Discover more from The Founder Thesis Podcast | Learn from disruptive founders
The AI powered call centre of the future | Rezo.ai
If you want a first-hand understanding of how AI is disrupting traditional services business, then this episode is for you!
Companies often spend a lot of money on new technologies to improve the customer experience, However, most of these efforts don't succeed because they stick to old-fashioned call centres or outdated phone systems that use pre-recorded messages.
Manish and Rashi Gupta, realising this problem, decided to create a smart AI system that can provide instant responses to customer questions around the clock.
Rezo is a technology company that runs call centres for brands using voice bots.
The platform is intelligently tailored to answer precise customer queries, eliminating the need for human intervention. Conversational AI improves service quality without burdening the company financially.
In this conversation (not automated!), the founder duo talks about building and scaling Rezo!
Other Ways to Listen:
Read the text version of the episode below:-
Manish: Hi everyone, this is Manish.
Rashi: This is Rashi and we are the co-founders at Rezo.ai. Rezo is primarily AI powered CX Cloud for enterprises. You simplify this a bit- you can treat us as a virtual contact centre. You reach out to a contact centre for all the sales related calls to generate all the leads. In our current capacity we are 100 people strong, working equivalent to 20,000 warm bodies and we are roughly doing around 30 lakhs calls in one single day.
In today's world, if you look at it sometimes we don't realize that how efficiently we could be. But the idea is: if you're not efficient, the end product is either the enterprise or the customers, it doesn't make sense.
So we are enforcing efficiencies in the call centres, taking away something which is a recurrent job that Rezo can do. If not, then bringing in efficiencies to the human agent. And end goal is basically helping enterprises do more at less other cost.
And the construct here is that we will be able to do a lot of outbound inbound calls just like what a human would have been doing.
We do collections, we do scheduling of appointment, anything and everything that a contact centre would've done, we are now doing it at the power of the bots.
Akshay: There must be like a spectrum of let's say simple calls to complex calls. Till what stage in the spectrum is Rezo able to handle, because I'm sure there would still be some cases where you still need human callers. Where's the line for you right now?
Rashi: The construct is very clear. Calls which needs human intervention, we don't try to attempt that. Imagine there is a death in the family. You're trying to get an insurance, get some money. You don't want to be talking to the bot, you are in middle of the road and you are stuck with your family. It's a mix that we bring on the table.
Where you think of time where Rezo can pitch in, can do the needful we bring that processes. From a simple to a mid-level processes is what Rezo automates. Something which needs a human intervention we let the human do the job and then for the rest we also have a combination that we pass on the ball to the human agents, if Rezo is not able to understand in maybe two probing, if Rezo still doesn't get the context, we pass it to the human agents and then there is a whole feedback mechanism, machine learning at the backend, which is constantly learning.
Akshay: How do you define simple to mid-level complexity?
Give some examples of what's a mid-level complexity. Let's say if I'm calling to change my plan to Airtel for example, would that be something that Rezo would do? Just define it a little like what is the current capability, what kind of complexity can it handle and what it can't handle?
Manish: I think what Rashi was mentioning about is one, if there is some kind of emergency it requires a human touch. And that human touch is something that we are saying right now the AI should not be interjecting on those emergency situations.
Everything else, it should be handled by the bot. Typically, any situation, any call where the average handling time is let's say less than 3 minutes, is handleable by the bot. Everything which requires much larger engagement per se, that's where the probability of the agents coming in will be helpful. That's the kind of way it is.
Akshay: That's a useful definition of 3 minutes. And why is it three minutes? Is it that the more longer the conversation, the more harder it is to contextualize something said at minute five with something said at minute one?
Manish: Three minutes is more like a ballpark kind of a scenario. What also happens actually is, at times the agents will have to look up the different systems and give the answers. And that looking up in different systems might take long.
What happens when solution like Rezo gets deployed. We do API integrations, backend integration with all the systems and the response rates are very quick. We have seen AHT getting reduced dramatically when solution gets deployed.
Akshay: AHT reference to?
Manish: Average handling time. And average handling time comes reduced dramatically because now instead of switching between the systems, looking up into the CRMs, the system is able to get that exact information through the APIs, the backend integrations and is able to serve the customer rather quickly.
Akshay: You have an agent with infinite knowledge in a way.
Manish: And agent with infinite knowledge, agent with infinite cloning capability and 24/7.
Akshay: Can you take me through the journey of starting Rezo. You and Rashi are not just co-founders, but you're also husband and wife.
How did you guys meet? How did you decide to take the plunge?
Manish: We met during our college days. We both were studying at IIT Delhi. We started off first startup in 2012. And that was into data analytics. That was into some sort of machine learning per se. And we had it for close to year and a half.
Akshay: That was a product or a service? What kind of startup?
Manish: It was a services but we were also working on seeding a product at that point of time which was more on the media mix modelling and media measurability, marketing measurability across all channels including the outdoor media.
So we did services as well. And we realized with time that it was way ahead of the curve. We ended up talking to different companies, telling them as to what machine learning is. We were getting compared against the average formula of Excel. How are you different from that?
It was like writing on the wall, probably we missed it. So we ran it for a year and a half. Didn't see much headway there and that was when we had to pull the plug. We took up the respective jobs.
