Many software companies put all their expertise, time, and passion into building great open-source tools with a wide range of practical applications. And although their tool provides cutting-edge technology, software organizations may struggle to communicate how their product solves users’ pain points — and why users should give the tool a try.
Such was the challenge that deepset faced back in 2021. As the makers behind the popular natural language processing (NLP) framework Haystack, the deepset team was working towards a new product, a managed platform for implementing end-to-end NLP systems. Though they had a loyal following of NLP enthusiasts and experienced developers, they were trying to figure out how to reach a wider audience, both for their existing open-source product and for the software suite to come.
After more than a year of working with us, the team at deepset has managed to update its blog with high-level, accessible, and educational content about their framework and modern language technologies in general. This translates into tangible results: since the beginning of the project, deepset has been able to increase its organic traffic for highly technical, niche terms by 800% and position itself as a thought leader in the natural language processing (NLP) space. Plus, the project’s visibility on GitHub has increased significantly — as evidenced by a 400% growth in stars.
For this case study, we sat down with Milos Rusic, deepset’s CEO. Milos talked us through how working with us helped deepset attract developers to their open-source product, generate traction in search for key topics for their area of expertise, and build a solid base for demand generation efforts.
deepset: bringing NLP to every application
Even though the term natural language processing (NLP) isn’t as well-known as the more buzzworthy artificial intelligence (AI), we all benefit from its advances on a daily basis. deepset is an applied-NLP startup based in Europe and backed by leading investors like GV (Google Ventures). The company is determined to bring semantic natural language understanding — the basis for a product like Google’s search engine — to every organization managing larger amounts of text.
Using deepset’s open-source framework, Haystack, everyone can build a production-grade NLP system on top of their own textual database — with the option of adding lots of useful functionality, like semantic search, question answering, automated summaries, translation, and much, much more. For instance, Airbus has used the framework to implement an intuitive search interface for its database of highly technical manuals. Rather than leafing through thousands of pages, pilots and maintenance workers can query the new system in natural language, and receive the right text passage within seconds.
Earlier this year, deepset added a managed solution to its portfolio in the form of deepset Cloud: a platform for building NLP-based features into software applications using a browser-based interface, with built-in capabilities for faster iterations through the NLP lifecycle based on user feedback and evaluation.
The challenge: creating regular, diverse content for a diverse audience
When we started working with deepset, they already had a blog on Medium, where they shared well-written, in-depth articles about their open-source framework. However, that content was mainly targeting a very specific group of people: NLP enthusiasts who wanted to learn more about the details of the Haystack framework and the nitty-gritty of implementing open-source NLP products.
What their existing content didn’t do was help deepset increase the visibility of its products, because most of it was too technical for people just beginning to wrap their head around the complexities and possibilities of modern-day NLP products. But the strength of the framework lies in its ease of use, which makes it ideal for a whole range of people to successfully implement their own NLP-powered projects — not just experts.
In the course of our research, we found that the existing articles weren’t ranking for many relevant SEO terms, and there was a lack of clarity about which terms deepset was trying to target in order to reach potential users.
To summarize, deepset wanted to move from a highly technical, product-focused blog to a growth engine for their company. They wanted to create a website that their core audience would find helpful to learn about all things NLP, and that would motivate prospective users to visit their homepage, join their NLP community, and use their OSS framework (and, later on, their commercial product).
Devising a content strategy for deepset
To be able to promote the value of deepset’s products, we had to understand them in depth. Part of that included diving into their product and building our own question answering application using Haystack — it’s great fun, and we recommend you try it out if you want to experience some “NLP magic” yourself. (In the process, however, we were able to point out some ambiguous sections in their documentation — which we later helped amend, separately from the blog work.)
Aside from talking to their team and experimenting with their framework, we also conducted our own market analysis and came up with broader topics to target the top of the funnel and draw potential new users to their product. These are folks who might be interested in making NLP work for their organization, but feel intimidated by the complexity of the topic. We uncovered keywords that were specific enough to deepset’s core persona, the “NLP-savvy developer,” yet had sufficient potential for getting organic traffic. We then came up with article ideas that put deepset’s unique angle and industry knowledge at the forefront.
