Ivy is a 24-hour virtual concierge service that smooths the stay of guests at hotels (Raj Singh)

Part of the Bot Master Builders Series

Ivy is a 24-hour virtual concierge service that smooths the stay of guests at hotels.

Ivy is available in more than 6,000 rooms within Caesars Entertainment Las Vegas Resorts, including Caesars Palace and Planet Hollywood Hotel & Casino. Ivy can check guests in and check them out, and also check in with guests mid-stay to see if they have problems (and “execute real-time guest recovery” as hotel insiders say). Ivy helps hotels improve their guest experience while improving speed and efficiency for desk staff.

Ivy uses IBM’s Watson-enabled Natural Language Classifier API to read incoming guests’ texts and directs them to the correct hotel department. She is trained around common questions and automatically responds within seconds to common questions. Ivy frees up desk agents to focus on the more complex guest inquiries. It would be nearly impossible to handle the volume and manually text with guests for large hotels, while smaller hotels can struggle with this too. A large hotel like Caesars can keep humans in the loop with a skilled team of universal agents in a call center that text guests back when Ivy doesn’t know the answer.

Raj Singh is the Founder of GoMoment, which makes Ivy. His family background is in hotels and he taught himself how to design and code as a teen. At one point, Sheryl Crow’s digital team reached out to him, and he did interactivity and full-stack development for them. UX has always been his passion, and he went on to work with Lady Gaga, Sony BMG, Virgin and other labels. While Raj got 1 billion page views for his projects, he realized he had no feedback loop after launching projects. So Raj started GoMoment in 2013, as he wanted to create a billion unforgettable experiences.

Raj’s insight came from his family business in the hotel and resort space, which was still recovering from the 2009 crash. Their staff was leaner than ever, the minimum wage had gone up, but top-line revenue was down by a third. Staff was overloaded with tasks and customer requests. Meanwhile Uber and Lyft had started and increased expectations of customer service. A $10 Uber ride came with service, but if you did a $200 hotel you often had to use a dirty phone with a recorded message and it could be hard to get a concierge on the phone. Hotels needed to digitize and automate. Initially, Raj partnered with IBM Watson for Ivy, a hybrid human and AI, but also had to work hard to create the category and educate the market about his hybrid turnkey service.

QUESTIONS

What was the original design vision for your bot? Does it have one or two clear functions, or many?
I took the approach of “we don’t know what the market wants” — so we studied the customer behaviour. I worked in service of the guest experience and did primary research on pain points. Hotels tend to have spiky demand around check-out and check-in times; people expect 3pm lines but it’s a bad experience many times.

We have eliminated the check-in and check-out times. Ivy will proactively message the guest to do check-out via text. Everyone needs to know the Wifi code, water, and restaurants at a hotel. We have an experience and story arc around each use case. Ivy has had 10s of millions of interactions. We have exclusive data to understand guest behaviours. We even the have the W Hotel in Boston!

Ivy can also do room service; it’s an omni-channel solution with SMS mostly, though we also have Alexa and have Google Assistant. It will also interface into the hotel back-end; there are dozens of property management and reservations systems. Still the 3–4 major vendors are 70%, and then 30% is fragmented over 15–30 vendors.

Ivy is a hybrid model — automated chat most of the time, but humans as backup to try to deal with fallbacks.

How do you measure success; what are your metrics? MAUs vs typical session length and completions?
We look at success in a funnel format — we optimize engagement and reach. “Reach” is 100 guests walking in and we want all of them to have access to the bot — it’s nearly 100%. The guest has to opt in by providing their hotel number after the front desk tells them about Ivy (once per hotel). Engagement is about being helpful first, service oriented. We will track campaigns on restaurant recommendation and social media review for guest. A key metric is our resolution rate — are all requests processed in a reasonable time period? We also have a staff leaderboard for the hotel manager. We will look at session length, but hotel guests may not message initially. In terms of deployment success, we deployed to Caesar’s Palace in 30 days, and this includes scheduling and training. Most onboarding is less than 30.

We have rules and behaviour patterns so she is respectful of the guest and service oriented.

Anecdotally, people have given great feedback. We have hundreds of guests raving about IVY in TripAdvisor (people have tried to tip Ivy and even offered to marry her).

What are the successful interactions? What are failed interactions?
A successful interaction is when Ivy checks in mid-day “Are things up to your expectations for your stay, are you having a good time, is there anything I can do?” We get initial feedback, detect sentiment, and send this to a hotel (these otherwise are silent travelers who would stew on a problem — the mildly dissatisfied). We can discover a “service recovery opportunity” autonomously. Caesars said some hotels went up 30–40 ranks on TripAdvisor — hotels normally need to invest millions of dollars to this.

Failed interactions are if a guest messaged in a tough question so Ivy escalates this to a human and watches their response time. A human could take too long. We have safeguards to mitigate this — we have a customizable escalation policy: if the front desk fails, Ivy can manage with a response and escalate to higher managers at a hotel.

Team — how did your team come together, and what are the roles?
I am a sole founder. We have some hotel technology people and we have SaaS people like a customer success team. We have an inside sales team — we don’t need to physically visit the hotel. Our team came from my consultant and hotel background.

Editorial and scripting — what have you learn from flows so far? How much customization by hotel or brand?

