Choosing the right Text Analytics partner

for your VoC platform

Adding text analytics capabilities to your VoC or Survey platform is no easy feat. This is why we put together a series of blog posts to help with your buying process. In the final blog post of the series, we look at different types of vendors in the text analytics space and how to compare them. As a special bonus, at the end, you can download an editable comparison chart to help navigate through your options, including detailed buying criteria we’ve seen with our customers use.

Types of vendors

There are three types of companies often found at the crossroads between text analytics and customer experience:

  • Customer Experience Platforms. Comprehensive platforms for managing customer experience, with an integrated text analytics component;
  • API-only providers. Back-end technology for companies wishing to integrate text analytics;
  • Full Text Analytics platforms. Specialized technology companies that offer full text analytics solutions, specialized in CX.

Customer Experience Platform

These solutions are well established both in the text analytics and the customer experience market. As a VoC provider you’re probably familiar with these vendors, since there’s a high likelihood they are your competitors. However, they do offer text analytics partnerships which makes for an interesting market dynamic.

Most of the leaders in this space use keyword-based technology. This means that if you choose to integrate with a company of this type, it’s important to plan time for set-up—the definition and configuration of your analysis—for each of the verticals your current clients are in, and any you plan to target. The vendor will help you to define and configure the dimensions of your analysis, taking into account industry-specific terms and words, and jargon related to your customers’ products or services, and provide support to your IT department. It can take 9-12 months to carry out. Before you start, you will need to ensure you meet the necessary infrastructure requirements. If you have to add these, you’ll need to budget for further time and costs.

Before anyone is able to use the platform, however, they will need to be trained. Training and support is often offered at extra cost. With built-in analysis tools and dashboards, trained users will find it quite easy to investigate results and identify insights and actions, and present their findings. Your agreement with the vendor will usually include an ongoing support contract, allowing you to request adjustments to your initial specifications. To avoid surprises, you might want to allow for additional set-ups in this support contract, as well as additional sources of feedback.

With regard to accuracy, keyword-based solutions are good at recall but not necessarily precision. For more information on text analytics accuracy metrics check out the explanations in the comparison sheet at the end of this article. An additional aspect to keep in mind is that once the setup has been made, any accuracy issues are hard to correct with these types of systems, but some measures can be taken. In non-traditional verticals, or if your clients’ customers don’t express themselves in the ways documented in the industry vertical libraries provided by the vendor and augmented by your specifications, accuracy can be very much reduced. Asking this type of vendor in advance about their experience in the verticals that concern you and your customers is highly recommended.

One final aspect to look out for is how the data is represented on the dashboard, unless you’re building your own. Most text analytics users expect to see the data not categorized in buckets, but in a organic way that represents how the customers express themselves. Are results are presented with their context? Is it easy to reference an individual customer story in order to understand what happened?

API-Only vendors

While API is just a method of helping different software communicate with each other, in our industry when vendors advertises Text Analytics or NLP API it usually means this company only provides the text analytics layer. With this layer alone, the initial functionality is reduced to a few basic features such as identifying key categories of topics and tag clouds.

Here is an example of Keatext’s stack (full text analytics solution) vs. a typical API-only solution:


Text analytics stack comparison: API-only vs Full solutionText Analytics layer– extract relevant information from a document.

The company choosing to go with this kind of provider will likely need to build up the remaining layers for more advanced features. Here’s an overview of the layers we built to cover a high level of use cases:

Aggregation layer Stores original metadata and analytics detected by the artificial intelligence pipeline for all the documents. This means the user can do advanced queries in a unified database across all their datasets across all communication channels, such as: “rising issues in the last week across all touch points”.

Normalization layer– Interprets data coming from the analytics layer, allowing it to detect changes over time such as trends, events or improvements. Depending on the technology used, this is where advanced features become possible, such as word grouping, or automatic metadata correlations.

Knowledge Management layer– This layer has built-in logic that serves the right information to the user when they request it. Users can create custom categories and capture specific industry and company knowledge.

Administration layer– Helps organizations manage users, access levels and keep track of their usage.

Application– Finally, an application containing a user interface needs to be built on top of the technological solutions. For a wider user base, this shouldn’t require specialist knowledge such as coding.

At first sight these API-only Text Analytics solutions seem very advantageous because of the low cost per comment rate. However, the costs and time associated with building a fully-fledged solution can be extensive and not visible at the beginning of the project. While it’s hard to estimate, from having personal experience of building such a solution, it takes about two years with a reasonably sized (6-8) dedicated development team.

Another aspect to watch out for is the accuracy level of the analysis. Vendors in this space vary from generic NLP solutions to CX specialized solutions. As a rule of thumb, if the requirement is CX, then the specialized solution should have better accuracy in this field. It’s easy to test the accuracy of these solutions since many vendors offer free trials.

Full Text Analytics Solutions

There has been a burst of innovation coming from small independent companies founded by individuals with close ties to the academic world. Their research has allowed for new solutions in the NLP space, moving away from keywords and rules, usually replaced by machine learning technology or other methods. One of the biggest challenges has been fighting the market leaders and getting the message out there. When thinking to partner with one of these companies, it’s important to do your due diligence. Do they have established customers? If they are a startup, where is their funding coming from? Do they have solid academic credibility?

The good news is that these systems are more likely to be built with the most recent technology. If your platform is also innovative and uses industry best practices, you’re likely to integrate the systems very quickly, within a few days or a couple of weeks, depending on the complexity of the project. The speed of data collation for a pure machine learning solution is very high compared to other. You can expect to be able to delve into the analysis within minutes. Then they’ll be able to explore the data, and analyze and segment it efficiently as well as effectively, using the built-in analysis and visualization tools.

If there is no setup or configuration required, any increase in volume of unstructured data from existing feedback sources is easily accommodated, as are new sources, products and markets. With a complete interface that makes it easy to interact with the results of the AI analysis, present findings and recommendations and share them, it’s the ideal self-service solution for your customers, requiring no data expertise to identify insights and actions.

In terms of accuracy it’s important to test it on typical documents as it can vary a lot. The good news is that it’s possible to retrain and adjust these systems very easily. Sample feedback data can be used to retrain the system and improve accuracy for new data sources, new verticals or products.

The biggest advantage of this kind of solution is that it’s less likely to leave the user with blind spots. Rules and keywords enable analysts to find what they’re looking for, but AI enables everyone, data expert or not, to find what they’re not looking for. At Keatext we’ve often found with our customers that if a high number of customers report a seemingly unimportant or unrelated mentions can be an excellent identifier of emerging trends or even an early warning for an impending crisis. You can rest assured, therefore, whether you offer value-added service or self-service, your customers will get the right information to act on, at the right time.

Tools to help you further

Choosing the right text analytics solution for your company is no easy feat. There are a lot of different aspects to consider. To help with your buying journey, click below to download the text analytics vendor comparison sheet.

Check out:
Bain & Co.’s blog post on what to look for in a text analytics solution. For more information of how a technology partnership could benefit you, click here.

In case you missed it the other posts in this series, I covered:


Are you interested in seeing Keatext in action?

Book a free product tour at your convenience.