April 16th 2018, posted by

In conversation with Margot: Lidl’s very own winebot

I’ve been chatting to Margot about all things wine. Lidl’s chatbot will help you pick out the right wine, whatever you’re planning to cook or spend.

So, how does Margot come across? Is she more suave sommelier or scripted cyborg? Here’s what the words she uses reveal.

She’s sophisticated, not snooty
Before you even start to chat, ‘Margot’ brings the right association to mind. She sounds just like the classy next door neighbour you could ask for advice on how to wine and dine that special someone.

When it comes to food pairing, Margot’s no stranger to pushing the boat out: “Personally I like to pair luxury with luxury, so if you’re having lobster, get the Champagne out.” But if flipping burgers is more your style, she can be more laid back: “Why not try a flavoursome Aussie red?”

Margot has the rich vocabulary you’d expect from a sommelier – but she won’t try to outwit you with wine jargon. She uses everyday language to help you choose: “Full-bodied reds with spicy notes are delicious to match the heat of a chilli con carne.”

She can be chatty and charming
It’s easy to think you’re chatting to a real person because Margot makes it all about ‘you’ and ‘I’. She’s a natural when it comes to using personal pronouns and contractions: “After cooking a roast turkey and all the trimmings, you’ll be ready for a glass of wine. I like Chardonnay and red Bordeaux.”

She’s also got a pretty impressive repertoire of idioms. According to Margot, Shiraz tends to “pack a punch” and chicken Kiev is “screaming out for a flavoursome white”. It’s this kind of language that makes for lively conversation.

Margot likes a little alliteration too. Paella’s a “treat for the taste buds” and “volcanic vindaloo” needs something off-dry. And for a bot, she’s pretty witty. Ask her what goes with oysters and you’ll get: “Don’t be ‘shellfish’ with the wine when eating shellfish.”

She knows when you’ve had enough
Margot’s a pretty convincing conversationalist, but she keeps things suitably concise, too. Her recommendations rarely spill over three lines – the magic number for engaging customers on a chat.

Her sentences are short and crisp, which makes it super easy to understand her in real time. She throws in the odd “Mmmm…”, “Oooooh” and “Hmm” here and there too, which switches up the pace nicely.

I ran a few of her chats through our readability test, which gives a Flesch Kincaid reading score out of 100. The higher the score, the easier it is to read. A good rule of thumb for chatbots is between 60 and 70. Margot scored around 78 – that’s an effortless read (even if you’d had a few).

But she’ll quickly switch from friendly to functional
You can also ask Margot to help you ‘find a wine’ by country, region, grape, colour or price. This is when you’ll notice a distinct shift in her tone. She starts by asking you to ‘send’ recommendations, which doesn’t sound like a free-flowing conversation at all.

The way she churns out these recommendations comes across as system generated:

Lidl winebot Margot search results Spanish wines

She shows she’s actively listening by blurting a list back to us – definitely more robot than real person. And if you’re on the hunt for a number of different wines, “Here are some suggestions…” becomes pretty repetitive patter.

Margot can also sound a little prickly if Lidl doesn’t stock your tipple of choice:

Napa Valley search term

Lidl winebot chatbot Margot search no results

She puts things down to bad luck by opening with ‘Unfortunately’ – which quickly strikes a defensive tone. And describing the criteria as ‘precise’ means she puts the problem back on the customer – she implies you’re being too picky.

And at her limits, she really is robotic
I was so impressed with Margot’s knowledge of world food, I thought I’d try and catch her out. So I picked her brains for some more unusual food pairings:

Winebot chatbot search world foods

Nobody would expect her to recognise every dish under the sun. But it’s a shame she turns parrot-fashion on us when she’s stumped for ideas. A real foodie might say something like: “Hmmm… that’s yet to tickle my taste buds.”

If you try leading Margot off-script and ask what her favourite wine is, she’ll say: “Sorry I missed that. What did you mean?” Asking an open question here is unlikely to steer the conversation back on track. Bots are better off offering options or asking yes / no questions, to avoid ambiguity.

And when something technical goes wrong, there’s no mistaking Margot’s a robot. She goes into meltdown mode: “Error: A server-side error was thrown.” That sounds very out of character for our new fizz-loving friend. Surely she’d just tell us she’s “feeling a bit squiffy”?

Our verdict: All that fizz makes Margot fickle
The variety of natural language Margot recognises is really impressive. And she’s certainly fun and friendly when it comes to food pairing. But depending where else you steer the conversation, you’ll see a few mood swings.

© The First Word 2018

September 1st 2020, posted by

Learning to love webinars

Read More

February 4th 2020, posted by

Customer service clichés to kill for 2020

What better time for a declutter than the start of a new decade?


Read More