Startup Explore

Dhruv (00:00)
Decisional helps any business agents for automating business workflows. typically, this used to be a very complex process of either using a no code tool, dragging,

blocks around in a workflow canvas. And Decisional simplifies that process because you talk to an agent that builds its own workflow and tests and maintains it.

Dhruv (00:24)
really is a tool for you know empowering businesses on the smaller side

to really use AI and with the existing tools that they're using.

it typically takes less than 10 minutes.

it's to build a more than 10 step workflow, is I think at least five to 10 times lesser time than it would take on like no code software.

Ali (00:51)
I sat down with Dhruv Tandon, CEO and co-founder of Decisional AI, a company that's using natural language to build AI agents, automating business critical workflows. No drag and drop needed. Let's dive in.

Ali (01:04)
Dhruv, thank you so much for coming to this show.

Dhruv (01:08)
Thanks, Ali.

Ali (01:09)
Awesome. Well, let's dive in. Can you tell me what you're building and what is that big problem that you're trying to solve?

Dhruv (01:16)
Sure, so my name is Dhruv. I'm one of the co-founders of Decisional. And Decisional helps any business build automation agents for automating business workflows. And so typically, this used to be a very complex process of either using a no code tool, dragging, you know,

blocks around in a workflow canvas. And Decisional simplifies that process because you talk to an agent that builds its own workflow and tests and maintains it.

So you don't need to get into managing and maintaining your business process or automation.

Ali (01:58)
Nice. That's very nice. So you basically use natural language to build automations versus maybe some of the traditional tools. And some of the tools that come to my mind would be like, and it ends a peer and make.com. have that drag and drop.

Dhruv (02:12)
one of the important distinctions between n8n and Zapier and Make is that you would have a human operator that would create the structure of the workflow diagram. And you would also depend on this person to maintain it and make edits to it, whereas we really have a completely natural language-based

Dhruv (02:35)
Yeah, I think it crashed as soon as I pressed share. Can you hear me?

Ali (02:37)
I do. ⁓ OK. That's odd.

mean, it dropped. You can try again.

Dhruv (02:43)
⁓ Do you want me

OK, great. I'm going to zoom in a bit. And so the way that you would build an automation on Decisional is that you would just come in and enter a prompt in natural language.

⁓ it will connect to all the tools that you probably are already using and you just hit submit and your agent is essentially going to help you build this automation and it has its own dedicated computer, it can write its own code and so it's really flexible in the way it can solve your problems and the goal is really to work with this agent in natural language and come up with the best approach for

building this automation. And so it's got access to tools. It can generate Excel sheets. It can read documents. ⁓ It can use APIs. And so it has ⁓ the ability of a coding agent, except that it's in a form factor where someone doesn't necessarily need to know how to code or how exactly the code works in order to be able to use these tools.

Ali (03:34)
Mm-hmm.

Mm-hmm.

Interesting. Do you still need to make those integrations like manually or does the agent does all of those integrations on its own once you give the instructions?

Dhruv (04:10)
That's a good question. if the integration is available in our library of integrations, and we have around 1,000 of them, then all you need to do is really click on a button, and it will just take you to the page where you can click on Accept, and it will link the integration, and that's all you need to do. If it's a slightly more complicated, let's say, the

platform doesn't, the integration doesn't exist on our platform, you can always get the API key and like put it in our platform and it'll still be able to read the docs and help you build that integration. ⁓ So the native integrations are easier to connect, but it can really connect with anything that has an API.

Ali (04:51)
Got it, nice.

Dhruv (04:54)
So here, it's really trying to help me think through the automation. So in this case, I asked it to build an agent that tracks leads in a Google Sheet and sends a personalized introduction email when a new lead is added to HubSpot. And so it's asking me a bunch of questions that I should have probably specified in the prompt. But because you can work with the AI, answer questions that it has, I'm going to say it should be AI-ridden.

Ali (05:22)
Mm-hmm.

Dhruv (05:23)
In case there's no email address, I'm going to say just flag it in the sheet. ⁓ These are the kind of things that it works with you on. And it finally will work towards creating a plan.

which is going to be essentially like a job description of your agent. And so typically like, you would spend time writing the role of your agent or writing the role of like, you know, potentially someone who's going to help you with this task. But now, you know, that can be just an AI agent and it'll help you like think through it. So it's asking me who should the email be from? I can say from...

And it's essentially able to use the APIs in real time. And so it's going to be able to read data from these systems and help me think through these integrations, which if you had to do it all in one go, you'd probably build the entire workflow and then run it. And then you'd realize something was wrong.

