Expertflow does not have it’s own Chatbot or Conversational IVR (Voicebot) product. We frequently implement such solutions however, and often get asked which product we work with, and how we compare to this or that alternative. The answer is - it depends, and often we don’t know at the beginning of a project what solution we’ll end up with. In most cases, we don’t use only one NLU or Dialogue engine or speech recognition engine in one and the same deployment. We see existing products more as a toolbox with different tools that apply for one specific scope. In some cases - for example CAPS DETECTION FOR SOMEBODY WRITING IN CAPS ALONGSIDE insult detection, we write our own code. Or we combine emotion detection in speech with speech recognition.
Selection criteria typically include:
- Shall the solution run on-prem or is cloud ok also?
- Use case for detection of speech or intents/ entities:
- Regular expressions (alphanumeric bank account numbers,…)
- Toxic language detection/ Insults
- Structured data gathering (= forms) or free
- company and location specific terminology
- language and dialect
- Use case/ Industry-specific terminology (ex. medication, Small talk,…)
- availability of or capability of the client to provide bespoken trained data sets
- Familiarity/ Availability of the client of AI (Tensorflow, KERAS, GPU, Dialogflow, IBM Watson,…) and DevOps (Kubernetes,…)
We haven’t seen any of the large vendors able to systematically address all these areas in all markets that we serve. This emphasizes the importance to remain flexible as for the choice of AI engines. What remains and what we urge clients to focus on is training data: Whatever AI engine(s) are going to be used, they all feed on training data (tagged speech or tagged chat). The more, the better.