Building a $3M/yr company without launching publicly

Jito Chadha grew Nventr to $3M/yr via word of mouth and expansions.

Here's Jito on how he did it. 👇

Structured workflows

My background has really been in building and leveraging in-house technology across complex organizations. A lot of that work came through family office portfolio companies, where we were dealing with operations, data, manual entry, cloud infrastructure, and business processes with many separate steps that had to connect.

There was so much opportunity there for making existing business processes more efficient. We were constantly looking at where people and cost were concentrated, assessing the complexity and viability for automation, and then going after the lowest-hanging fruit with the biggest impact. That became a repeatable way of looking at problems.

That environment is where Nventr came together. It's a platform for structured workflows, software pipelines, and AI agents. The bread and butter has been structured workflows, and today, we are building on that foundation with Nventr Agent and Agent IO, which extends those workflows into fleets of specialized agents that can work together. Nventr Agent is now publicly available, while we’re continuing to build out Agent IO.

The platform has been live for years, with nearly a trillion dollars in transactions per year flowing through workflows we host. The processes connected to the platform support around 10,000 employees and roughly 200,000 freelancers. Our annual revenue is $3 million.

Getting the product dialed in

I founded Nventr while I was already working at HGM and running Rule14, so it developed inside that portfolio environment from the beginning. We already had operating companies, technology teams, data problems, and workflows in front of us, which gave Nventr a very practical place to develop.

The initial product was a substantial engineering effort. Some of the workflows became several layers deep, with multiple steps, loops, branches, and independently scaling components. We would build around a process, put it into use, and keep refining it as the volume and complexity increased. Our engineers were central to that work, along with the people using those workflows day to day and showing us where the system needed to become more scalable or flexible.

Growing out of the HGM portfolio shaped the product as much as the engineering itself because we were constantly building against requirements that were already showing up inside the businesses.

As far as validation, it came from people actually wanting to use what we were building. Independent management teams and boards advocated for the product inside their companies and with customers and third parties. For most of that period, we had never really launched Nventr to the market or offered self-service sign-up, so that organic adoption gave us a way to get the product dialed in before pursuing outside growth.

The challenge of permissions

One of the biggest challenges has been building something modern while doing it the right way. Companies are already using AI, but once employees start putting documents, agreements, data, and workflows into third-party systems, permissions and security become a much bigger issue. The company may have no control over where that information goes or who can access it.

We have had to build around data permissions, user permissions, and controlled workflows so organizations can get the efficiency of AI without jeopardizing their information.

Our approach is to give the organization one controlled environment where approved tools, workflows, data permissions, and user permissions can be defined by role. For companies with stricter requirements, we can also run the models on dedicated infrastructure so access to those interactions can be limited to the customer’s own employees, and in some cases even Nventr engineers would not have access.

A flexible stack

At the core of our stack is nQube. It handles the structured workflows and scales the underlying computing processes as those workflows run. Around that, we can connect databases, file repositories, and RAG vector databases, do fine-tuning, and use different large language models in parallel depending on the job.

The stack has become more flexible as we have moved further into agents. One provider might handle speech-to-speech while another model handles processing inside the same workflow. We select the toolkit around cost, performance, security, latency, and whether the customer is comfortable with third-party providers or wants dedicated infrastructure.

A two-sided model

Our model has two sides. On the enterprise side, the larger deals are the ones that move the needle. A CTO, COO, or CEO might buy in on behalf of the company and allocate access across a larger group of employees. On the self-service side, individual users and smaller teams can now start with Nventr Agent and expand from there. We currently offer a public free-trial pathway for Nventr Agent.

The expansion opportunity comes from both directions. A company can add users across more of the organization, while one employee can become an entry point into a larger account by showing how they are using the product to automate parts of their job. That creates a path from individual adoption to a much broader implementation.

Organic growth

Early on, we did very little traditional marketing. Most adoption came from people using the product inside the companies that we were serving, advocating for it internally, and sometimes introducing it to customers or third parties. We were much better at building the product than marketing it, so for a long time, all our focus stayed on getting the technology dialed in.

Now, we are putting more attention into expanding our digital presence and increasing organic growth from outside the ecosystem.

For example, I launched AgentCraft, my newsletter on LinkedIn, where I share how I think about agentic workflows, voice agents, automation, and the infrastructure underneath all of it. I'm also publishing more on LinkedIn and through outside media around specific problems like AI permissions, agent trust, browser automation, and what makes voice agents useful inside an organization.

On the Nventr side, we are putting much more of the product and the use cases into public view. We have built out webinars, demonstrations, case studies, and use cases around areas like call centers, document processing, reporting, healthcare communication, and other workflows. Specific use cases can become their own marketing campaigns — if somebody sees that a process that normally consumes days or weeks can be compressed dramatically, they immediately understand why they would want it.

We also joined NVIDIA Inception, which gives us access to a broader technical and go-to-market ecosystem as we scale. 

If I had to start over, I might put more attention on the go-to-market side earlier. We spent years focusing on the product, but I think we could have started building outside organic growth sooner.

Make your job easier

My advice for indie hackers? Automate your work.

I would start with the lowest-hanging fruit with the biggest impact. Look at where you are spending the most time, what you are doing manually, and then assess how complex and viable that process is to automate.

For an individual entrepreneur, workflows can keep things moving while you sleep and do a lot of the manual work far more scalably than one person could. Start with your own role and make that one job easier before trying to automate everything.

What's next?

The near-term goal is to continue opening Nventr up through self-service and outside organic growth. Nventr Agent is now available to a broader range of users, and we are continuing to expand Agent IO as the operating system for fleets of specialized agents that can work together across an organization.

The larger goal is to make those capabilities useful from individual entrepreneurs through enterprise teams while keeping the workflows, data access, and infrastructure controlled as adoption scales.

You can follow along on LinkedIn and Instagram. And check out nventr.ai.