Operations and Technology: How Modern Publishers Simplify Workflows, Use AI Effectively, and Scale
Speaker
Executive summary
AI should not be treated as a collection of isolated task tools. It cuts across leadership, workflow design, security, architecture, quality, policy, and the wider publishing value chain. Arantza argues that senior leaders first need enough understanding to connect AI to the company's mission and decide which problems are worth solving. Middle managers and employees then need approved tools, clear rules, and a route for experiments to be reviewed and integrated.
The session also highlights a less obvious risk: incentives. When performance measures reward speed without rewarding verification, employees who question AI output can appear less productive than colleagues who accept it uncritically. Effective governance therefore has to define not only which tools may be used, but how quality, judgment, disclosure, and responsibility will be evaluated.
Key takeaways
- AI adoption is a leadership and operating-model challenge, not only a technical project.
- Senior leaders need practical understanding before they can set priorities, infrastructure choices, and governance.
- Middle managers carry pressure from both directions and must be involved early.
- Employees will use accessible tools even when no formal program exists, creating security and confidentiality risks.
- Bottom-up experiments need a clear route for review and integration or useful work will be lost.
- Generic models, generic objectives, and generic data can push competitors toward the same recommendations.
- Governance must cover authors, agents, publishers, booksellers, libraries, and platforms, not only internal employees.
- KPIs should reward verification and quality as well as speed.
What publishers can do next
- Give the executive team enough practical AI education to make informed strategic and governance decisions.
- Identify priority workflows and risks, then produce a short, usable policy covering approved tools, prohibited data, review obligations, and disclosure.
- Create a formal route for employees to submit experiments for security, architecture, and operational review.
- Involve middle managers and users in tool selection and workflow redesign.
- Review KPIs to ensure careful validation is not treated as lower productivity.
- Discuss AI use with libraries, retailers, agents, and other partners before scaling new products or features.
- Delay major workflow redesign until teams have enough hands-on experience to understand what should actually change.
Examples and evidence
- Arantza described a useful vibe-coded departmental tool that was stopped after the CTO objected that it did not fit the existing platform or API model.
- She described employees using personal AI accounts with company content, including potentially sensitive material, because no approved route existed.
- An educational publisher reorganized its knowledge and tested adaptive tutoring and personalized exercises based on student interests.
- A publisher brought 800 AI-produced audiobooks to a library tender and discovered that the library did not want AI-produced audio.
- A bookseller proposed automated reading recaps, while publishers objected to the system deciding what readers should remember.
- An employee who took time to verify AI output was criticized for slower delivery, exposing a conflict between quality and productivity measures.
Important nuance
Arantza did not recommend redesigning every workflow immediately. Her sequence is deliberate: leadership understanding and governance first, practical experimentation and employee involvement second, and deeper workflow redesign after the organization has learned enough to make informed decisions.