Your team needs AI ContentOps
When incorporating AI into content development, AI is the easy part. The challenge is getting surrounding processes and tools right.
AI ContentOps
The discipline of AI ContentOps combines three areas:
- Artificial intelligence (AI)
- Content development (eg. writing product documentation)
- Operations (eg. publishing, translation)
AI ContentOps has three goals:
- Infuse AI into internal processes and tools
- Use AI to improve the external customer content experience
- Help content team members keep their AI skills up to date
AI ContentOps focal
Because AI is moving so rapidly, discovering how it could be usefully applied to content-related processes and tools requires continuous learning and a lot of experimenting.
But writers are busy with their day-to-day responsibilities of collaborating with design on interface content, writing topics, recording how-to videos, and creating samples and tutorials. Infrastructure and operations teams are busy publishing content for different products and versions, managing translation, troubleshooting build problems, and performing site maintenance for online documentation. Even if an enthusiastic team member decided to carve out time every Friday afternoon to work on AI-related projects, they would struggle to make significant progress in that limited time.
So, as with anything else, if you want your team to benefit from the potential of AI, the least effective approach is to make it everyone’s responsibility: “Hey, everyone! Try to use AI more.”
Instead, a more successful approach is to make sure someone owns it. To make sure that the buck stops with someone, find one or more people to focus on AI ContentOps.
Benefits of adopting AI ContentOps
Specialists have deep knowledge of their own area and less knowledge about other areas:
- Writers are experts when it comes to crafting effective content to support users
- Infrastructure or operations teams have a deep knowledge of the workflows, pipelines, tools, and tasks for publishing content
- AI developers are skilled at building models, using AI to perform tasks like classification or named entity extraction, and prompting large language models (LLMs)
AI ContentOps focals who are generalizing specialists have knowledge of and experience with all three of these areas. And they take a holistic view of their work. They can therefor see opportunities to use AI in content creation and publication where specialists, who are immersed in only their own area, cannot. AI ContentOps focals will be more likely to design solutions that are a good fit for the AI technology, integrate well with existing processes, and add real value (as opposed to using AI for the sake of using AI.)
Examples
- Someone who has experienced wasting time hunting for meeting recordings containing design discussions for a new feature that needs to be documented and who is familiar with speech-to-text and AI search techniques would naturally look for ways to implement video searching solutions. (See: video search)
- Someone who has experienced the pain of constantly rerecording how-to videos when product features change and who has experience with text-to-speech and machine translation would naturally look for ways to streamline video maintenance by generating video audio in multiple languages from easy-to-maintain, single-language text transcripts. (See: video transcript translation and maintenance)
- Someone who has experience leading user testing of documentation, running workshops with customers, or supporting customers in the field as they navigate product documentation and who has experience with retrieval-augmented generation (RAG) would naturally explore using RAG to make it easier for users to consume documentation. (See: watsonx.ai search-and-answer)
- Content teams need to perform the same planning, brainstorming, and reflection activities that other team do. Someone experienced with leading those activities and experienced with NLP and LLMs would naturally use AI to facilitate the exercises and analyze results. (See: seeding a brainstorming activity, classifying reflection comments, analyzing and summarizing employee feedback)
- There is no shortage of AI training available online, but a content team member would want to focus their learning on tools and techniques that they can use for content-related work. Someone experienced with content work and expert with AI would be able to tailor training for a content audience. (See: NLP workshop, Chatbot workshop, MURAL workshop)
The risks of not adopting AI ContentOps
Not approaching the adoption of AI for content-related processes holistically risks failing to benefit from the opportunities of AI.
If you leave your content developers and infrastructure/operations teams in silos to adopt AI as best they can, or if you bring in an AI specialist to bolt AI onto content and publishing processes and tools, you risk wasting time building solutions that don’t improve anything.
Examples
- An AI specialist might propose using LLMs to generate documentation based on code comments or commit messages without understanding the complexity of content development: content design (the process of understanding what users need), content taxonomies (how documentation topics are structured based on the type of information they contain), legally approved terminology, trademarking requirements, globalization, and accessibility requirements, just to name a few. The problem with the naïve solution would be that the generated content simply couldn’t be published as documentation, even with significant rewriting. Someone taking a holistic view might instead integrate code comment and commit messages into a search solution for writers to quickly find needed info. (See: What many people get wrong about using LLMs in content development)
- A content developer who is enthusiastic about the potential of named entity recognition to analyze customer feedback might build a solution based on the assumption they can use default, out-of-the-box, general language models. However, because customer comments would likely contain domain-specific jargon, this use case is not a good fit for general-purpose language models. Someone with experience building this type of AI solution would anticipate having to build and maintain domain-specific dictionaries and plan for that in the overall solution and related processes. (See: general vs customized language models)
- An infrastructure team member might hope to use LLMs to edit error messages to make them more effective. They might suppose that fine-tuning a model with message writing best practices would do the trick. Unfortunately, the challenge with trying to improve error messages is that without an understanding of the error handling that returns the message, it is difficult to know what details to put in the message. Even knowing that guidelines say a message should be specific about what went wrong, trying to rewrite the message “An error occurred” without any additional information about the underlying error is an exercise in futility. (See: Error messages)
Supporting your AI ContentOps focals
Because driving change is an uphill battle, it’s crucial to support the efforts of your AI ContentOps focals:
- Give them an appropriate title, like “AI ContentOps Lead” or “AI ContentOps Architect”, to signal that management sees these people as leaders in this space.
- Boost the visibility of the work and actively support efforts to drive change, such as by having senior leaders post blogs or announcements about AI ContentOps projects, to signal that leadership agrees with the direction of the work.
- Include some aspect of AI ContentOps work in organizational goals (eg. KPIs, if you use those) to signal that senior leaders expect everyone in the organization to support and align with these efforts.
- Reward people for contributing to AI ContentOps projects.
- Regularly check in with your AI ContentOps focals to see what assistance could help them with challenges or blocking issues.
In other words, leaving your AI ContentOps focals to their own devices and expecting them to persuade people to contribute to projects, change processes, or use different tools on their own sets them up to fail.
Conclusion
Without AI ContentOps, your content processes won’t benefit from AI.