Yeah, so it is a... So in the first stage you have like a plan.

Ali (06:35)
Okay.

Dhruv (06:36)
which essentially tells

you what the agent is going to do, and it's completely in natural language. It's like a resume for the agent of what its role is. And it tells you what tools it's going to have access to, because you really want to control the tools that you want to give to your agent. You don't want it to be having ⁓ root access to everything in your company. And so you limit the number of tools. And then finally, what it's writing right now, it's writing a workflow.

Ali (07:01)
Mm-hmm.

Dhruv (07:06)
for which so that you can visualize really what's happening. So the agent will write code but really you need a way that you can kind of like operate at a higher level of abstraction where you just see what's happening, you know at a workflow level and so in Decisional every node is actually just a chunk of code and it's just there as a way for a human operator to better manage this workflow and what that means is that you could say that I

Ali (07:09)
Mm-hmm.

Mm-hmm.

I see.

Dhruv (07:35)
I want you to send this email in such a way. First you search the web, then you figure out what's the company. You go to the company website, look at the customers that they have, and then craft something. And so your human taste is really involved in crafting the workflow. So then after it's done generating the workflow code, you click on it and you get something like this.

Ali (08:00)
Very nice. ⁓

Dhruv (08:03)
It's like ⁓ the way it planned it, it says I'm going to fetch the contact, I'm going to set up a sheet, I'm going to start appending leads, then I'm going to check the email, then I'll compose an email, send an email, and then log it as sent. And I could be in any of these steps here that, hey, I think you should work like this or give feedback because I'm really the owner of this process. And I care about the way the work is done. And I want control over how it happens. And so you could just talk to the agent and it would

Ali (08:24)
Mm-hmm.

Dhruv (08:33)
modified and so ⁓ do you think that you have any request here I can actually try to like make a change on the fly so if you were doing this process of like you know reaching out to people like how would you you know personalize the email

Ali (08:49)
Yeah, I think when it comes to personalization, it's a must have these days. And even if it's like small thing, hey, I came across your company and ⁓ what you're building is pretty interesting and then kind of get personalization on that front. Even that would be interesting.

Dhruv (09:03)
Yeah.

Makes sense. So let's add something here. So I'm going to be like, let's use Firecrawl to search the web about something that this person may have talked about recently and incorporate that in the personalized email. So I just entered an actual language prompt, gave the feedback to the agent, and it's going to change the workflow based on just the

the input instruction. And so it's basically using the fact that I know my job better than the AI, so I can really control it and tell it that, hey, you're going to use some lousy web search. You might as well use Firecrawl, because I know I've played around with it a lot. I know it works. So ⁓ that's kind of what we're trying to do, really take the human input into the way the agent works.

Ali (09:34)
That's pretty sweet.

Got it. By the way, your voice input, is that built-in feature of Decisional or is that a separate tool that you're using? Okay. Yes, yes.

Dhruv (10:08)
No, that's a separate tool I'm using. It just speeds up the demos because

I talk a lot faster than I type.

Ali (10:16)
Yeah, yeah, I'm a big

fan of those tools. I use one of the tools as well, Whisper. But I was kind of wondering. OK, awesome. Awesome. Yeah, but I was just wondering if you were able to build into the product.

Dhruv (10:20)
Yeah.

Yeah, yeah, I'm using Whisper too. Yeah, Whisper is cool. Yeah.

Well, ⁓

I recommend people to use one of the tools, either Whisper or Willow Voice. And I think eventually they'll probably all give APIs where you can just integrate it in your product. ⁓ So I think for us, it's great that there are tools out there. We don't want to build them. But we definitely think that people should use them.

Ali (10:35)
⁓ huh.

Yeah.

Yeah.

Makes

sense.

Dhruv (10:54)
Yeah, and so what's happening now is like the agent is actually editing the workflow, trying to incorporate the feedback that I give and put it back in the workflow. And so once it's really done with that, ⁓ and if you want to wait, have fun while waiting, like we have a fun mini game that you can play where you can...

Ali (11:03)
Mm-hmm.

wow, you take care of everything.

That's pretty neat.

Dhruv (11:23)
Yeah, it's

like I have a high score of like 5,000 on this mini game. But the cool thing is that once the workflow is ready, the game will automatically change. So it's just.

Ali (11:37)
That's

very sweet, very creative

Dhruv (11:42)
yeah, so there, like it says now the workflow is ready, it's time to stop playing the game.