Then again, as part of our jobs and different responsibilities, we saw this opportunity of unstructured data coming into the industry, which is like a road blocker for the enterprises. Anything which is structured can still get consumed very easily. But the moment you have unstructured data, it's a roadblock and you have to employ humans to handle it.
Akshay: What's an example of unstructured data coming in?
Manish: Unstructured data would be voice, would be free text, would be documents. Anything which is not numbers is unstructured data, which is not consumable. You cannot inference from it directly. And that's what we saw.
As part of the business operations, almost all the enterprises generate this data. And in order to consume it, they have to have humans, they have to have people solve that thing for them. So it's like a very simple example, you want to take a feedback, you want to take a NPS score, you want to take a CSAT.
If you can do it over the forms, that's okay. But a larger volume, you'll have to call them up, understand what is the feedback, what is the voice of the customer. And then someone will have to punch into the CRM in a 1 to 5 or 1 to 10 band, give a score. And once the score is received is when the workflow will continue.
That's where we realized that given the evolution of technology of AI, cloud, why can't this be solved? So we started deep diving into it and that's where we realized it might not be a 100% solvable problem, but it's a problem which can be solved to a great extent.
So rather than solving a 100%. Probably we could solve 80% of the cases. That was the leading hypothesis with which we started. And I think we started in 2017-18 window. Spent couple of years in terms of getting the product definitions, the business requirements, because you cannot just take the hypothesis, you cannot take the technology and just sell it like that.
It has to have the business meaning to it because ultimately someone will have to sponsor, someone will have to give the money for it. So we spend good two years making the business viability out of it, the product feature set out of it. Lot of small pivots, lot of features here and there, putting the things together, stitching it together.
And that's where we found the product market fit, where the value is high and that's what we've been selling in the market the last three, four years.
Akshay: When you both quit, you were CTO of RateGain and Rashi was at WNS. You quit without any business on the horizon.
There was no prospective client or a lead where you thought, I can start generating revenue in a couple of months. It was purely the idea that let's build this product even if it takes us two years?
Manish: Typically one would expect something to fall in place maybe in 8 to 12 months. I started doing some odd consulting jobs just to make sure that the pressures should not mount beyond a certain limit.
You have an IPL, you are limited over match, required or net should not go beyond 12. So for first year, one year we were like putting the things together, trying to see what can be done, but then just to make sure the pressures don't mount up.
I took some consulting jobs on the side to keep the ball rolling and one thing we were very clear from day one; that we do not want to get into the services site. There were ample opportunities for us to pick up the services projects and services is a very different ballgame.
And we were very clear on that. We don't want to go that route and we want to go to the product side.
Akshay: Between the two of you who was trying to get business, who was selling and who was building.
Manish: Given the kind of background and expertise that we bring on the table, I was more on the tech side. Rashi was on the data science side and the solution side. And collectively we both were selling because, I would say we were both equally novice and equally specialist in sales. It was like, we have to go out, divide and conquer, let's both sell. And with time now I think Rashi has picked up the sales pretty well.
At the moment it's Rashi who is selling. It's like anything in the front of the office, Rashi takes care, anything back of the office, I take care.
Akshay: Did you start by selling what you're selling today? There must have been a journey where you would be selling something and over time based on what you heard from customers, the product evolved, just tell me about that.
What were you selling initially? What response did you get and how did you finally discover product market fit?
Manish: What we started essentially was more in terms of, automating or automating the chat responses.
Then we moved towards the email automation, social media automation. And all these nuances short text, long text, their understanding, conceptualization, semantic analysis, all these things is where we spend good amount of time. Incidentally voice was the last major segment that got added.
Akshay: Because voice would need more time to build. It would be more time and investment to build voice capability. Texts sounds relatively easier to implement.
Manish: Actually, I would say voice in the end didn't take us that long. It was much faster than we anticipated but somehow we undermined the potential of voice.
Akshay: So when you were selling text, I'm assuming it would have been a crowded market. There are a lot of other players also offering text, chat automation and social media listening. Is that what led to voice? What led to discovering that voice is the niche which you can dominate?
Manish: I think competition has never been a concern for us. I don't want to sound ignorant or arrogant here, but competition has never been a concern. Two major reasons. One; the market is super huge and you can have 10 unicorns operating in this space. So if there is so huge of market, competition is not a concern.
It will be a concern maybe three years down the line, but not today. There is enough land to grab. What really we were looking at is what is the value that we are getting for those enterprises. If I, for example, do a chat automation, what is the ROI the enterprise is getting?
Let's say I go to a 2000,5000-crore company and I talk about a chat bot and they're giving me 50,000 rupees a month, 1 lakh rupees a month, the ROI they're getting is 1.5 lakh, 2 lakh, 3 lakh rupees a month and of which they're giving me a portion of it. The question was: why am I not able to give a bigger value to the enterprises?
That was a question I was solving. If I start giving a bigger value to the enterprise, I can go and ask for a bigger part.
That's the problem we were solving because, whenever you are in a B2B space, enterprise space, you have to give a bigger value. If you are just a superficial or a periphery service provider, you are like a fly on the windscreen. You might stick onto it for some time, but you might get replaced anytime.
Akshay: And did this come from customers asking you for voice or you only started checking with customers, 'Hey, would you be interested in voice'?
