Some Documentation Managers and Technical Authors are now being asked to demonstrate how they are using AI and to quantify the benefits. Here are some suggestions on how to approach that challenge.
Before you start
It is hard to measure benefits if you have no baseline. Before making any changes, identify what you currently measure and document your starting point. This will help you demonstrate progress over time.
Useful metrics to capture include:
- Time to market
- How long does it take from when a writing task is created to when content is published?
- Output quality
- How many errors reach production?
- What does user feedback say?
- Team productivity.
- How much content does the team produce?
- What types of content can it create?
- Return on investment
- What are current production costs?
- How many support calls does your documentation deflect?
- What does your Knowledge Base analytics show?
You should also check whether your team has a way of logging activity consistently, as this data will underpin any reporting you need to do.
Option 1: Start small
Look for situations where AI can be applied quickly and with low risk. This lets your team build confidence and demonstrate early wins while you develop more ambitious plans.
Good starting points include tools for:
- Estimating the time required for each piece of work
- Scanning existing docs to flag outdated content, broken links, or inconsistent terminology
- Generating and editing metadata
- Generating alt text for images
- Drafting release notes from change logs or ticket descriptions
- Suggesting improvements to readability or plain language
Starting small also gives you a chance to spot problems early, such as AI-generated errors or tone inconsistencies, before they affect high-visibility content.
Option 2: Focus on the big wins
Treat this as a change management or process improvement project. Investigate where the biggest problems and opportunities lie, then assess whether AI can address them. This approach takes longer but it can produce more significant results.
Activities can include:
- Waste analysis (applying Lean methodology)
- Where are the biggest time sinks
- Where are people doing work that adds the least value?
- Gap analysis
- What is undocumented?
- What do users want that you’re not currently providing?
- Quality analysis
- Is your information hierarchy logical from a user perspective?
- Do you have redundant, outdated, or trivial content that could be removed?
- Skills assessment
- Does your team have the skills to use AI tools effectively?
- Is training needed?
Change management matters here. Make sure AI adoption is framed as improving the quality of work, rather than replacing human judgement.
Option 3: Focus on the workflow
Rather than improving individual stages of a documentation project, examine the end-to-end pipeline. Can stages be chained together automatically, in the same way that CI/CD (continuous integration and continuous delivery) works in software development?
For example, a workflow might automatically:
- Pull content from a product changelog or ticketing system
- Draft initial documentation using an AI writing assistant
- Run a terminology and style check against your style guide
- Route the draft to a human reviewer for approval
- Publish on sign-off
This is the most ambitious option. It offers the greatest potential for productivity gains, but it also carries the most risk. Automated pipelines must include human oversight checkpoints. Speed is not worth the reputational or support cost of publishing incorrect documentation.
Option 4: Wait
Sometimes the best option is to do nothing yet, at least not publicly.
AI is changing rapidly, and a strategy built on today’s tools might look very different in six months.
AI might not be the best solution. You might be able to make improvements through non-AI automation, better authoring tools, or changes to how people do things.
A team that starts with a clear, evidence-based plan will almost always outperform one that rushed in without one.
Waiting is not the same as doing nothing. You can use the time to gather the information you will need to make a well-informed decision when the moment is right.
- Monitor the market. You can use AI to do some of the research for you.
- What are peers in similar organisations doing?
- What has worked for documentation teams that have already adopted AI?
- Audit your current processes
- You can start to document your workflows, pain points, and metrics now, so you have a solid baseline whenever you do start.
- Talk to your team. Early buy-in makes later adoption much smoother.
- What are their concerns and ideas?
Other considerations
Choosing your toolset
Before deciding on an approach, it helps to understand what AI tools are available for Technical Authors. They can range from general-purpose writing assistants (such as Claude or ChatGPT) to specialist tools with integrations built for documentation platforms.
When evaluating tools, consider:
- Whether the tool integrates with your existing authoring environment
- How the tool handles your content types, terminology, and style guide
Governance and ethics
Whichever option you choose, you should also address:
- Who is responsible if things go wrong?
- Is it clear what the AI system is doing? Are the process clear and documented?
- Are you checking AI output for unintentional bias or accessibility issues?
- Are you complying with the organisation’s AI usge policies?
- Data security and privacy
What next?
The options are not mutually exclusive. A practical starting point for most teams is to:
- Establish your baseline metrics (time, quality, cost)
- Identify one or two low-risk, high-visibility tasks where AI can save time (Option 1)
- Run a focused review of your biggest pain points (Option 2)
- Use what you learn to decide whether a workflow transformation (Option 3) is appropriate for your team
Don’t forget Cherryleaf can help you, with our AI consultancy and training services.
The teams that get the most from AI adoption are those that treat it as a continuous improvement process rather than a one-off project.
Leave a Reply