Leena — Automating the HR Helpdesk (Adit Jain)

Part of the Bot Master Builders Series

Published in

Chatbots Magazine

7 min readOct 10, 2018

Leena AI is an AI-powered HR Assistant providing instant responses to employee queries and improving the employee experience. Leena builds HR chatbots for enterprise employees, basically automating the HR helpdesk. Leena has 8 modules across the HR lifecycle: Onboarding, FAQ, health and wellness, travel and expense, HR services and transactions, performance management, learning and development. A customer’s employees don’t have to go multiple interfaces; they only need to go to one interface and Leena does all the integrations. This summer, Leena had 12 enterprise customers on annual plans (including Coca-Cola, Viacom, Pearson, Axis Bank), with 10 doing Proofs-of-concept and proposals.

Their differentiators: Most HR technologies have been made as full-featured systems; they are a tool system or record-keeping system. Leena’s goal, in contrast, is to build for the end user, to make it really easy for employees to use it.

Leena AI, was a member of the Y Combinator Summer 2018 class, after which they raised a $2 million seed stage round. The team wants to change HR by building HR bots to answer questions for employees instantly. The bots can be integrated into Slack or Workplace by Facebook and they are built and trained using information in policy documents and by pulling data from various back-end systems like Oracle and SAP.

Adit Jain, a co-founder at Leena AI, says the company has its roots in another startup called Chatteron, which the founders started after they got out of college in India in 2015. That product helped people build their own chatbots. Jain says along the way, they discovered while doing their market research a particularly strong need in HR. Having a generic platform wasn’t as inspiring or profitable as solving a vertical need. So they started Leena AI last year to address that specific requirement.

Jain says when building bots, the team learned through its experience with Chatteron that it’s better to concentrate on a single subject because the underlying machine learning model gets better the more it’s used. “Once you create a bot, for it to really add value and be [extremely] accurate, and for it to really go deep, it takes a lot of time and effort and that can only happen through verticalization,” Jain explained.

What’s more, as the founders become more knowledgeable about the needs of HR, they learned that 80% of the questions cover similar topics, like vacation stuff, sick time, and expense reporting. They have also seen companies using similar back-end systems, so they can now build standard integrators for common applications like SAP, Oracle and NetSuite.

While many employees may ask similar questions, there may be dozens of ways to ask the same thing, or intent. Jain says that’s where the natural language processing (NLP) comes in. The system can learn these variations over time as they build a larger database of possible queries.

INTERVIEW Q&A

What was the original design vision for your bot? Does it have one or two clear functions, or many?
Leena started with HR service and transactions — the first 3 or 4 enterprise customers had issues with things like payroll or benefits, and were integrated with SAP or Workday. Our customers told us what other modules they wanted.

How do you measure success; what are your metrics? MAUs vs typical session length and completions?
We have many different metrics for each module; each sub-flow in a module has their own metrics. Some are:

Accuracy: % of times the agent could answer a user’s query from FAQs
Retention: How to keep people coming back and use a module or sub-flow, how often do they come back. We found a 70–75% retention rate for employees coming back — if 100 employees come on day 1, over 120 days, 75 come back.
Messages exchanged: it shows high-level how much interaction is happening.
Session length: we want transactions sessions to be short.

What are successful interactions? What are failed interactions?
The most used module is Service and Transactions, then the HR FAQ. When a default fallback, or a misfire (thumbs up or thumbs down) happens, a human figures it out and takes over. We always have humans in the loop at some point, or the HR department takes over. We use our own internal ticketing tool, or sent to a 3rd party tool like ServiceNow and Neverday.

Onboarding for enterprise client is on a 28-day implementation cycle: we first onboard the policy documents and employee handbooks. Then we have to do the integrations with the top HR software packages. We give the HR people a form with all the features they want.

Team — how did your team come together, and what are the roles?
We have 3 co-founders: we are IIT Delhi 2015 grads who lived in the same hostel in the last 6 years. Two have a mechanical engineering background. All 3 of us loved NLP in college — we did a few projects in auto-content summarization and built our own NLP library. For chatbots, we thought the enterprise use case was the best area.

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

Creating a database of questions is easy — we have to translate an employee handbook into knowledge via an automated system. Still, there have to be humans who do a sanity check at the end.

What are the most common things users ask outside of the main function — do they ask for jokes or other advice?
Employees generally ask seasonal, transactional questions. For example, at the start of the month: payroll. Later, open enrollment and so on. Other employees ask IT, admin, and finance queries. Our next step is to look at other verticals.

User acquisition strategy — how do get enterprises to hear about your bot and start using it?
We do a lot of out-bound (website and emails) and in-bound sales (customer referrals). We will do a lot of content marketing on HR and chatbots, linking back to Leena.

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

The Leena chatbot also reaches out to employees. So if your annual leave expires in the next 20 days, the bot lets you know that (and may even suggest a place to go vacation, given where other employees went). Every morning at 9am, the boss or HR person gets sent any standing leave applications, and what approvals you gave as reminders.

Personality profiles — how important are they and what have you learned?

We see that large multinational banks like their agents to be serious and official. Smaller media companies want a funny and informal one. We have four different identities to offer out of the box.

Monetization? Many bots are a great free service — how do you monetize your service?
We have a per user per year charge, and it differs by module — we charge up front. We have not entered the renewal cycle — Leena AI is young, though we expect high renewal rates.

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

We do this on-cloud and on-premise. Our NLP is completely in-house — it’s an ML/RNN system. We use all of the above for a cloud system (AWS, GCP, or Azure) — most clients run on AWS. We use virtual servers and a standard database stack (noSQL and relational). Our core happens on O-desk; all NLP in Python/NLP. We also use NodeJS, search algorithms, and microservices for integrations to other HR systems. We are on Google Assistant, Alexa, any phone number, Slack, and Workplace for FB.

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

Voice has been ticking up for chatbots. If Skype for business opens up, it could win for the enterprise. Google Assistant and Alexa for the car and home will win — harder for the office. Large companies don’t use Slack and larger companies have big pockets to spend. Microsoft Teams has recently launched new bot support; it should take over from Skype.

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 are in a very early stage — people are trying to understand what works and doesn’t work — a lot of experimentation and data gathering and analysis. We see testing and rollout, constant iteration. A few people are doing well in small spaces: X.AI for meetings.

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

I like a company called Hoogalit, they build voice applications for teachers on Alexa & their focus on making teachers life easier is amazing.

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

It is hard to build value-adding chatbots — as a startup, always verticalize and focus on one small domain/problem to solve it.

Read the prior articles in this series:

Ivy — a 24-hour virtual concierge at hotels (Raj Singh)

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

Woebot — Your AI Cognitive Behavioral Therapist (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)