: generated documentation was reviewed by a person before it reached a customer, and automation was trusted only where its output could be checked against a source of truth.

Guarantees held through the change
  • OAIS alignment (ISO 14721)
  • Fixity and integrity checking
  • Geo-redundant, self-healing copies
  • Standards compliance

AI-forward, but not AI-unsupervised. The trustworthy version was the only one allowed to ship.

Results

A foundational engagement, described by what is now in place.

The deeper platform work is still underway and the value compounds after the function is established, so the honest measure here is the direction of travel rather than any single metric.

Organisation

From an informal, engineering-led setup to a real product org: vision, PMs on a competency model, structured teams, OKR-driven work.

Documentation

From a manual task that got skipped to an automated by-product of each release, generated from code and Jira.

Reporting

From hand-compiled updates to dashboards assembled automatically, so the team reads and decides.

Knowledge

Meaningfully more structured and AI-legible, the groundwork for everything that follows, with the platform direction set and moving.

What's next

The fractional model is built to conclude.

The handover

Establish the function and operating model, then hand a working organization to permanent leadership, the natural next step as our engagement winds down.

The platform

Platformisation continues from the foundation now in place: from bespoke installations to repeatable, reliable deployment so the business can scale.

The systems

The AI-forward systems, once embedded, keep paying off without needing us in the room, which is the point of building them properly.

About Odysi

We design and build websites, web applications and automation systems for B2B companies, and we take on product leadership where a company needs the function built rather than just advised on. On this engagement our role was the fractional CPO: the product organization, the operating model, and the AI-forward systems underneath it.

Hands-on product leadership.

Outgrown an informal product setup, or want to be AI-native?

Related solution AI for digital preservation Read the solution