/ Engagement

The Science of Bots and the Art of Automation

An uncomfortable truth - customers have a choice!

Consider three barriers you must hurdle to make a sale:

  • The need is big enough to warrant a purchase
  • Your solution meets the need in the customer's eyes
  • The customer chooses to both engage and buy from you

The buyer's circumstances affect how hard it is to jump each hurdle. The need may be relatively simple or have several elements.

Buy now, later or not at all?

For example, the decision to buy fuel at the net filling station changes not just on the amount left in the tank but several other factors. Distance to travel, route and terain of upcoming journey even other occupants in the vehicle will determine whether a purchase is to be made now or later. My own propensity to 'fill up' is much greater heading up a mountain with my wife nagging about fuel to reach the destination!

Does your solution 'cut it'?

Similarly, fully understanding all facets of the need and connecting a solution that meets all requirements is essential. You may have the best energy saving lightbulbs on the market. However, future savings are not relevant if the purchase cost is beyond your customers current means. Spreading the purchase cost may be a critical factor in a viable solution.

Who gets the business?

Lastly, there is engagement. An emotional desire to do business with you in particular, as opposed to looing for an alternative. The desire that must be sustained through to the point of order. This element is very much overlooked in automatation.

I find the current dialogue about artificial intelligence or 'AI' far too narrow. In our experience, successful automation needs to deliver in several areas. The choice of technology and the method of engagement varies significantly based on the environment. Lets examine a few of the key decisions.

Technology factors: ChatBot Science

Is 'probably' good enough?

There are really only two ways to select what offer to make:

  1. Take lots of historical data and compare that with the current situation and predict... making an informed guess about requirement
  2. 'Learn' enough about the situation, dependencies and options to be able to reliably work out what the best available choice is.

And the rub: the first search method only works for high volume environments, where there is sufficient data to feed the statistical calculations. Brutally inaccurate where there is insufficient training data available - in particular the 'long tail' where higher margin products typically sit.

As with so many decisions it is a case of factoring:

  • how likely is a failure
  • whats the impact

Both routes are equally described as Artificial Intelligence.

How to converse

Should you ask open questions? This is often a psychological trick. Customers perceive their requirements as special, and offering a free-form method of collection suggests that responses are bespoke.

But be warned, natural language processing is cumbersome. Even if conversation is restricted to a specific 'domain' or series of topics (the vendors product range). The alternative is to take more control and offer prompts. The good things with this are:

  • Selecting prompts also gives buyers guidance
  • Answers are unambiguous
  • Responding only requires one click

The outcome is faster progression through the dialogue. Further more, providing suggestions forms a reward by sharing information and ideas the customer may not have actually thought of.


Does pretending to be human improve results? It depends.

The more human the bot is positioned - the name, avatar, language choice and response delays etc, the greater the expectation of human like behaviour. Showing empathy, interpreting indirect language such as irony or metaphor, these are expectations created by humanisation. They are hard to handle and the price of failure is to discredit the bot. It is also a big risk to take.

Art of Automation

The art of automation is about engagement. Getting customers to use your technology to make purchases. Engagement contains several parts:

  1. Making contact
  2. Exchanging information
  3. Staying the course

The practice of making contact are common with many other aspects of design. Position, imagery, movement, sound etc. There are many examples to draw from particularly from manned chatplugins which are now established technology.

When exchanging information, aside from technology, some important guidelines are:

  • Only ask for information that you must get - use what you already have such as browsing path etc
  • Ask real world questions that devide the population into broadly equal sets - 4 or 5 areas ideally with 20-25% in each, order questions efficiently
  • Accept 'dont know' as an answer
  • Recap responses regularly to keep the customer on track
  • Use the variety of media available, never be afraid to signpost to existing content on your website
  • Customers probably know what hey want to achieve, but may not know what to buy, or how to express what they want. Be ready to help

When staying the course, nothing is more frustrating than an abandoned conversation or basket. Some principles to keep in mind:

  • The customer is visiting with a purpose. Demonstrate progression towards the goal
  • Acknowledge what you hear, summarise back
  • Re-assure by showing expertise and objectivity in advice
  • Manage expectations: are there 100 questions to come or only 1

Pulling it all together

Picking the right combination of technology and approach dictates the likelihood of success. Expert help is invaluable.

Here at AMO, we are specialist in automating sales in the business to business (B2B) environment. In our experience, the precision of prompt based interaction performs better than the novelty of natural language exchange. Similarly, probability based solutions lack the precision required when the buyer is accountable for achieving the right purchase. Obtaining and retaining engagement remains the single most important factor in design and execution.

If you would like to explore how automation can apply to your particular requirement: please get in touch.