Manish: In the first, our customer said, we like the team, we like the concept, we like what you're bringing on the table. But we would want to have a single vendor which can handle everything. In a way our clients pushed us into it.
Akshay: I believe Maruti was the first big client for voice for you. Tell me about that. What was their ask and how did you deliver it?
Manish: Essentially we applied; Maruti runs an incubation program with the name MAIL: Marti Automobile Innovation Labs. And it’s a program which enables startups to apply and they would then want to invite the innovative startups to come work with Maruti's different verticals, different teams and do something big, something different. And at the same time, it probably also allows a balancing act for the enterprise in terms of the size they operate at and maybe a combination of little agility. So honestly we applied in that, I think we were in Cohort 2. We were in Cohort 2 and we applied for it and luckily, or I would say destiny, we got selected and we were one of the winners of that program and that's where we got onboarded. We got a paid pilot and the success, the results of the paid pilot were satisfactory for Maruti.
Akshay: And Maruti wanted to do this NPS survey through voice.
Manish: Yes. One is doing an NPS survey, but it was also about getting the VOC; voice of customer. So the calls that were happening, what were the customers talking about?
Akshay: And this was after servicing, like when someone goes, gives their car to a Maruti dealer for servicing. After that, a call goes asking about the experience.
Manish: That was one. Second was also in the normal conversation. In the customer support conversation, what were the customers talking about? Were they really happy? Was there any hidden dissatisfaction aspect?
Akshay: So that was like monitoring the human calls that are happening and creating a dashboard based on those calls, how effectively the call was handled, what was the customer's emotion? Was he positive or negative about the brand and things like that.
Manish: Absolutely. I think your sales pitch is better than mine, I believe.
Akshay: And Maruti wanted this to be multilingual, right?
Manish: Essentially any business if you see in India and not just Maruti, but even otherwise what we've seen is unless any enterprise or company is regional in nature. Typically, 60 to 70% of volume lies in Hindi and English.
For everyone, the first version, first iteration of deployment is- do this in Hindi and English. Once this milestone gets achieved is when the regional languages are very critical. So if I was to say till date, we do around 10 languages across India, all the major languages of India, we do.
Akshay: From the technology side, what is the product doing? Is it essentially a text to speech engine and vice versa, a speech to text engine and then that text is analysed for intention? Is that what it is?
Manish: The reality is that there are nuances in the speech engines, whatever engines are available in the marketing, we have couple of our own versions of speech engines.
So we have to create an ensemble layer on top of that and see which engine works the best in what scenario. So we do that.
Akshay: What is a speech engine?
Manish: Speech engine is text speech agent. It's ASR, you have speech detection, text to speech. Those are the speeching things. Then obviously you have a NLP layer which kind of works and which is our own proprietary in-house.
Akshay: And what does NLP do? What's the full form of NLP? What is it?
Rashi: What we do under the NLP engine is; we hear the voice and this is at the backend of the product, there are a lot of these elements, a lot of features which are embedded in the product. It could be understanding the free text because what we enable the whole system is to understand the voice of the customer and just get it to thrash that data and bring it to the enterprise, this is not what you wanted to hear, like five things that you wanted to check, but your customer is also talking about this 50th thing and the 60th thing that you never even imagined. So that was what we wanted to bring to the enterprise. With that as a thought, now you imagine anything that you built as a function on the voice or on the chat, this was the backend enablement that we wanted and we wanted this to be enabled in all vernacular languages. Now, the complexity when you look at the landscape right there is, anybody can talk whatever. It could be any verbatim, linguistic comes into play.
Slangs comes into play, short forms comes into play, anything and everything because it's like what you and me are talking to figure that context. And then basis that there is this whole generative AI where you basically have to revert in a manner. For example, I may just say my remote is not working.
The customer is saying this, now the brand needs to handle it in a way and we are the one which are handling on behalf of the brand. So all interpretations, the backend engine has to be so powerful to be able to get the gist of the customer, do an intent identification, figure out from the backend of the enterprise; what kind of a response, what kind of a API do I have to fetch to be able to give out a response to it.
Now you do an amalgamation on this simple construct, with all vernacular, with all sentiment, with all emotion embedded into it. In simplistic, like Manish said the base was that, but then you built in the complexity of the sentiment, of the emotion, of the open verbatim, of the slang of the background noise, because we work with a lot of people.
You are driving and you're talking to the bot, there is a lot of this honking, road noise, vendor all sorts of noises are there. You have to cut that and the bot should still be able to interpret everything in a unified manner and should be able to roll appropriately. It can't do any hangups in between.
We did chat, we build the engine on the chat. We started interpreting chat, either on social or on the website. Then with the ask of our brands that we were working with, they said this is good. What you've built is good, but can you take it to the next level? Can you bring in speech? We said, okay, speech. We've got into voice. Then they said, can you do vernacular? And we did vernacular. Now, can you do vernacular, you make it open-ended. We don't want a 1, 2, 3, sort of a bot.
You make it open-ended. So that is how the complexity for us kept happening. And now the ask is bring in variations, bring in generative AI's, incorporate the ChatGTP, then the other ask is add emotion, add sentiments, add tonality, interpretation.
How the product gets carved is basis our deployment and basis when we see the ask coming from our brands that we are doing for a rollout. And the competition is where, you look through that. If you do this, then you are there.