Ali (11:43)
It solves the problem.

That's awesome.

Dhruv (11:49)
So yeah, and so now the workflow has incorporated a research lead node. And at this point, it's ready to test.

And so the way this would work is that you can run a test. It would be really annoying in other platforms to run a test because there's no intelligence involved in the testing. Because you'd have to find a record from some system, try to create some mock data. But I can just tell the agent to run a test on mock data, and it'll figure out all of the fields that need to be populated.

Ali (12:28)
That's nice, that's very handy.

Dhruv (12:30)
Right, basically like it's going to run the test and then finally once the test is completed ⁓ you can publish this agent and the cool thing is that you know once you publish the agent you can like

And if you're like me, you're using hundreds of these agents. So there's a dashboard where can track all your live agents, where it's working, where it's not. ⁓ finally, there's just one single agent who's a manager of all these agents that you talk to. And they help you keep these agents working and alive. In case something goes wrong, it'll try to fix itself. Let's say I ran out of five credits. And it would basically fix

Ali (13:00)
Mm-hmm.

Dhruv (13:13)
out that you know this agent is failing because of this reason and it will reach out to me or slack and tell me that you you need to like fix this issue and so I typically don't have to worry about maintaining and managing these agents.

Ali (13:27)
Nice, nice. Yeah, so.

Dhruv (13:29)
So

⁓ yeah, I'm gonna stop sharing my screen now.

Ali (13:35)
That's pretty sweet, especially the gaming feature. ⁓ Going back to things that would make your product different from the closest competition. So I feel like you're taking the best of both worlds. You have visual flowchart, but at the same time, you're taking advantage of natural language to drive the process and orchestrate the process.

Dhruv (13:52)
Yeah. Yeah.

Yeah, so I think that the three major differentiators, one is the control that you get. ⁓ So once, I probably didn't go into this, but once you run the agent, you get a visual feedback of what all ran successfully and what failed. In fact, maybe I should show you one quick example

Yeah, and so you can look at every run that the agent has completed, and ⁓ you can click on the specific run and get node level details about how that particular ⁓ node was executed. ⁓ You can go back and see different nodes. And if there was a failure, really the hardest part of using an AI agent is figuring out how to fix it and how to keep it maintained. ⁓ And so that's kind of what we help with. ⁓

Initially, we thought that the main power of the product is you've probably used something like Cloud Code, right? And so we were like, nobody needs to drag these boxes around because the AI is going to be smart enough to come up with an approach. And ⁓ we didn't really want the visual layer, but then we quickly figured out that that's a really efficient way for humans to operate at a higher level of abstraction. And what that means is that building the

Writing the code and getting the automation to run is not a problem, but making sure it keeps running, making sure that you can make modifications, maintain that automation, that's the really hard part. And unless you know what's going on with your agent, how it's going to do the work, it's really hard to keep track of all of this. Especially when you have multiple agents, you want to know what this agent is up to. And ⁓ as you make it do more critical

tasks, you want some element of control. Like you don't want a wrong invoice to send up, to be sent to your customer with like, know, a wrong zero. So the question is like really giving control. That's one thing that we try to do. also like, but also like give more amount of like reasoning and intelligence. So that's the other aspect. And the third is like having this kind of pattern of an orchestrator that sits on top of all of these automations.

and

helps you manage it. So it's really about software that's agent-native that you can essentially delegate to an agent and it's going to keep maintaining these automations for you. Whereas in the previous world it would be about building a software that's going to be used by someone who's going to click buttons and try to navigate a UI.

Ali (16:37)
Mm-hmm.

Got it. That makes sense. Now I know one of the kind of pain points from my own experience when I tried to automate things, especially when it comes to like node based drag and drop is if there's a bug, it's kind of hard to fix that bug. It takes me some time to try to identify, isolate and then fix it and then test it, retest it. ⁓ How do you guys handle like bug fixes and all of that?

Dhruv (16:55)
Yeah.

Yeah.

Ali (17:07)
Does it have like built-in self-annealing, self-healing feature and then it gets better and better every time it fixes something?

Dhruv (17:11)
Yeah.

Yeah, so what essentially happens is because we split the automation into a step of nodes.

