Leveraging AI to accelerate content generation

The challenge

Improving return on investment for Nordic content production

Our client has a large portfolio of well-known brands stretching across the globe and Comprend is their strategic creative partner in the Nordics. A long-standing challenge has been to grow market share in a region with smaller market potential and diverse local languages.

Investing in content creation, such as producing articles, in each Nordic language requires burdensome translation and localisation work, which means a higher cost and potentially less return on investment.

In close collaboration with the client, we embarked on an exciting pilot project where we used  AI to create a more efficient process to translate and localise articles to meet the specific needs of local audiences. Our overall objective was to accelerate production without compromising on quality or performance to influence audiences in the Nordic market.

The solution

A customised GPT to drive efficiencies and produce targeted content with precision

The key to working with AI is remembering that the quality of your output is only as good as the input you use for training your GPT. This insight led us to a three-step project plan:

1. Defining success

Doing things faster doesn’t necessarily mean doing things better. So, first, we worked together with the client to define success, looking at the following parameters:

• Speed of content production
• Quality of content
• Performance: ranking in search, time spent on site, and sales generated

2. Training the AI

To create a customised GPT that met our expectations, we started by training it on key brand materials – some of which we needed to create. This included:

• The company website
• The Nordic version of the brand tone of voice (which we also developed)
• 11 high-quality articles which we translated and localised to meet the expectations of Swedish audiences.

With the right KPIs and training materials in place we experimented with several different prompt sequences to train and align the AI on our client’s TOV, products and audience needs using a secure OpenAI Teams subscription and ChatGPT-4 model.

3. Drafting content

Once we started getting reliable and reproducible draft content, we created an article brief which meant we could input some basic information to the pre-trained AI and get a high-quality first draft for editing.

Key learnings...

1. It’s best to train the AI in the language in which you want to generate content.
2. For best results, the article brief should include an article title, desired length, SEO keywords, header structure and purpose. 
3. ChatGPT-4 consistently produced good first drafts but only between 300-500 words. Our expert copywriter would then go through and expand upon each section to increase the length and sharpen the focus of the article. 

The impact

High-quality articles at a lower cost

After creating five completely new AI articles, we could see several measurable results:

• Article quality: The new AI articles passed strict brand and scientific approval without requiring any edits for the first time ever. This saved the client a lot of time.
• Article performance: The articles are still newly launched but analytics show they perform as well if not better in rankings and stickiness than previous content.
• Efficiency: We can now reliably create a new high-quality SEO article four hours quicker than before. This means we can produce 33% more content for the same cost.
• ROI: Based on the learnings and outcomes of this project, we predict to be able to drive even greater efficiencies at scale with a 50% ROI over time.