So that is where a lot of debating internally we do basis the ask. Then we put forward the ask to the product team and the tech team that this is what we are looking at and then we build this whole size together.
Akshay: There is a handover between the speech engine and NLP engine or they work together? Like the speech engine would convert it into text and then the NLP engine looks at the text.
Rashi: It goes seamlessly, because the revert has to happen within couple of seconds. So the whole embedding is- the speech to text will happen, interpretation will happen, API calls will happen at the backend, generative AI responses will happen and then a revert would go to the customer. Imagine all of this like the way you and me are talking, it has to happen like that. So it goes seamless.
Akshay: And if it is vernacular, then it's translated to English for the NLP engine or the NLP engine is processing in the vernacular language?
Rashi: It's processing in the vernacular because once you do the interpretation, the context changes. It's good for if you have to do one or two pieces, I think it may still work. But if you have to look at a broader picture what we are trying to solve, like open-ended conversation, it will be a show stopper. Initially we tried that, but after a point of time we realized that this is not working out. So we rolled back.
Akshay: It'll sound very unnatural. Translation removes the flavor of the language from it. So, the speech engine for vernacular language, did you have to build this or was it available? English of course there would have been a lot of work done and would be available.
But how did you build capability for, let's say something like Gujarati for which you may not had enough off the shelf speech engine available or something like that. Just help me understand that.
Manish: There are two things here. One is- the corpus is available. Indian government is putting a lot of effort towards building this corpus.
There are universities working in silos. All this time they're working in silos now, coming together and set up a collaboration amongst themselves along with the Government of India.
Akshay: By corpus you mean the dataset, labelled data.
Manish: So that's getting built up. Then this dataset corpus is also available with universities, online it's there. In fact so much so that I know you'll be a little surprised. Why not the movies? The closed captioning is already done, what better data set would you want?
So all the movies that are there in the market and someone has put in the effort or has been paid to create the closed captioning manually. That's the corpus you need.
Akshay: So you were able to develop the speech engine for vernacular using the corpus which exists.
Manish: We were able to build this to some extent with this corpus available. We also leverage some third-party engines which are there. And a mix and match is what kind of is working for us.
Rashi: See, the mantra is not to say no to business, when the business is coming. You don't have to say no and you have to figure out a solution.
If you have something built-in you roll, if you don't have built-in, you don't shy away from figuring out how can you leverage. Third party is competition, whatever it is but the crux is never say no to business. Just keep rolling. So that is what we have adapted and moving on.
Akshay: And you were talking of things on the roadmap in terms of incorporating ChatGPT? How would that become a feature in the product? What would that imply?
Rashi: We are testing it out because the idea is: we are also working with a lot of NBFC's, lot of financial organization, insurance companies, banks.
Few weeks back, we have rolled out a version, incorporated something which is already existing. We are testing it out because it can't be open-ended. Tomorrow it's going to be talking something, and then the price will be paid by the enterprise, we have done the solution.
The construct is that a huge, in fact the government themself, there is so much investment going into seeding this. But then on the other hand there is a larger investment, the grants have been issued to basically also put the boundaries because if you allow it to dock, it can do anything.
Akshay: What does it look like? This ChatGPT powered version? Does it allow the bot to give answer in longer sentences and show more empathy and emotions while talking?
Or why incorporate? That's what I'm trying to understand. What is the value it'll add? How will ChatGPT add value?
Manish: For example if a user, let's say Manish goes to a bank's number trying to play a prank and start ordering a pizza, I mean us-Rezo system and obviously what the permissions we have from the bank because there are certain guided rails within which you have to operate. So our system will say, 'sorry I will not be able to help you out with this'. Whereas ChatGPT will say, 'pizza checkbook kya sath ly gy, account kholo gy toh pizza mil jaye ga'.
Akshay: Slightly more humanistic, quirky responses. That's what ChatGPT will allow.
Manish: Yes. And this is again a hypothesis and probably will get tested with time, but remember in the beginning we said a 3-minute AHT is a limit. This is something which might help go beyond 3-minute part.
Akshay: Because you're able to engage people more.
Manish: That's correct. It's same as; if you see a parliament session or if you see a keynote by anyone and which is let's say half an hour or one hour, they need to put some jokes in between to get everyone's attention.
Similarly ChatGPT one use case that we see at the moment is, getting that engagement factor there.
Akshay: And currently you are using Generative AI already, you have been using rather. So Generative AI here refer to the fact that you are generating the responses of the bot. Is that what it refers to or what is Generative AI?
Manish: Generative AI overall, if I was to say, is to generate content which doesn't exist. So typically ChatGPT has two components. One is semantics or understanding the content, intent, context, whatever you might want to call it.
And then basis that come up with some solutions, some answer, which chances are very high doesn't exist in totality, because if it just gets you or fetches you a solution or an answer which was there in one of the websites then it's a search engine, then it's not a generative content. But if it can with certain thesis, hypothesis in the system which it has been trained on, it can generate the content and give it to you, assuming plagiarism isn't there to that level. And it was not copy pasted from any website directly. So that's what generative content is all about.
Akshay: And this you have been using from when you started Generative AI because your answers of the bot are essentially Generative AI.