If a failure is detected, it will spawn the agent that built the automation to try to fix it. there are various reasons for a bug in these sort of automations. ⁓ One is that it could be an external issue. Let's say you get your external system, like you're trying to use HubSpot. HubSpot is down for some reason or it rate limits you. That's an external issue. Then there's business logic issues where maybe ⁓

you've asked the automation to always send a particular field. And once that field had a different value, it started rejecting it. So there are different kinds of bugs. What we try to do is when something fails, we try to spawn ⁓ an agent to fix it.

And if like depending on the type of bug, it will either try to ask for your help or it will try to complete the job. And so, you know, there's a lot of different classes of these kind of bugs. And so it's really important to figure out before trying to like, you know, solve the issue. So we are self-healing in the sense that if it figures out, you know, there's a rate limiting error, then, you know, the resolution for that is just like trying it again after like 30 seconds. So, you know, that's something that is possible.

to fix without your intervention, but in some cases, you may need to get some sort of clarification from the person who built the automation. And so you don't really want to try to fix everything. And so as soon as a failure happens, we try to categorize the failure type. And if it's one of those issues that should be resolvable, we try to go ahead and make sure that the process completes, or we will loop in the human operator and try to get some sort of clarification before.

trying to fix it.

Ali (19:12)
Makes

sense. ⁓ Let's talk about the technology a little bit more. What's under the hood of the product? How would you describe that?

Dhruv (19:20)
there are different moving pieces. ⁓ Basically, it's a two-agent system, I would say. You have a...

an agent that's like a supervisor agent that's responsible for just like helping you run these agents or run these you know these sub agents and also like you know look at their performance look at their runs review the data figure out what's wrong give you updates and anywhere that you wanted like you could be in slack you could just ask there you know how is this agent running you could also like come into the platform and ask a question and so that's like the supervisor agent and then

Ali (19:55)
Mm-hmm.

Dhruv (19:57)
there are like each agent has its own workflow that it's responsible for. And so that allows it to be, you know, in a self-improving kind of loop. So we talked about like, you know, having the role and role of each agent described upfront. And so it knows that it's responsible for this workflow. It needs to make sure that this task keeps happening. ⁓ And, you know, we in order to do that, we the agent is writing code to maintain its own workflow.

within that workflow, it could be using another AI agent node, but it could also just be like pure code, right? Because you don't want to be burning tokens for every situation. Like it depends how much intelligence you require for the workflow. And so, you know, we have ⁓ a sandbox that is assigned to that agent so that it can write its own code. And then we have like, you know, security guardrails where, you know, we tokenize

credentials and we never expose it directly to the agent. Instead we sort of give it a sort of a proxy that it can call so that we have control over whether you want to stop allowing this agent to use that credential etc. So it's really about access control also. And then finally you have a runtime environment which is separate from the code generation environment so that you can isolate these two

And so, yeah, it's basically, we started off with the...

problem was like we worked with these businesses and they had like really like their day-to-day workflows were quite messy in the sense that it has multiple steps, you need to go to different systems, some parts of it are an Excel sheet. So we started working from the customer problem backwards and then we realized that you know writing code is the most effective way to execute these workflows. But finally like when you want to manage all of this like people want a digital worker of sorts that's able to manage and orchestrate all of these workflows.

And so we started with, you know, how do solve the problem of like getting this workflow to run in most kind of business workflows? And then we went backwards and how do you make it easier to maintain it? How do you make it easier to manage it? How do you make it easier to kind of like just figure out what's going on? And that's how we ended up with this kind of

Ali (22:21)
are you guys targeting both B2C and B2B space and where is like more you guys are focused?

Dhruv (22:27)
Yeah.

So we're more focused on the B2B space. We want to help businesses automate workflows. ⁓ think there's a different kind of profile of these kind of workflows. They're more specific in nature, ⁓ tend to be more complex compared to consumer workflows where they tend to be more like... ⁓

more like a, I would say, a chatbot agent that you want to speak to that can do one-off tasks. So.

Our product is focused more on repeatable business workflows that you need to have control over. That's the target segment. And we typically work on things like revenue operations workflows, sending quotations to customers, which may be involving some sort of inventory, figuring out how to read documents and quote correctly. Then GDM workflows with enrichment, like something that I just showed you.

And then you know you may need to like hit multiple systems do a waterfall enrichment get data And then there's like ID procurement things like that So it's more focused on you know business workflows because those are more complex and repeatable ⁓ But it really is a tool for you know empowering you know businesses on the smaller side like not like the fortune 500 enterprises ⁓ But more like you know empowering smaller businesses with maybe

to 1000 employees to really use AI and with the existing tools that they're using.

