Shahed Islam
CEO
SJ Innovation
CEO
SJ Innovation
I’m the CEO and co-founder of SJ Innovation. My wife Shahera and I started the company in 2004 in Queens, New York. Before that I came to the US from Bangladesh, studied computer science, and worked at Domino’s Pizza to pay the bills. That job taught me more about customer service and showing up every day than any class did.
We started small, building websites and custom software for local businesses. Today we’re a team of about 120 people across the US, India, and Bangladesh, and we run 60 to 70 active projects at any given time. Over the last four years we’ve turned SJ into an AI-first company. We still build custom software, but now we also have our own products: CollabAI, a platform for AI agents, and Control Tower, an AI operations orchestration layer that we first built for ourselves.
My role is clients, strategy, and partnerships. Shahera runs operations and finance. But honestly I still spend a lot of my time building. I built much of Control Tower hands-on, because I think a CEO selling AI should know how it really works.
Most small and mid-size businesses know AI matters. They just don’t know where to start. They’ve tried a chatbot or two, maybe paid for some tools, and nothing really changed how the business runs. The owner still chases status updates. The team still copies data between five systems. Everyone is busy, but nobody has a clear picture.
We fix that. We build AI that works inside the tools a company already uses, like HubSpot, their email, their project system, and their calendar. Then we put agents on top that do the repeat work: summarizing meetings, following up on leads, flagging a project that’s slipping, preparing a client update. The people stay in charge. The AI does the digging.
A good example is our own company. We had the same problem. Information about every client and project was spread across emails, tickets, and invoices. So we built Control Tower for ourselves, with dozens of agents running on it, and it saves our team more than 40 hours a week. Now we bring that same system to clients in agencies, healthcare, mortgage, non-profits, and real estate.
Culture is the business. We’re a services and product company, so what we sell is really the work of our people. If the team doesn’t care, the client feels it in a week. Our motto is “Empowered Teams, Thriving Clients,” and the order matters. Clients do well when the team has the freedom and the tools to do good work.
We keep our values simple so people actually remember them: put the client first, focus on solutions instead of problems, deliver on time, own your results, keep learning, and support each other. Personal accountability is the one I push most. When something goes wrong on a project, I want the person closest to it to raise their hand early, not hide it until the invoice.
Culture has also changed with AI. A few years ago we asked people to be good engineers. Now we ask them to show a working demo, not a slide deck. We run demo days, where teams present what they’ve built with AI agents. Almost all of our internal work now runs through AI tools, and that only happened because people felt safe to experiment and share what they learned. Our team is spread across New York, Goa, Dhaka, and Sylhet, so we also respect how each office works and keep everyone connected with shared rituals and regular updates.
Our sweet spot is small and mid-size businesses, usually between $5 million and $30 million in revenue. These companies are big enough to feel the pain of manual work, but too small to hire their own AI team. The leaders want to adopt AI. They don’t want to become AI developers. That’s where we come in.
We focus on industries where the work is heavy on people, documents, and follow-up. Healthcare is a big one: medical practices and clinics that need help with scheduling, insurance checks, and patient records, all with HIPAA in mind. Finance is another, including mortgage lenders, community banks, insurance advisors, and fractional CFOs. We also work with legal teams, marketing and digital agencies, HR and training companies, non-profits, real estate, and consulting firms that serve the government.
We’ve delivered more than 2,000 projects since 2004, so we’ve seen a lot of different businesses up close. That history helps. When a mortgage lender or an agency owner explains their problem, we’ve usually built something close to it before. We also know who we’re not a fit for. We’re not the right partner for a Fortune 500 company with a large internal AI team, and we’re not trying to be.
We use it everywhere, and we use it on ourselves first. Almost all of our internal work now runs through AI tools, mostly Claude, plus Lovable for fast builds. Before we sell an agent to a client, it has usually been running inside SJ for months. If it doesn’t help our own team, we don’t sell it. The center of it is Control Tower. It connects our CRM, email, meetings, project tasks, and invoices in one place, and 46 AI agents run on top of it. They summarize meetings (we’ve processed more than 2,500), score client health, prepare employee reviews, and remind our team about follow-ups.
One example I’m proud of is what we call the contact “brain.” For every lead, the system reads the full email history, filters out the noise, and checks if the person is waiting on a reply from us. Then it writes a short profile with open questions and the best next step. Our sales team no longer reads 100 emails before writing one. We’re now building the same brain for projects, so a PM can see hours vs. estimate, scope changes, and risks without digging.