Manish: That's correct. We were having this version there.
And it was fine-tuned or perfected for the B2B, for the enterprise space. Because there are certain boundaries in which you'll have to operate when you work with enterprises. You cannot go beyond; you cannot go all over the place. In fact, I just want to cite an example here and without naming the enterprise; what happened was they rolled out a chat bot- ChatGPT on their website and within two days they had to take it down because the answers it gave were not in lines with the legal compliance of that company.
Akshay: This was a chat bot for answering customer queries. Take me through your pricing journey. When you started how did you price it at, what did you learn about pricing a B2B product? What do you price it at today?
Manish: In the beginning when we started the best we could charge was 50,000 rupees a month, 1 lakh rupees a month. Obviously for any company, as a startup also the ticket size, the average selling price or ACV, which is annual contract value- you have to try and maximize it. So we tried doing it, and what we realized was the value that is coming on the table for the enterprise is not that exorbitant or is not in lines with what we are looking at. So we had to flip it and we started doing a value-based selling which really helped get a much better ROI to the enterprises and basis that even our ticket sizes started increase.
Akshay: What is value-based selling?
Manish: Value-based selling is all about what is helping the enterprises derive and understand the overall value that this offering that we have, brings on the table.
It's same as, if I say that I sell you a pen for 10 bucks. Probably you'll say, I have more pens but this 10-rupee pen, I might buy it but I typically lose pens. I have that habit of losing pens every two weeks, the ink starts leaking and this and that.
One approach is I start selling you 10-rupee pen. And I say, I'll sell you now, I'll sell you again after the week, I'll sell you again after a month. That's one approach. Second approach is I try and study typically, what is the average shelf life of a pen with you.
And I see on an average you need a replacement every six day. So what I'll say is, don't worry you use as many pens as you want. This is the package. You want to change the pen on a daily basis, please be my guest.
Instead of charging 10 bucks, I'll be charging you a different structure. And you'll say this is too expensive. I'll help you understand as to typically last year what you spent was this amount. I'm giving you a discount over that because since I'm selling you a bigger ticket size, my unit price will not be 10 bucks.
My unit price might be 8 bucks or 7 bucks. So actually I'm helping you save 30% and your pain point goes away that, every time you have to think again about buying a pen which when to buy, it's a win-win for you. It's a win for me because I have a larger, a bigger horizon with you.
Akshay: In your case, you would probably look at the salaries that they would be paying to agents and how many of those agents you would be able to replace through Rezo. That would be the basis of pricing.
Manish: Let me try and explain as to how we typically take it. We say what is the total cost.
Akshay: Of running a call centre.
Manish: Of running a certain function, let's say running a certain service, say capturing the NPA score. What is the total cost associated with getting the NPA score? What is the delay that you get?
What is the opportunity loss you have? What are the pain points you have? We start with that and then see what all we can solve for them, the solution solves. What other things that we can help solve and basis that we say, today let's say you're spending 50 lakh rupees on getting the NPS score structured with a delay of one week.
What we bring on the table is something which will cost you 5 lakh rupees real time and it's a plug and play, you deploy it and you forget about everything else. You'll get the results within a matter of, let's say two hours. Whereas you were getting something in seven days and things like, was it really transparent because it could have been managed, whereas this is like a automated system everything there, no human intervention.
It's more about comparing as to what it is as a status quo and what we bring on the table and the value that we bring on the table. And in this there have been ample of cases where, for example, if they're spending 50 lakh rupees today, annually and when we did our math and we saw that best case scenario, I'll be able to give this service for 45 lakh rupees, I tell them, do not change.
There are ample of cases even today. In fact, I had a call with one such customer a week back where they wanted automation rolled out and we said, 'sir, do not change it. What you are doing is the right way of doing it. Do not deploy solution of Rezo'. In fact, I would suggest do not deploy from anyone because the size that you operate at, the use cases you have, automation solution will not be able to help you out and you will not get the ROI and nor will the implementing part.
And it'll be burning the bridges three months, six months down the line. It's better to be upfront.
Akshay: You sell directly to organizations or you work with the implementation partners and is there like a channel sales model or is it direct sales?
Manish: At the moment, we are selling directly to the enterprises but we do have some channel partners, who have started selling now, little early in the construct. But yes, it's picking up.
Akshay: Once you sell, let's say this NPS survey, how much time does it take to go live? How long is that onboarding journey? Because you would need to collect data or from their systems and you would need to read data from the systems. You would need some access to the phone numbers.
You would need some sort of a calling line set up or something, how does it happen? What's the onboarding journey?
Manish: Our typical onboarding journey, if the enterprises are ready, the decisioning is made and they have the information, API documents, data handy. I'm just counting the delay because of the enterprises or our clients. We can get started anywhere between two days to maybe one week's heads up.
Akshay: This kind of selling approach of value-based pricing, doesn't it reduce the speed of sales? Because every time you go to a client, it's not like you could give them a rate card, but you have to first ask them.
Manish: Again, we do give them the rate card also. I'm not saying that we don't, our rate card is fixed. We have a certain range.
Akshay: What's the rate card based on?
Manish: On the complication of the use case, the volume that is coming in, what kind of commitment we have. So rate cards are fixed. It's fixed in a sense, it's in a certain range and it's a very well defined. I'm not saying that we inflate our prices basis, the value. We don't do that.
