What’s Twitter got to do with consumer protection?

September 4th, 2019

Social media is changing customer service by shifting the ways in which consumers seek resolution of problems and the communications channels that firms make available to consumers. The ability for consumers to directly, instantly, and publicly praise or chastise the service of a firm has led to increased accountability and new ways to remotely resolve customer issues swiftly.

In Kenya, the “#KOT”, or the “Kenyans on Twitter” community, has become an important channel for raising issues with financial products. This is often borne out of a frustration that traditional channels like the regulator or a bank’s customer care hotline have not given satisfactory complaints resolution.


But Twitter can be more than just a place for venting frustration. It can be a place to get complaints resolved for individual consumers, and to flag emerging consumer protection issues across the financial sector. As more and more Kenyans share their issues with financial services on social media, a new evidence base is being created that can help providers and government to better measure, monitor and resolve consumer protection problems. All we need is a way to make sense of the data…

From August through November 2018, FSD Kenya and Princeton University partnered with CitiBeats, an Artificial Intelligence (AI) company that collects and analyses textual data such as Twitter messages, to track and analyse consumer protection-relevant tweets directed to 29 different Kenyan financial institutions. Using keywords and machine learning we tracked tweets across 10 different consumer protection categories (see Table 1 below).

Table 1. Consumer protection categories monitored through the Kenyan Twittersphere
Category Category description
1. Agents Monitoring of the behavior of banking agents who conduct different services on behalf of banks such as funds transfers, bills payment, cash withdrawal, cash deposits and balance inquiries.
2. Blacklisting Refers to the word often used to describe negative loan repayment data in credit histories. This topic included that and other related discussions of credit history and Kenyan Credit Reference Bureaus (CRBs).
3. Frozen account Tweets regarding the inability of a consumer to access their accounts and conduct transactions.
4. Functionality Consumer concerns related to the performance of financial technologies such as mobile banking, apps and websites.
5. Insurance claims Requests to insurance companies to compensate for loss or occurrence in relation to the loan terms.
6. Loans A wide range of topics related to loans such as hidden interest rates, being denied loans, or delays in money remitting to borrowers’ accounts.
7. Privacy Primarily concerns regarding how private consumer data is collected and the ways service providers share consumer data with third parties.
8. Resolution/ customer service How banks communicate and serve their customers, both physically in branches, and through nonphysical ways such as calls and social media platforms. This topic received the most Twitter traffic of all 10 topics.
9. Scams Illegal or unauthorised transactions that take place without consumers’ knowledge and consent. Unauthorised debits on bank customer accounts was a concerning problem discovered in this three-month pilot.
10. Charges This includes costs to consumers by financial service providers for provision of services such as account maintenance and transaction fees.

To complement the data analysis, we also encouraged a public dialogue on consumer protection through the creation of a Twitter handle, @pesastory, which provoked discussions on consumer protection and consumer experience with financial services in Kenya. The @pesastory experience is discussed in a forthcoming blog.

What Kenyans tweet about on consumer protection

Our three-month pilot demonstrated the value that social media data can have for financial consumer protection monitoring and enforcement. First, we were able to sort and measure the volume and type of consumer protection issues. An analysis of tweets sent to 29 different financial service providers demonstrated that most consumer concerns related to customer service or how products were working. However, some providers do see disproportionate percentages of tweets on other consumer protection categories—such as “Charges” for the Mobile Network Operators (MNOs), or “Loans” for the digital lenders, in our sample.


Figure 1

An early warning system for serious consumer protection concerns?

Beyond measuring how many tweets were directed to which providers, we wanted to use Twitter to identify problems as they arise. We developed a daily and weekly alert system to make it easier for the research team to identify spikes in consumer protection issues with providers. The daily alerts would be sent when they matched two criteria:

  • The number of tweets in a category today must be at least 10% higher than the average number of tweets in that category for the last 30 days for that provider.
  • There must also be a minimum of 10 tweets in the category, to protect against small increases in very small sample sized (e.g. 3 tweets in a day when the average is 1 per day.)

Figure 2

The daily alert system would send an email to the research team whenever there was an increase in tweets to a provider about a particular topic. This alert tool allowed us to flag larger than normal volumes of complaints relative to each provider’s normal chat levels—helping to control for relative size of firms’ Twitter presence. For example, we received an alert of a spike on the topic of Customer Service on November 2, 2018, with Equity Bank. Using the CitiBeats platform, we can see the spike on November 2 in Equity Bank’s daily tweet records (See figure 2.) By selecting the date and topic, we then are told there were 65 tweets sent to Equity Bank on November 2 regarding customer service issues. Finally, we can then read these Tweets (Figure 3) to investigate the issues raised by consumers, and determine if there is a need for follow-up with either the customers or the bank. In this case, we notice from the tweets that there are recurring issues with deposits and transfers on mobile banking platforms.


Figure 3

Some of the other noteworthy daily alerts for November are summarised in Table 2 below.

Topic Provider Relevance
Customer Service NIC Bank An influential Twitter account raised a complaint about long waiting time at branches, prompting other customers to share their experiences with this issue.
Charges Safaricom A new feature on “Lipa na M-Pesa” was introduced that eliminates the need to share phone number with merchants, improving data privacy and subsequently receiving significant coverage and praise.
Charges Airtel Complaints regarding disappearing data and the need for a data manager function to manage use of bundles.
Loans Safaricom Customers complaining about loan balances that belong to another person being deducted when they send that person airtime via M-Pesa. (Two alerts generated on this topic)
Charges Equity Bank An influential Twitter account shared that Equity Bank refunds not just the transaction amount, but also the transaction fee, for wrong transactions in their mobile banking. This was praised as it differs from the practices of other providers.
Agents Safaricom Consumer raising concerns about a till number being used by police in Nairobi to solicit bribes.
Customer service Cooperative Bank Several customers raising unresolved problems that have not been fixed yet.
Functionality Commercial Bank of Africa Poor customer care response on their app and double debiting on the app that was not reversed.
Resolution / Scams Imperial Bank Customers raising issues related to the resolution of the failed bank.
Functionality Diamond Trust Bank Charges related to a multi-currency card

The weekly alert system would share the five provider/consumer protection categories that had the most significant increase in traffic over the past week, so that even when there were less than 10% triggers on daily levels, we could still see what the most active conversations were for that week.

Testing the alert system required some modifications during our month-long trial. For example, we quickly learned to filter out all Twitter threads originating with the provider, as they were usually marketing messages, and would lead to false alerts. We also experimented with different threshold levels for triggering an alert before deciding on a 10% increase in volume as the trigger for a daily alert. There will always be a need to balance between comprehensiveness in reporting and the efficiency of the filter in this system, and so the settings should be periodically tested.

Manually reviewing and making sense of the tens of thousands of relevant tweets by Kenyans would have taken weeks of researcher time. But the testing of the email alerts in November 2018 demonstrates how thousands of tweets can be effectively flagged and investigated in just a few hours using an artificial intelligence platform like CitiBeats.

As our second blog in this series will demonstrate, there are latent problems that financial consumers bring to Twitter because they are not resolved to their satisfaction via other channels. By using artificial intelligence platforms such as CitiBeats, we can make sense of these conversations and prioritise the more serious and recurring consumer protection issues to efficiently allocate the time of customer care staff. For financial service providers and regulators in Kenya, social media monitoring can help make complaints handling more efficient and improve the customer care experience in financial services to weed out some of the abusive practices we have seen occur far too often in Kenya.

Read the full report here.



FSD Kenya newsletter

Stay informed with regular updates from FSD Kenya

Subscribe to our mailing list

Our partners