One rule never changes: the AI proposes, and a person approves. Our agents draft the email, flag the risk, or suggest the next step. A human on our team still makes the decision and hits send. That’s how we build trust with our own people, and it’s how we build it with clients too
They overestimate what a generic chatbot can do. Many companies buy an AI tool, give everyone a login, and expect the business to change. It doesn’t. A chatbot that doesn’t know your clients, your projects, or your data is just a smarter search box. Most of the “we tried AI and it didn’t work” stories we hear come from this. The tool was fine. It just wasn’t connected to anything.
They underestimate how much value sits in their own data. Every business already has years of emails, meeting notes, tasks, and invoices. That’s where the real answers are: which client is unhappy, which project is losing money, which lead has been waiting on a reply for two weeks. When AI can read that data and answer clear questions from it, it becomes useful very fast. We saw this ourselves. One fixed-price project had used more than three times its estimated hours before anyone raised it. With the right agent watching, that shows up in week two, not month four.
They also underestimate the work of adoption. The hard part is not the technology. It’s getting a busy team to trust the output and change a habit. That’s why we start with one small workflow, prove it in a few weeks, and grow from there. A small win that people use every day beats a big system nobody opens.
Yes, and more often now than when we started. In the early years we said yes to almost everything, because we needed the work. Over time I learned that the wrong project costs more than no project. It takes your best people, it frustrates the client, and it usually ends badly for both sides.
Today we say no in a few clear cases. If a company wants to build its own AI research team and really needs full-time AI scientists, we’re not the right partner. If a very large enterprise wants a vendor that just follows orders, that’s not how we work either. We also turn down projects where the budget and the scope don’t match, and the client isn’t open to changing either one. I’d rather tell someone at the start that their idea needs three months, not three weeks, than take the deposit and disappoint them later.
Saying no is also a form of honesty. When we pass on something, we try to point the person to a better option, even if it’s another company. A lot of those people come back later, when the timing or the project is right. Scoping honestly and keeping a client for years is worth much more than one oversold deal.
Trust, and trust comes from telling the truth when it’s not easy. Every project hits a problem at some point. A deadline slips, a feature takes longer, or the scope grows. The client doesn’t leave because of the problem. They leave because they found out late, or from someone else. So we tell them early, and we come with a plan, not just the bad news.
The second thing is showing up the same way every week. We give clients a clear update, a real person who knows their project, and quick answers. That sounds basic, but most vendors get busy and go quiet. Our Client Success team exists for this reason. Their job is to sit between the client and the build, so the client always knows where things stand and never has to chase us.
The third thing is caring about their business, not just their project. Some of our clients have been with us for many years, and those relationships grew because we kept bringing them new ideas. When AI came along, we didn’t wait for them to ask. We showed them what we’d built for ourselves and how it could help them. When a client sees you thinking about their business even when you’re not billing, the relationship changes.
Success is when the client’s team uses what we built every day without being told to. Launching is not success. A lot of software goes live and then nobody opens it. For us, the real test comes a month later. Is the team logging in? Did the agent save them time this week? Would they be upset if we turned it off?
We try to measure this in plain numbers. How many hours did the team get back each week? How many leads got a follow-up that would have been missed? How fast did a project risk show up? Our own agency version of Control Tower saves our team more than 40 hours a week, so that’s the kind of number we aim for with clients. Control Tower also tracks agent usage, so we can see what’s working and what isn’t, instead of guessing.
That’s also why we like to start small. We prefer a pilot of two to four weeks on one real workflow, with one clear goal. If it works, the client sees the value with their own eyes, and we grow from there. If it doesn’t, we learn fast and change course. Either way, nobody spends six months and a big budget waiting to find out.
Start small and stay close to the customer. When Shahera and I started SJ in 2004, we didn’t have investors or a big plan. We had a few clients and a promise to do good work. Every client we kept led to the next one. I still believe the fastest way to grow is to make your current customers so happy that they bring you the next customer.
Second, don’t be afraid of hard jobs early in life. My time working at Domino’s while studying computer science taught me more about business than I expected. I learned to show up on time, to treat every customer with respect, and to fix problems fast. Those lessons still apply when I’m talking to a bank or a hospital today.
Third, and this is for anyone in tech right now: build with AI yourself. Don’t just read about it or delegate it. I built much of Control Tower hands-on, and it changed how I lead, how I sell, and how I hire. Today, the gap between an idea and a working demo is a few days, not a few months. The founders who win will be the ones who use that speed to test ideas with real customers, not the ones with the best slide deck. Show, don’t tell.