Akshay: But you're able to demonstrate value to get the conversion
Manish: Correct. Our rate cards are in the similar range. It's just that this approach helps the enterprise come to the conclusion faster. Actually this approach helps selling faster.
Rashi: In addition to what Manish said, what we bring on the table is- they may have a current process and they're sitting on large data. They're doing things in a certain way.
When we come in and as Manish was saying, there are rate cards broadly. It's a volume-based pricing. It's a fixed sort of a structure which is there.
Akshay: Which is like a permanent or something like that.
Rashi: Yes, it's a permanent depending how the larger the volume and all of that is.
But broadly, when you go to the enterprise and you talk to them, it is very important to understand, they want to automate a process or they are looking at something was running in with, in a certain way. And they're saying, ki mery ko toh itna hi response mil raha hai ya itna hi loog mera pick up kar rahy hai. What we bring on the table is the success.
Like today the selling is majorly on a success base. Today you are getting x as output. I will be able to take it to 2x or 3x in such and such time. And that time duration as Manish said, our deployment times once the client is ready on the basic construct, like a lot of my competition where the deployment time could be three months. We do it in flat 10 days.
Along with the UAT built in. And success is what we drive. We say that, if you today are able to get X as an output, we will be able to do it 2x, 3x depending upon the complexity of the problem. And that is one. The other thing that when we talk to the enterprise, when they are just talking, there is a lot of base.
These are your customer base and you are making revenue from your customer base in a way. You're serving them, you're making revenue from that. There would be always an untapped bet that you don't even know you have not attempted. So when Rezo comes into the play, we are more data.
Our decisioning are majorly powered by data. Where are you sitting today? What is your current input? What is your current output? What's broadly your cost? There's not too much of conversation. What you're thinking that this is going to be taking a lot of time for me. No, it's very straightforward because, you just say, where are you today and where are you wanting to be?
Can you do this with the current bandwidth or the current infra that you have? The answer is no. Okay, so we will come in, we will look through your data and we will get you from here to here and you can measure the success. So that's the construct. And when you put that construct, irrespective of who's by competition. These discussions goes much faster.
And once they generate faith in us. They know, they value my data, they're able to hear me out, they're able to understand. That is where they say, this is my untapped customer base. I have not even looked at this base. Can you do something about them?
So then you do a lot of co-inventing with them and you talk to them, you figure out this is what it is. And it doesn't take a long because the idea is; if you start looking at the data, you start looking at the problems that they are facing. And you propose simple solutions not very complicated, simpler ones, you're able to roll in a much quicker manner.
Akshay: Does the product get customized for each customer? Let's say male voice, female voice or stuff like that? What all customizations happen?
Rashi: You can define that. We have seen that; we were working with a lot of driver base at certain point of time. Now, when we were working with the driver base they were not liking the female voice. It was not just working out. They wanted to hear a male voice telling them instructions.
Another time we were working with a brand where collections were happening. We got a mandate that only the women voice is going to work, bring in a women voice from age 30 to 35. Because they have run these units for a very long time, they know this much clearer than what you can do. So we amalgamate and put things together and this agility that we have to be not hard coded everywhere and listening to the client and bringing things. This basically is playing in our favour.
Akshay: So these are available off the shelf voices, like a 30-year-old Indian woman voice. Does it sound close to human or it's obvious that it's a machine?
Rashi: No, it's pretty human. In fact we have rolled it both in the urban and the rural areas. The rural latches onto these voices beautifully because it's easy, we're rolling it out at the farmers, people hearing about various policies, about what is happening, konsa seed kab lagana hai you can ask anything and everything. As far as the urban is concerned what I can sense is; for example unlike you being hogged by the multiple calls during the day if you make one single call for an insurance policy and then you get 10 calls in a day till the time you pick up the call, you answer 15, 20 question. These kinds of calls are happening.
We leverage a beautiful concept, which is a smart contact centre strategy. Once I've understood the DNA of Akshay, I know ap kab phone uthao gy, what are your plus and minus, I know your transactional history and this is what we bring on the table. It's not just conversation ki ek messaging hai jo mery ko brand ki taraf sy apko dy na hai.
Akshay: When the call starts, maybe the bot would speak in a language let's say bot speaks in English, the customer responds in Hindi, so that change over happens on the spot. The bot will also start.
Rashi: The introduction is basically in one or two languages because you can't do it in all languages. For example if the call is in Hindi and the other guy started to talking in Telugu for example, so the switch will happen in Telugu and the rest of the conversation will be in Telugu.
Akshay: And then you have data that next time you should use Telugu with that customer. How does the cost compare to a human agent? Let's say maybe a human agent would give you 2-300 minutes of calling time a day to 20 days, say 5,000 minutes in a month.
What would 5,000 minutes of Rezo cost?
Manish: So, basically the cost reduced dramatically, let me answer it this way.
I would say ballpark. Depends on the complexity, depends on what kind of a volume is there, a lot of factors. It could actually be lesser as well. It ranges anywhere between, I would say a 40% to maybe a 60, 70%. Depends on a lot of factors.
Akshay: A cost of a human would be at the very least, maybe 25-30,000 a month. Because you also have the cost of providing a laptop and the space, et cetera.
