No Code Chatbots and why you should machine generate relevant Utterances

Let's get one thing straight, actual customer data, at scale, with spelling mistakes and huge variance, is the best training data if you can get your hands on it and have the skills to curate it.

Many clients we have worked with have struggled to get real, cleaned data quickly, even within their own organisation. This struggle can be due to how new the project is, privacy policies, internal security policies, complex requisition and handling process. The most significant of these issues is private data embedded in customer text that needs cleaning or makes it unusable.

When you are looking for rapid warm-up data to get a chatbot going, it's essential to have non-overlapping intents and a lot of utterances so that when you show budget stakeholders your inspiring conversational automation, it works and is impressive enough to secure next-stage funding.

After you have exhausted your team's internal generation skills and remember you may be limiting the seed utterance diversity doing that, then next on the block come consultants and crowdsourcing, both of which can work well.

Typical crowdsource utterance generation takes anywhere from 10 days upwards to push through. Drafting scope and budgets generally are £500 upwards per 'job' at the economical end of the scale, but costs can then spiral with project management, QA, labelling and other services. All of this while onboarding a new vendor organising the budget, and finding a working relationship that truly is productive.

An alternative is Consultants, who may already be working with your company; however, they are not cheap (£750 per day upwards), and they may want to do some background research about the use case, draft a scope, meet up…. and you get the picture; days of billing may roll by before you get any data.

With many chatbot platforms being low-code or even no-code, the days of leaning on IT to create a service are reducing. There will always be a market for large scale IT implementations using technology from IBM (Watson), Microsoft (bot framework), Amazon (AWS/Lex) or Rasa. Still, most projects do (and should start with quick iteration platforms.

In a low-code or no-code environment, there is tremendous power and opportunity to make an impact, and with it comes the even greater need to train your chatbot on a self-contained basis. LevelFish provides Utterance variance generation at scale and is the type of service you can access on your credit card to get cracking with lots of data.

LevelFish can provide near-instant training data sets - so that you can press on with your voyage of discovery and prove that a problem can be automated fast. If you have a Chatbot, you can even integrate our Paraphrase API from our cloud service to give your customers a more quickly, cheaper, and more effective training inflow. Complete the contact form below or email us at sales@levelfish.com