Ali (24:05)
Can you maybe walk me through a customer success story that really helped you to validate the product market fit?

Dhruv (24:15)
Yeah, I think...

When we think about customer success, it is about getting the job done end to end 100%. And so in the initial days, we had different patches of point solutions that probably helped someone solve 60 % of their problem, but they needed to do some manual work. And then we just figured out that's not something that makes someone really happy about using the software.

flows we target are all 100 % completion that you can completely delegate to our software. And so we started off with speaking with a bunch of design partners. we went and asked them what was the hardest thing in your business right now? And then we took a seat back and tried to understand how we could automate it.

In many parts, it isn't about using more AI, it's about using it in the right places. And so that's how we arrived at ⁓ this. We take a design partner on board, we look at their workflow, and we only feel that it's completed when all they need to do is let it run on autopilot. And so that's how we tested our product initially with these design partners. ⁓

complicated quotation system so you had to figure out you know you had a population of

of certain equipment ⁓ assets and you needed to figure out at the right time figuring out what parts to quote and that involves some sort of technical knowledge reading these complicated manuals about the parts of the system because this was some sort of machine equipment and then going into an Excel sheet, writing it down, then coming up with an email template, sending it out to the customer, getting the data from the right system. And so that's how we work backwards.

down the workflow and then we build the product that could basically automate that workflow end-to-end but without our intervention of having to do anything with the customer.

Ali (26:33)
Yeah, that makes sense when you ask for a problem and then you work backwards. I think that's the way to approach it. Now switching gears to maybe to market traction. I'm curious to hear what key metrics you pay close attention to. And if you could share some of those key metrics that would show the either user adoption or any other market traction.

Dhruv (26:56)
Yeah.

Yeah, so we monitor the number of times the agent has run. And we monitor the time that it takes to set up and take an agent live. These are the biggest health metrics that we track. And so so far, we're in closed invite-only kind of release. basically, our agents have run around 35,000 workflows so far. And it typically takes less than 10 minutes.

it's to build a more than 10 step workflow, is I think at least five to 10 times lesser time than it would take on like no code software.

So these are the two things that we track for market adoption.

when we think like, you know, it's going to be a certain types of workflows start becoming more natural fit for us, we'll probably have native integrations for those. And that's when we like open up access to everyone.

Ali (28:06)
Makes sense. I'm curious to ask, ⁓ how did you land your first customer? Because I believe that's one of the most difficult things to do for any startup, land their first customer. So I'm curious what like specific actions you guys took to do that.

Dhruv (28:21)
Yeah, so we spent a lot of time trying to ⁓ generate educational content about

I think there's a lot of noise out there. trying to educate customers about how these things work, ⁓ mostly about ⁓ how you can use agents to automate processes. Not even telling them to do it on Decisional, but how can you be just educating them about the important trade-offs to consider. ⁓ I think that's how we started. And eventually we built enough trust with a few people that they were willing to

try out us versus like 10 other folks in like a of a blind test. And that's how we got our first client where we just, we tried to educate some people. They actually just, ⁓ they just said, okay, let's throw this, throw this like startup in the mix and see how they perform. And we ended up doing well because like we were listening to their requirements. And so we,

kind of we were you know tuned to what they needed and it was just like the performance of the product because you know in early days it's really hard to get you know anyone to trust you but if they can try it and test it and see that it works then that's that's something that's helpful in in helping you close that first customer.

Ali (29:55)
Makes sense. And then if I double click on that, like in order to educate those prospects, did you do like cold outreach to those prospects and then try to educate them? Or did you do some sort of like a blog and content

Dhruv (30:09)
Yeah, it's mostly like LinkedIn content, a lot of blogs, and off late I've been starting to create some videos. But I think most of the initial prospects were from ⁓ social media on LinkedIn.

Ali (30:29)
Now, speaking about future product features, what can users expect from Decisional? Some of the features may be that excite you the most.

Dhruv (30:39)
Yeah, I think, you know, having the...

having a orchestrator agent really manage everything for you. We're in early stages of having that. But that's going to be really fun when you really are able to delegate an agent to manage your hundreds of workflows. And it's going to keep improving them over time, and so that your business is going to get better while you sleep. And I think because the way that we've built our decision,

⁓

Eventually we can start reducing even token consumption for our users without any impact to... ⁓

you know, their workflows, which is interesting because, you know, in most products, it's like you only increase spend on tokens. But because of the way that we've built the platform, you can make a trade off on that ROI. So I'm excited about a future where you could like eventually say that, you know, here's my budget for, you know, this and this is the problem. And you have like a bunch of agents that are able to solve that problem for you for that specific budget.