Our team and deepset’s were united in the conviction that rather than using buzzwords like “AI,” it is much more useful and empowering to explain the concepts behind those terms in language suitable for their target audience — not too complex, yet not so basic as to obscure the actual workings of those cutting-edge technologies.
How we put the strategy into practice
After we had decided on the topics that we wanted to write about — following our strategy of having a mix to target readers at the top, middle, and bottom of the funnel — we applied our technical content creation workflow. This consists of cycles of outlines, drafts, and reviews that produce helpful, detailed articles, which are fun to read and transmit the client’s unique knowledge and expertise in the space.
Of course, text isn’t the only means to communicate complex technical topics to a wide audience, and so we made good use of technical illustrations as a visual aid in our articles and landing pages. Our illustrator collaborated closely with the writers to produce informative, visually appealing diagrams to explain technical concepts like NLP pipelines, model distillation, and information extraction.
To meet the regular publishing cadence deepset was looking for, we maintained a dedicated communication channel on Slack for their team and ours. There, we would share our outlines, drafts, and sketches, and remind them about the things we needed them to do before publication. Our project manager sent regular updates and reminders, and we reviewed the results of our work on a monthly call.
Evaluating the results
After working together for more than a year, we’ve managed to create the kind of blog that deepset was looking for: a solid foundation of content that targets both NLP experts and novices. Those who already have a good understanding can learn more about specific product features, and those just starting out can find in-depth introductory articles. Many of the articles we created — for example, a comprehensive article about the technique of text vectorization — routinely show up at the top of results of relevant Google searches. Through pieces like these, we’ve been able to increase organic visits per month to deepset’s site by about 800%.
We found that deepset was the ideal partner, because they were quick to answer all of our questions about their product, and gave us access to subject-matter experts whenever we needed it. This is essential when you’re operating with such cutting-edge technology that is, at times, not even properly documented.
Compared to pre-Wizard on Demand days, deepset is now ranking for a number of relevant non-brand keywords. This means that they are able to attract new customers who have never heard of their particular product before but who have a specific problem that can be solved through open-source NLP. In this time frame, deepset has also seen a rise in inbound inquiries and increased activity in their community channels on Slack and, later, Discord. Their greater visibility has also brought about an increase in the quality and rate of collaboration within the Haystack repository, which is reflected by their increasing star count on GitHub.
As well as blog articles and landing pages, we’ve also produced two ebooks for deepset. The first ebook, “NLP for Product Managers,” explains the details of modern NLP development to a non-technical audience. deepset has seen great success distributing it to existing users and prospects, and is using it as a demand-generation tool on the website. Through this and the next ebook, “NLP for Developers,” deepset has been able to reach its target personas and introduce them to its product in a thoughtful and friendly manner, by educating them about the technology behind NLP successes — and by showing them how they, too, can build the same.
Working together on a range of different projects, we at Wizard on Demand have found a way to communicate complex NLP concepts accessibly — and in a language that greatly assists deepset’s product marketing efforts. We did this by listening to the deepset team, who in turn listens very closely to its users. These days, deepset doesn’t have to worry about the content production anymore because they trust our process.
Wizard on Demand's holistic approach to content production has helped us build a solid foundation for future marketing and sales efforts. We're very happy about our partnership, and quite proud of the results.
-Milos Rusic, CEO at deepset
We’re continuing to diversify our content portfolio with deepset in order to drive the next stage of its growth. In the coming months, we’re looking to promote its most recent product, deepset Cloud, more actively: following our established strategy, we’ll get qualified leads for deepset Cloud to help the company grow its commercial offering and increase revenue.
At the same time, we’ll continue working on technology marketing, to enable users who have problems that can be solved with applied NLP to find the right solution. By explaining industry best practices and sharing useful opinions, we’ll create more content to illustrate how the new language technologies that deepset leverages can solve many pain points.
The field of NLP is growing constantly — and we’re honored to play a small part in it through our work with deepset.