Sanoma Tried to Build an AI Tool. It Ended up Rebuilding Its Workflow

Why It Matters

Standardising the data capture process is a prerequisite for reliable newsroom AI, highlighting that automation gains stem from process discipline rather than technology alone.

Key Takeaways

• Inconsistent interview recording blocked scalable AI at Sanoma
• Standardizing input with Elisa Ring enabled AI pipeline rollout
• AI drafts still required editorial editing due to hallucinations
• Defining “good” output proved harder than building the model
• Workflow redesign, not AI, delivered productivity gains

Pulse Analysis

Newsrooms worldwide are racing to embed generative AI, yet many overlook a fundamental prerequisite: a clean, consistent data pipeline. Sanoma Media Finland’s experience underscores that without a uniform method for capturing interview audio, even the most sophisticated transcription engines falter. By consolidating disparate recording habits into a single, automated service—Elisa Ring—the Finnish publisher eliminated manual transfers and created a reliable "miracle wire" for AI ingestion. This shift mirrors a broader industry trend where legacy workflow quirks, not algorithmic limits, become the bottleneck to scaling AI.

The technical layer that follows a standardized feed presents its own set of challenges. Transcribing Finnish speech demands high linguistic precision, as minor errors can alter meaning dramatically. Sanoma’s pilot revealed that setting quality thresholds—whether by word‑error rate, speaker identification, or contextual relevance—was more complex than engineering the model itself. Moreover, the generated summaries and drafts suffered from typical hallucinations, mis‑ordered facts, and occasional quote misplacements, forcing journalists to act as quality gatekeepers. These findings highlight the importance of rigorous output evaluation frameworks and the need for AI systems that can be tuned to domain‑specific standards rather than relying on generic benchmarks.

Strategically, Sanoma’s journey illustrates that AI adoption is as much an exercise in process engineering as it is in technology procurement. By first solving the input problem, the organization unlocked a modest but measurable efficiency gain, allowing journalists to treat AI output as a trusted starting point rather than a replacement. The lesson extends beyond phone interviews: any media operation seeking AI‑driven productivity must audit and standardise its upstream workflows. Companies that embed these disciplined practices stand to reap faster ROI, reduced editorial risk, and a smoother path toward fully automated content pipelines.