And over time, it just keeps getting better. Your customers are getting the same quality of experience, ⁓ but you're able to ⁓ time and save money over a long period of time. And you don't need to worry about maintaining and ⁓ dealing with ⁓ broken automations.

Ali (32:07)
Mm-hmm.

Okay, ⁓ let me switch gears and take you back. ⁓ Can you talk about your background and your co-founder's background? And how did you guys came up with this idea and start building the company?

Dhruv (32:24)
Yeah, so my background has been working in startups and product roles. And so I started my career at a bank, hated it. And then I switched over from the dark side, working at startups ever since, worked at a series.

A to C startup that was based out of San Francisco called Drip Capital, which was also a YC startup, then joined this company called Razorpay, which is Stripe for India, also in a product role. ⁓

that company also saw its growth from early days to it becoming a unicorn and ⁓ potentially IPO-ing very soon. And so had a seat on that journey. And then I spent a year working in venture capital before starting to work on this. And I think in early days, we were all excited, at least my co-founding team and I, we were all excited

about how AI is going to change knowledge work. ⁓ But the models are just not good enough. And even though we were very excited about it, it was really hard to get any useful work out of the models. ⁓

you could build a little bit of a knowledge research assistant and maybe integrate AI in a workflow in a very targeted fashion. But I think it's only in the last six months that things have started getting real. But my other co-founders' background, so one of them used to work in machine learning and AI research at Uber before

He left and joined Decisional. And the third co-founder is ⁓ one of my colleagues from Razorpay. And he's a distributed systems engineer. ⁓ We worked together at Razorpay. And then he worked at this company called Arthur AI, which is an ML ops company. And then he joined us. we had a good complementary set of skills in terms of product, AI research, and decision

Ali (34:25)
Mm-hmm.

Dhruv (34:39)
Distributor Systems Engineer. And we've been ⁓ excited about building agents in general. And we actually come from a business ⁓ family background. And so we saw our own family businesses struggling to adopt AI, whereas enterprises had a direct line with all the foundation model labs, especially on the bigger ones. And everyone in between really trying to figure out how to use this technology. And ⁓ that's how we

Ali (34:51)
Mm-hmm.

Dhruv (35:09)
kind of started building out ⁓ something for the folks who may not be technical enough to code, but they really understand their business process really well. So how do we give them tools that can sort of even the odds against ⁓ the bigger ⁓ companies that are able to have specialized talent to adopt AI?

Ali (35:36)
I know you guys went through YC program and I'm curious to ask, what is your biggest takeaway from YC experience?

Dhruv (35:46)
I think YC teaches you to be extremely ambitious.

So ⁓ you are essentially surrounded by ⁓ really amazing people, and that really pushes you to perform more. And ⁓ there's this community of ⁓ your group that meets every few weeks, and you tell each other about ⁓ the progress that you made. so that continual progress in a time pressure

environment.

I think we've tried to carry that forward. ⁓ Our founding team meets every week. We try to consistently track some goals and I think that's the biggest takeaway that I took from YC. ⁓ Trying to have that consistent review of some sort of ⁓ performance ⁓ and really try to compress the time in which you can achieve that with ambitious goals.

Ali (36:48)
sense. Now, thinking about the future, let's say three to five years, where do you see Decisional AI? What is the North Star for the company?

Dhruv (37:00)
I think ⁓ it is about evening the odds, like I said earlier. Our mission is to help ⁓ businesses really adopt and integrate AI into their workflows ⁓ in a practical way. ⁓

I think we would have achieved that if we could help businesses get 100 % completion of their workflows ⁓ using our AI agents. ⁓ I think ⁓ if we could have done that for millions of businesses, would be ⁓ where I would want to see it.

Ali (37:39)
Lastly, I know we've talked about the company, about the product and future vision. Is there anything else that you would like to share with the audience as a final thought?

Dhruv (37:50)
I would be interested to know what they are trying to automate or if they have used any of these tools and what pain points they faced. And I'd love to have a conversation with anyone who would be interested in doing that.

Ali (38:05)
Yeah, that's awesome. We can leave your contact details and then you can reach out and ask questions and tell their problems and hopefully get the solutions. Well, Drew, thank you so much for coming to the show.

Dhruv (38:09)
Yeah, for sure.

Sounds good.

Thanks Ali, thanks for having me. It was a pleasure.