With the new features now live, I thought it’d be interesting to use Kapiche to analyze customer review data for Skymesh, an Australian ISP that provides me with my own home internet.

I found 900 reviews to use for this analysis, all with unstructured text data, along with structured data: date reviewed, service type, a number of people who found the review helpful, and a 1-5 star rating. I uploaded the data to Kapiche and ran an analysis. Shameless plug: I timed how long it took to upload and process the data: 15 seconds. Let’s get to it!

Overall data

With an average star rating of 3.6, and an overall sentiment which was slightly more positive than negative, Skymesh seems to be in a good place with customers compared to the other ISP’s I browsed (many ISPs had an average star rating of 3 or below).


We can see an overwhelming majority of the reviewers had an FTTP connection.

In the top 20 concepts (high-importance terms) the language model has identified, it seems like people talk about customer service and support the most -- sounds like a good place to start digging! But before that, I wanted to quickly check the distribution of reviews:


Hmm, there’s a huge spike in January 2017 and a smaller spike in June. This looks very abnormal, so let’s get more specific and show the exact dates.


Interesting. 99 reviews on a single day. Let’s get even more specific and show only 4 and 5 star reviews.


That’s a lot of 4-5 star reviews in one day! Remembering the large majority of FTTP connections in the overall data, I added connection types to my query. This revealed 73 of the reviews on this day also had the same connection type (FTTP).

Keen to see what your customers are saying about your product? Sign up for a trial, upload your CSV file and get results in seconds!

It’s an unusual event to see so many reviews come in at the same time, but not unheard of. Back to the customer service concept:

Customer service is key


Customer service makes up a large portion of the reviews, being included in 24.7% of the data. As shown below, Skymesh do a pretty good job here, as over 60% of the reviews which include this concept are 4 or 5 stars. The strongest influencing terms (shown above) are also mostly positive: “excellent, great, superior, top notch, quality”. Overall, a decent outlook for customer support, but there’s always room to improve!


Customers happy with competitive pricing and no lock-in contracts

A query for discussions around pricing shows a large positive sentiment and correlation with 5 star ratings. We can see that Skymesh’s competitive pricing and price-match guarantee along with no-lock in contracts is significantly boosting high scoring reviews, whilst being negatively correlated (unlikely) with low scoring reviews. Good job Skymesh!

Sample verbatims for contract query

Sample verbatims for pricing query

Low-score culprits: Slow peak hour speeds, no connection, long hold times/no callback

Let’s see if we can find something Skymesh should look to improve on. I queried 1 and 2 star ratings to try and find any correlations. There are 285 1 or 2 star reviews, making up 31.67% of all the reviews.

Concept & Segment Charts for 1-2 star reviews

Firstly, the Segment Comparison chart above shows some interesting correlations with connection type. FTTN, HFC and 3g/4g Mobile Broadband are likely to be correlated with 1 and 2 star ratings.

Digging into the Concept correlations, three concept groups stand out: long hold times, weeks without connection, and slow peak hour speed. These concepts are highly correlated with 1-2 star ratings which tells me resolving/improving these issues would see a great improvement in overall rating. Reading the verbatims for these topics show the specifics of the problems (Kapiche highlights words I queried for).

Sample verbatims for slow peak speeds

Sample verbatims for no service/internet/connection for weeks

Sample verbatims for long hold times

These three issues account for around 15% of the total reviews, and 50% of reviews which gave 1 or 2 star ratings! Hypothetically, if Skymesh converted these to 4-5 star reviews, their overall score would change from 3.6 to 4.1.

There’s always more

The analysis I did took less than an hour (writing this blog was more time consuming). Digging through this data with Kapiche allowed me to find dozens of interesting things with a few clicks here and there. This blog could have been so much longer, but I’m going to self-moderate and conclude my little adventure here.

To summarize some of the insights I found today:

  • Customer service, as expected, is of great importance to customers and a strong reason Skymesh’s average rating is higher than other ISPs.
  • In addition to customer service, Skymesh’s competitive prices, price-match guarantee, and no lock-in contracts are the next contributing factors to its overall positive rating. People talked about these concepts in 15% of the reviews.
  • The main areas to improve are long hold times, extended periods of no service/connection, and internet speeds during peak hours. These issues account for 50% of low score (1-2 star) reviews. If Skymesh converted these to 4-5 star reviews, their overall score would change from 3.6 to 4.1.
  • Skymesh received 93 positive (4-5 star) reviews on the 16th June, 2017.
  • People with FTTN, 3g/4g Mobile Broadband, and HFC connections are likely to give 1 and 2 star ratings.

Note: this was not in collab with Skymesh, and I have no preference towards them. This was just a curious experiment to see what I could find using Kapiche!

With the Kapiche platform, customer insights teams are finding success with the discovery-led approach. They aren’t spending weeks manually coding datasets with hundreds of thousands of pieces of feedback or waiting for vendors to update their code frames. They are getting time back time to understand the behavior drivers and reporting actionable insights across the business. And this is happening because their text analytics solution is doing all the heavy lifting for them.

Curious to see how it work with the Kapiche platform? See this demo video.