Manish: It's a setting cost, it's the training cost, it's the hiring cost, it's the onboarding, it's the offboarding, it's the PF, ESI everything, then the TL cost, then the QC cost everything put together.
So there are multiple factors. In fact, I'll give you a very interesting perspective as well. I was talking to someone in terms of helping them derive value-based, ROI based positioning. And this person was telling me we pay 20,000 to an agent.
I said, 20,000 you pay, but that is not your cost of operation.
Akshay: It would be double of that probably.
Manish: Yes, but then obviously you can't just say double. You have to justify. One thing that really came out was, as per the government law, you have to give them certain leaves.
The leaves could be around 25 to 28 offs, in a year. This is besides the Sundays, because typically any employee gets 18 leaves in a year. Plus there are festivals, plus you have 52 Sundays. There are medical leaves, there are paternity, maternity leaves compassionately there.
Akshay: And then health insurance and the office infrastructure.
Manish: Typical setting cost can range anywhere between 1000 rupees to a maybe a 10,000 rupees per seat.
Again, depends on lot of factors, then you have the bosses above TL to maintain this entire hierarchy. There are account managers, electricity cost, internet cost, laptop.
Akshay: And in your case, you're at half of what it would cost for a human to be running the same thing.
Manish: One thing, I want to really call it out is, the first impression or the first thing that comes to the mind is job security for these agents. And interestingly, when we work with these enterprises, not a single agent has been let go.
Akshay: They are redeployed for higher value work.
Manish: They are redeployed for higher value work, which essentially requires their upskilling and upskilling I mean training more.
Akshay: I want to understand as a business, how you're doing.
What is your current ARR? What kind of customers do you currently work with? What's like your biggest sector from where you have customers? Is it NBFC's or is it automotives. Just give me an idea of what your revenue and the breakup of revenue looks like.
Manish: We currently focus on automobile and BFSI and telecom. Three verticals that we are focusing on at the moment. Currently we are purely based out of India and operating in India. We have around 15 paying clients today. Another five in the pilot stage.
Akshay: Who are some of the clients that you could name? Are you at liberty to name some? Like Maruti of course, we've already discussed.
Manish: So Maruti is there. We work with L&T Finance, we work with TATA AIG, we work with Delhivery, Usha.
Akshay: Delhivery would be like, when there's a customer who's expecting a parcel. So a call would go and say that your delivery will come at 11 o'clock or like feedback for the delivery or something like that.
Manish: Delhivery, it's been a non-voice function, non-voice automation for them. The use cases are those straight-line use cases. And where we are in terms of our revenue and others.
Let me put it this way, we are targeting to hit 10 million ARR USD by end of this financial year.
Akshay: How much is that in crores?
Manish: It'll be around 80, 85 crores as a run rate. And in terms of our otherwise health, we are EBITDA positive company. We are a profitable company. We've been profitable for last eight quarters now.
Akshay: And this includes your salaries also,
Manish: Absolutely. Everything.
Akshay: Do you need to raise funds? Are you planning to raise funds?
Manish: We are in discussions to raise capital, some early-stage conversations.
Akshay: But why, you are profitably growing.
Manish: We have just set up our US subsidiary and what happens now is, at the growth stage you need to kind of experiment. You need to invest in the sales and marketing engine. And also as part of the growth, the experimentation needs to be quicker which requires capital infusion. And that's the reason why we cannot raise in capital.
Akshay: But your product is already mature.
Manish: Product is already matured. It's already there, in terms of the scale is there and everything is there, but still the feature sets will keep evolving, keep improving basis the customer feedback.
Akshay: Like going beyond that three-minute handing time.
Manish: Absolutely. So what happens essentially, just for example, sometime back the feature that got added was about switching the language on the fly. You and me are talking in English, but ek dum sy mai ny hindi mai baat kar li toh shayad ap b hindi mai baat karo gy.
We have designed the system in such a way ki bot b swith karta hai.
Akshay: And the voice sounds like the same person only talking?
Akshay: Are you getting used for sales calls also or it's mostly just data gathering?
Manish: We are doing a lot of sales calls as well. We help setting up the appointments.
Akshay: What kind of sales? Say tech companies generally do a demo of the product. So setting up those kinds of appointments?
Manish: One is setting up those appointments that need verification. In fact we've started selling high ticket size items as well in Automobile.
Akshay: And this would be like a cold call or a call to someone who filled a form on a website or something?
Manish: It's kind of both.
Akshay: You're adding a lot more value because then you're doing revenue generations. Sales is obviously every business lifeblood.
Akshay: It seems like the opportunity is huge in India. Why go to the US right now? For example, like telcos is an unscratched market for you right now and telcos would be a massive market.
Manish: In India itself, absolutely. You are right, India is a huge market no doubt about that. And in fact, as a market it's just scratching the surface.
Akshay: There is so much potential for you here.
Manish: See, it's like there is a certain consumption or a capacity for an industry, for a geography and we are already maximizing and we are expanding as well.
As Rashi said, we are hiring for 50 positions right now. While we are expanding, while we are consuming this market, we would need more geographies to be added, to fill up our capacity or our hunger for growth. And if you see a new market, that new market itself has a gestation period of 12 to 18 months. So if you start now, you'll see the results in next 18 months, not today.
