This week the Estonian Government AI top-level meetup Kratitreff presented insights from AI-related pilot projects in governmental organisations (read more here). The main challenges relate to missing hypothesis, unrealistic targets or inadequate project teams during the POC-s or trials. We at Feelingstream share this view. Therefore, I would like to share the experience we had during 2019.
Year of pilots for Feelingstream
We have executed over 20 pilot projects with large and medium enterprises or government organisations over the last 12 months. The data volumes in the pilots have varied from 100 to 500’000 customer service calls, emails or chat/chatbot conversations. During the pilots, all customer service calls have been transcribed to textual format with Feelingstream speech-to-text models. Results are astonishing as top management will have first-ever visibility on a large scale into actual customer interactions.
Our most important lessons from 2019:
Lesson 1: Pilot is not technical validation but business value validation
Corporates often create POC-s as technical capability validation projects. Our POCs are different – we focus validating on business value replying to a simple question “what the organisation could benefit if customer service is fully visible for management level?” Therefore we work closely with business stakeholders from sales, marketing, quality and product management units.
Lesson 2: Always set a hypothesis
An organisation often sets the POC-s without a clear hypothesis. Based on our experience, working with the unclearly defined targets will not lead to a good outcome. Especially when the technology works well – then the need for showcasing clear business value is crucial. Otherwise, there is nothing as the next step for the business side to decide.
Lesson 3: Nobody knows what is really happening in customer interactions
Believe it or not, we have recognised that organisations live far from their customer frontlines, customer perception and actual customer service. After the POC organisations know what is actually happening in their customer service. It is about audio and text. If analysed on a scale the real discoveries are unhidden. We believe that organisations need to go deep into the conversations and service situations to improve their services and customer perception.
2019 has been successful for Feelingstream
Looking back to the ending year 2019, Feelingstream has made great progress. We have launched the most complete platform for customer conversational analytics for the Nordic languages. The latest achievements are:
- Speech-to-text (Automated Speech Recognition, ASR) in Estonian, Finnish, Russian, and Swedish. Read more about our Finnish ASR.
- A new feature to get a list of sales leads and contact for close the deal;
- Assign a task to win back team, when there is churn risk detected;
- Classify calls which indicate negative sentiment and take actions immediately;
- Use industry-specific audio characteristics to cluster your business signals.
Our customers – Nordea Bank in Finland and Telia in Estonia are happily gaining business value out of customer interaction analytics. Thank you to all of our customers!
These achievements were possible because of super cool team members at Feelingstream. With the team, we are building a new analysis approach to the data unused so far. Therefore there are a lot of experiments, collecting ideas and bouncing concepts with customers. I’m proud to have all you on my team!
The first sight to 2020
We code until the last minute of 2019 but continue in 2020. Still, the focus goes more to growing our sales execution capabilities and focusing on talking to more customers in the Nordic and the Baltic regions. We see our uniqueness in the market in the ability to anonymise audio data to comply with the GDPR and data security requirements. Moreover, there will be next ASR models for Danish and Norwegian languages, plus massive UX updates. But most surely we have a lot of challenging tasks and fun with our fantastic team!
Wishing you wonderful holidays!
CEO of Feelingstream