IBM has been hyping the Watson Platform for Conversational Commerce, so we decided to put it to the test. 1080bots keeps an eye on all the latest Ecommerce Chatbots to help our retail clients design the very best chatbot customer experiences. Given the amount of hype, I wanted to provide every opportunity for Watson to strut its stuff with Natural Language. So we reviewed Gywnn, the 1-800-Flowers Chatbot.
Gywn appears on the 1-800-Flowers website and works well on the mobile version of the site where we tested it. The quick input of delivery zipcode, occasion, and delivery date is a great option, easy to do on a responsive UI but quite tricky in a messaging app like Facebook Messenger or Kik.
Gywn asks what you’re looking for and provides an example, “I am looking for some flowers for my wife”. This prompt is really trying to engage in natural language conversation, perhaps to gain a richer set of terms than a site search query. However the words “I am looking for some” are probably redundant in this dialog.
I opened it with the term that upset the Facebook Messenger version of their chatbot, “Large Thanks Flower Bouquet”. The Watson Chatbot did much better, and returned a bright and inspirational carousel of Thank You bouquets. Among them was an item named “Sunset Passion”, so I typed in “Passion”, but the bot didn’t seem to understand.
I tried the word “Sunset” and the bot failed again. This made me wonder if the 1-800-Flowers product catalog has just been attributed with a set of curated terms corresponding to possible user intents. To prove this, I used the whole term “Sunset Passion” as if I had previously purchased this and was looking for the same thing. Once more the bot did not recognize the input.
At the foot of the chat window there were a set of buttons, I clicked the far right one “Products”. I was shown a carousel of the items I had seen so far, a handy feature as I browse – and it included the elusive “Sunset Passion”. I scrolled through and saw “My Pet Plant: Dog or Cat” so I typed in “My pet plant dog”. Apparently, Gywn (a chatbot) loves animals – which may be cute, but quite unhelpful.
I give up on this search and try a different tack – what might the bot select if I was buying for a man? The bot suggests flowers for 18 and over. I tried again “Man’s birthday small arrangement”. Gywn is not keen to send flowers to a man, so she offers up some meat – so I checked to make sure I didn’t say ‘Cave Man’. Nope.
Another unsuccessful attempt at “Flowers for men” and then I moved on to try to describe what a florist might suggest for a man. “Flowers dark colors lots of greenery” hoping to give the 1-800-Flowers Chatbot enough words to define the intent. But it’s looking very keyword-ish when Gywn comes back with “Dark chocolate is so delicious and has health benefits”. Too much conversation, not a lot of commerce right now. I tried something simpler “Brown and Green Bouquet” Gywn comes back with a very bright and not very masculine set of options. There are shades of green: Kelly or Mint – but why refine green when the more significant “brown” was not understood.
The selection of colors was clearly baffling to Gywn Watson (reminding me a little of the Sephora bot and the red lips). I typed “Flowers, Blue Bouquet” returned white flowers. Maybe Gywn Watson was trained against a floral encyclopaedia which seems sensible, so I added a flower I know to be blue the “Cornflower”. Bingo – sort of, because finally, we’re getting somewhere as I see a white and blue bouquet on the carousel.
I kept on the trail of blue and white until the 1-800-flowers chatbot realized I did not select flowers and offered me chocolates. This was a great move, because the 1-800-Flowers Chatbot is getting closer to detecting ‘intent’ or in this case, lack of intent. I jump from blue and white to “white roses, small bouquet” and it asks the next question about the occasion.
I give up on the White Roses and try to look for some thematic flowers for Thanksgiving. Surely there are some bouquets specially created for Thanksgiving gifts. In fact, I even visited their website to check and found this page: 1-800-Flowers/ThanksgivingFlowers . However the 1-800-Flowers Chatbot couldn’t understand the concept even though I really tried to help it.
Trying another idea – “Dinner Party Gift” – for some reason it did understand and gave me some suggestions. I noticed a summary tab on the bottom menu and I clicked on that to see all of the search terms I’d previously used. Perhaps I’ve built up such a mess of search criteria that there was nothing matching. So I removed some and got a better match. I wanted to refine this, so typed “Autumn Colors”.
Although it did not recognize “Autumn” it did recognize “Fall” – surely this is simple stuff to correct in an NLP engine that has won Jeopardy. We proceeded to find a matching bouquet that I decided was going to my “Uncle”.
And finally, I was sent a link to the 1-800-flowers mobile shopping cart, where I was able to complete the purchase. Gwyn was interested enough to ask me for some feedback, but she wasn’t smart enough to see the arduous trail she had me follow to get there.
So as I keep saying, Chatbots are coming, but right now, the really good ones seem to be a long way off. At 1080bots, we’re not focused on the technology of the chatbot, but we are focused on creating convenient and useful customer experiences.
Some thoughts on where this service could be improved:
To help customers shop the retailers store with the minimum of friction. If you’re thinking of building a chatbot to help your customers, drop us a line at email@example.com
As Gywyn seemed to have so much trouble with the term ‘Thanksgiving”, I wanted to see what keywords the page was able to generate. So I went to the IBM Watson Developer Cloud page for Alchemy Language.
Here you can provide a URL that the Alchemy Service can analyze. I then typed the 1-800-Flowers page of Thanksgiving Products:
The primary term extracted for that page was “Thanksgiving” with an 88% relevance, positive sentiment and of type “Holiday”. And there on the page was my old friend, the Sunset Passion bouquet.