That's the reason why adding new market is important. Plus it also gives us a much better stability, much better learning of the feature sets or the use cases and the product evolution.
Akshay: Obviously going to US would help the product also grow because that customer is a lot more demanding.
Who are the other companies in this space?
Manish: There are companies in different kind of sections. There are the likes of Unifold and Observe, which are setting more on the data analytics part or the speech analytics part.
Akshay: So they would be like monitoring human conversations and sharing like a dashboard with feedback.
Manish: Also nudging them, prompting them as to what next to speak. Those kinds of things are there.
Akshay: How big a part of your revenue is this? You also do this, like monitoring?
Manish: This doesn't have that much of contribution to our top line today. But in terms of our pipeline, I would say almost one third of the sales pipeline in the pilot that we have is this. Then there are obviously players like Cognigy, there are ASAP and then Variant to some extent. So these are the players.
Akshay: I've not heard of these companies. What do they do?
Manish: So these are the companies in the US I was talking about which kind of focus on the voice automation, voice watch or getting those data points, completing the workflow.
Akshay: They're doing the conversations and transactions, et cetera like talking to customers. There is none in India, these are all US based.
Manish: Yes, these are all US based. There are some in India but in India things are at a very little early stage, I would say. There are players like Naani, there are players like Skit. Skit has now moved to relocate it to US.
So there are players there, but again everyone is little early stage working on different kind of use cases. And what we've seen is for example Saarthi is big time into collection process only. Although we have a massive rollout and massive deployments in collection as well.
So we do see some competition from that perspective. Naani on the other hand is majorly into the government side.
Akshay: Government side like political campaigning and all or promotion of schemes, government schemes, et cetera.
Manish: I think that as well, we're working with different government entities.
As I said, there is ample of opportunity, ample of stuff to be done. And the market is opening up. Then at the same time market really wants to have evolved products because no one wants to have unfinished product and get backlash from the customers.
Akshay: What's your plan to open more accounts in India? Because any company which has more than let’s say 10,000 headcount would have at least 10, 20% of their workforce doing this kind of calling role. So there is a lot of opportunity here. What's the way in which you think you can capture it?
What's the way to grow your sales pipeline?
Manish: We want to grow fast. In fact, we've been growing almost 3x year on year for the last two years. And we expect the same growth going forward as well.
I think probably one of the big thing that we bring on the table is, we do not do sales that much. We are focusing big time on the product, on the technology. We keep sharpening it. There are certain outreach programs that we have, which we run and we have some awareness programs letting the enterprises, the businesses know that this is what can be done. This is what we do.
And we prefer it that way because our clear mantra is whatever we do, we need to do it good. We do hear a lot of enterprise coming to us and saying, we haven't even heard about you. We didn't even know that a company like you existed with this kind of a numbers.
That's a scale we were not even aware. And we are like, that's okay. Because it's okay to talk to 100 companies and not 1000 but whatever hundred companies we talk to, we need to do a good job. We want them to be satisfied with the deliveries that we do.
So it's like a balancing act that we can now follow.
Akshay: I'm wondering why none of the telcos like Airtel, Jio, Vodafone, none of them are using you.
Are they already using some products? Or is there a resistance because I'm guessing they would have the maximum number of such calls happening.
Manish: We work with Airtel.
Akshay: Which could eventually become your biggest account, because Airtel would have so many such less than three-minute calls.
Manish: Well, I would say there are hidden gems in the industry. We know Airtel for the scale. There are other enterprises which actually operate at a much larger scale. The need is much larger.
Akshay: Like NBFCs would also be doing a lot of collection calls.
Manish: I'll also tell you a very interesting use case. Let's say in the collection process only, let's say given an enterprise X, they would have let's say 1000 agents doing a certain function, a certain collection process. But their books are so huge, so much of work to be done. Thousand people cannot reach out to each and every customer, whoever is there in the queue for the collection, they'll just probably pick the top 10% of buying.
Akshay: They prioritize.
Manish: So that prioritization, they'll only be able to touch 10% of the customers. But the 90% goes uncertain. But a solution like Rezo, there is no scale problem. Probably even with our kind of offerings, the bottom 10% would not make sense in terms of the ROI, but it's the middle 80% which is there up for grabs.
So when we look only at the call centre, we're probably looking at the thousand, which is like handing the top 10%, but the actual opportunity is for 10,000. These are the hidden gems in industry.
Akshay: What's your advice to founders who are planning to start? And because you've scaled up to soon to be 10 million ARR with no external fundraise.
Manish: We had one funding, it was a seed round three years back and at the beginning of the Covid.
Akshay: What advice would you like to give to aspiring founders?
Manish: My suggestion would be that you have to be very clear in the beginning itself. VC money is good, but you need to be very clear that are you going or doing your business to raise money or you are raising money to grow the business. That visibility, that clarity has to be there. Both have their own pros and cons. That is important. That's point number one. Point number two is have your family, your friends, your social circle, a good circle around you which can morally support you because this is going to be a hell lot of a journey.
Before you go……the analytics only tell me so much, I want to hear what you feel and think about the conversations.
Mail me at email@example.com with your comments & feedback or if you just want to hear my comments on your startup idea - I love getting your emails!
Until the next founder's thesis📕,
Your host, AD