The VoC market is shifting and most leading survey platforms are either scrambling to integrate text analytics into their solution or have done so already. We’ve prepared a blog post series that provides guidance for survey platforms or other feedback collectors on how to choose the right method to integrate a text analytics solution in your VoC platform. Finding a text analytics solution is not enough, it needs to fit with market requirements, which are becoming increasingly sophisticated.
In this blog post we will be exploring:
What customer expectations your solution needs to satisfy;
What building text analytics in-house entails;
Key questions to consider before making a decision.
Customer expectations demand an advanced solution
For years, survey platforms have spoiled their customers with advanced dashboards that present data in real time, finely tuned to their customer’s needs. However, the data displayed tends to be structured. Valuable open-ended survey answers are left to be perused manually, and therefore only partially when the amount of feedback is too great. The market now demands that advanced reporting is provided for free-form customer feedback too: accessible, easy to use and in real time. Furthermore, the market perceives text analytics as an add-on to surveys, so the text analytics price must be aligned with the core product price.
There are also some important internal clients to think about. Normally a project for building text analytics in-house is led by the CTO of the company or in larger enterprises by a senior product manager. However, before a decision is made, the requirements of other stakeholders need to be carefully considered for the project to be successful. Executives need to be assured that the project will be profitable and that it will strengthen the business model in alignment with the overall strategy. Sales leadership within the company usually demand fast time-to-market, they want a solution that has the “wow factor” such as having the capacity to do demos with client data and, most importantly, a new story to tell their customers.
Building in-house, a significant undertaking
The first step towards assessing if building your own text analytics solution is understanding the scope of the project. In broad terms, the solution will need to have two components: The AI component that does the analysis, and the front-end interface that users will interact with.
Building the AI core
The AI in a text analytics solution needs to undergo several complex steps to perform well, each with its own error potential. These steps are littered with fundamental NLP problems that top experts are still struggling with. Here’s a taster: different words have the same meaning, so your system will need to make decisions on how to group together the same customer opinions, regardless of the way they were expressed. Your MVP might perform well at first, but might perform a lot worse once it’s exposed to the challenges of understanding diverse expressions, with grammar or spelling mistakes, slang, new words, different languages and so on.
Building the interface
Once your AI solution has demonstrated an acceptable level of performance, it needs to be made accessible to the end user through an interface. Presenting data from free-form text is much more challenging than structured data. Customers expect a versatile solution that can adapt the results and visualizations to a wide range of use cases. They need to access both a comprehensible overview and reference the comments at a granular level.
Survey platforms tend to have clients across all major industries, which means that the vocabulary is different from one industry to another. If the text analytics solution implemented works based on libraries of predetermined keywords, each industry vertical needs to be added manually. This is a long and painfully manual process where all variations need to be fed into the system. These libraries need to be maintained periodically because new words are used, especially when it comes to younger generations that constantly invent new slang. Even after investing a lot of effort into maintenance, such solutions still miss words that are misspelled or unexpected issues. This can be a serious shortcoming if the clients are interested in detecting emerging trends or impending dangers they can’t anticipate.
Key Questions you should ask yourself
In order to assess if this project is worth it, it’s important to ask the following questions. Going through this thought process will help to decide the best course of action for you.
Do I have the expertise already, or will I need to build it up?
Survey platforms usually have advanced expertise in VoC feedback collection, data analysts, development and support teams. However, they are likely to miss some vital know-how that is crucial to develop a future-proof solution such as natural language processing engineers and artificial intelligence experts. The availability of open-source components might induce a wrong sense of ease. However, these components usually need to be adapted and integrated, which still requires specialists. Furthermore, experts in the field are in short supply. Even market-leading companies have trouble recruiting in this area, bringing recruitment time and cost way up. Relying on consultants will put you at risk, because of the need for ongoing maintenance.
Is text analytics part of my core business?
Undertaking a project of this size and assuming the risks associated with it is ultimately only truly worth it if the text analytics will be a vital part of your core business and long-term vision. You need to ask yourself if text analytics will become your main value prop, could you possibly sell it by itself? Most survey platforms are using text analytics as an add-on to their existing service, which does not constitute a core business requirement and is easily covered by a partnership.
Can I afford to wait?
If all goes well, you should have a solution ready within 1–3 years. Is this horizon acceptable for the Sales Executive in your company? There’s also the possibility that once the solution is complete it does not perform as expected. The AI field is notorious for having unpredictable results. In general, the market expects an accuracy of 90% and over which is a high performance to achieve. It’s also highly likely that in 1–3 years the market will shift. User expectations are hard to predict especially when it comes to rapidly evolving technology.
What else could I do with the resources this project will absorb?
You’re confident that you will manage to secure the executive buy-in, the resources, and support for your project. There’s only one last question to consider: What else could I do with these resources? Perhaps work on the core competencies of your product, build new integration, expand on your localization projects or add those sexy new features you’re sure will make your product stand out in a crowded marketplace.
Could a partnership satisfy your needs better?
If text analytics does not constitute your technological core, a partnership will eliminate a lot of risks. Depending on the solution you choose, it’s possible to see results right away. This will enable you to quickly validate with your customers base if they are interested in this kind of add-on to your services. You can start selling right away, which will make a lot of your stakeholders happy. Furthermore, you can rest assured that specialists are working on keeping your text analytics solution up to date without you losing sleep over it.
Consider the alternative: technological partnership
In the next blog posts, we will be exploring integration options, where we will be learning more about the differences between text analytics solutions and what to look out for when choosing the right one for your company.
Meanwhile, if you would like to find out more about the Keatext partnership program, click above to download more information. Our next-generation text analytics solution does not require any configuration, industry customization to set up, and it’s easy enough for strategic managers to use without specialist support.