We co-create the copy alongside each hotel — there’s not much canned copy. We have small boutique hotels to very large hotels — no standard copy makes sense. We do have best practices on what copy will perform better. We have our own messaging model and top requested queries based on the hotel type, size, location — we will try to mimic the best agent but simplify for the messaging medium. Message compression is important, and also texting regulations like TCPA. We offer white glove service for the hotel.

Our biggest learning on message compression is from guests directly: don’t present too many options. In the travel context, people are so busy they don’t want to think. Keep it simple, interpret the message as someone who doesn’t text often, an older person, or a foreign weak English speaker. Have empathy for the user. A sweet spot is to give a user only one option — it shows the most understanding for one person (one option for check-in, one to give wifi, etc, but for restaurants can be 3 to 4 options is good — 5 is pushing too many).

What are the most common things users ask outside of the main function — do they ask for jokes or other advice?
People ask Ivy to come watch Netflix and chill — people will ask how to dress for the nightclub — a group of special needs guests who are hearing impaired came to one hotel and they loved the texting

User acquisition strategy — how do people hear about your bot and start using it?
For hotel guests, we have digital signage and desk tents in the hotel- also poolside. It’s a number that anyone can text. To get hotels signed up, we have an inside sales team in terms of how the deals get approved — indy hotels may be 1–2 people, but we target a dozen people for the larger hotels (ownership, mgmt, operators, brand and flag).

Re-triggering and re-use strategy — as many bot developers know, how do you get initial users to engage

Ivy always checks in once for any engaged guest who wants messages. We have an automated check-out by text — we give them a time and email address. We can also do a daily housekeeping opt-out for hotels who wants to ask. The hotel saves money, guest saves on hassle (some hotels may want a more subtle signal).

Personality profiles — what have your learned?

Ivy does not have a single personality — it’s empathy above all, you have to be egoless. You don’t know what day the guest had or why they are in town. You cannot even say Mr. or Mrs due to gender issues. Positivity and resilience are key and each hotel picks a brand personality.

If someone attacks Ivy and says “you suck”, then Ivy hands them off to a human to defuse — “I’m sorry this happened, tell us what happened, let us help you make it right.” Ivy can be used for event crowd control and hurricane evacuation. Cornell did a survey on Ivy and got hundreds of responses from the hotel workers at the front desk. They reported that Ivy made them feel more human by getting robotic work (“what is the wifi”) off their plate, so they had less emotional strain.

Monetization? Many bots are a great free service — how have you tested monetizing it?
Our business model is to charge a hotel on a per room, per month fee basis (it depends on the hotel) — we have a set-up fee to charge for training and responses. A full-service resort with amenities would pay more versus a hotel with no restaurant.

What can you tell us about your tech stack? Do you do NLP in-house, what external services do you like?

We wanted to be agnostic with anyone AI system — we want to interface with multiple systems. We did IBM Watson for a while and will launch with Google’s DialogFlow — we also have internal NLP. You cannot have a regular chatbot and ask for leeway; you need to have the human in the loop (always hotel in the loop).

We use AWS generally and have our own RESTful API for clients’ systems. We will use a messaging service for SMS. We have an internal system with patents pending; it’s an overlay system to monitor and ensure quality service on top of everything else.

Thoughts on the different platforms? FB Messenger vs Amazon and Google or Kik? Voice platforms?

We have many hotels using Ivy with Alexa — “Alexa, tell Ivy to get me more towels” — it looks like Google and Amazon are the two heavy hitters here. Samsung’s Bixby and Yandex’s Alice are there, but we see Google and Amazon as the best positioned. Google has a multi-device advantage with Chromecast and Android, more than just a speaker in your room. Parents with young kids like to text in a room service order, can also say don’t knock but text me. One challenge is training customers on how to use a speaker, better to have visual and speaker — travelling hotel guest would prefer this.

What other bots have you looked to for inspiration — what other bots made you say “WOW”? Are there other use cases you’ve thought were simply brilliant?
We have not interacted with most bots — the best business messaging experiences were manual. It’s hard to do chat automation with this scale for customer-facing with such little room for error. Mezi was one of the better experiences — but it seemed mostly manual, like they were throwing humans at the problem.

Who are other smart people in the bot world you’ve met, whether on the tech stack, UX, scripting, or even financing sides?

The Machine Learnings newsletter by Sam DeBrule is great. We have written for their guest column.

Any lessons to share with other bot-builders on useful tools?

Murphy’s Law — anything that can go wrong will go wrong! This is true for chatbot devs and for entrepreneurs. Find the inevitable failures to turn into fuel and to get your product better. Mistakes will be made in a new field or technology — always anticipate what could go wrong. You only get one shot with your user for a good experience. You have to be very responsive to solve problems — see a small problem, and fix it and thousands of similar problems like it. Know what to expect: ask yourself, “what experience have we created?”

Read the prior articles in this series:

Edwin — Your English Language Tutor (Dmitry Alekseev and Dmitry Stavisky)

Woebot — Your AI Cognitive Behavioral Therapist: An Interview with Alison Darcy

X.AI’s Amy and Andrew Ingram (Diane Kim)

Rose the Loebner Chatbot Winner (Bruce Wilcox)

Poncho the WeatherCat bot (Greg Leuch)

Howdy and Botkit (Eric Soelzer)

Statsbot for Business Metrics (Artyom Keydunov)

Earplay: What Chatbots can Learn from Interactive Voice Games (Jon Myers)