From agency to martech: the products that grew out of our own problems
In 2024 our plan said outsource development. Today btodigital has eight in-house products. The pattern, what we learned, and how to tell if you have a product.
AI, digital marketing and medical marketing consultant · btodigital
On December 19, 2024, btodigital’s strategic plan said we would “outsource graphic design and web development to partners.” Twenty-two months later we have eight software products of our own. None of them came out of an innovation workshop. Every one of them started as a problem that hurt us first.
This is the second article in a series about how btodigital, the marketing agency I founded in Medellín, Colombia, changed with AI. In the first one I described how we worked in 2025 and how we work now, process by process. This one covers what came next: once an internal solution works, the temptation to sell it shows up, and sometimes that goes well and sometimes it does not.
It is written for the owner of a services firm who feels that selling hours has a ceiling. I will walk you through the pattern that worked for us, what we learned and a checklist to tell whether an internal problem could become a product.
Why does selling hours have a ceiling?
Because every new client needs somebody’s hours. If you want to serve twice as many clients, at some point you need more people, more coordination and more room for mistakes. We are 28 people in Colombia, and in September we invoiced 62 companies. Each of those accounts takes time from specific people.
A product works differently. Atendio answers a WhatsApp message at three in the morning without waking anyone on the team. The audience report for a regional media outlet has been assembling and sending itself since October 2. That work no longer depends on someone having a free afternoon.
At our October team meeting I put up a slide with a single line: “We are no longer just an agency.” It does not mean we stopped being one. It means we now have two ways to grow.
What are the eight products and where did they come from?
| Product | What it does | Where it came from or where it stands |
|---|---|---|
| Atendio | AI agent for WhatsApp | Thousands of unanswered chats at El mejor DOC. Started on June 16, 2026; paywall since October 1 |
| Cerebro IA de btodigital | Per-client memory, reports, community, commissions | The team’s need to stop starting from zero with every client. In production since July 8, 2026 |
| El mejor DOC | Doctor directory and appointment booking | Its chat volume is where Atendio began |
| ShopiUP | Shopify app that optimizes product SEO with AI | Available to Shopify stores |
| Dictado Médico IA | Voice-dictated medical records | Voice dictation was switched off over cost; it now works with templates |
| Medellín traffic simulator | Simulates road closures and one-way changes | A mobility analysis in El Poblado; published at btodigital.com/lab |
| btomant | WordPress maintenance without server access | Our own web maintenance service |
| AI video | Video production with artificial intelligence | More than 15 videos of Medellín made with AI |
Look at the third column. None of them came from asking what product we could sell. They came from asking what was costing us time, money or clients.
What is the pattern?
I sum it up in four words: inside first, product later.
1. A problem that hurts us. El mejor DOC was receiving thousands of WhatsApp conversations and many went unanswered. It was not a market hypothesis: those were potential patients slipping away every week.
2. We solve it for our own use. Atendio was first built to filter and answer those El mejor DOC conversations, not to be sold.
3. We use it with real clients while the team keeps watch. Once it worked internally, we took it to service clients, with the team watching closely. That is where the problems internal use never shows tend to surface.
4. Only then do we put a price and a door on it. Atendio’s paywall arrived on October 1, 2026, more than three months after the first line of code.
The Cerebro IA de btodigital followed the same path, with an interesting twist: clients started asking for things. An industrial brand wanted to see its social media comments in one place, understand what people were saying and reply faster. We built that inside the Cerebro IA de btodigital as a community module. On its first sync it classified 85 real comments (72 positive, 12 neutral and 1 negative) and drafted a reply for each one, which a person approved before it was published.
We would never have imagined that module in a product brainstorm. It came from a client with a concrete problem and a foundation that already existed.
There is one more fact that explains why this was possible now and not five years ago: I built a large share of these products myself with Claude Code, without putting together a large engineering team. The 2024 plan listed development among the low-margin activities to hand off to third parties. AI changed that math.
What we learned
Technology that works is not enough if the cost does not add up. Dictado Médico IA was built so doctors could dictate and the AI would write the medical record. Dictation worked, but the cost did not add up. We switched it off (reversibly) and the tool now works with templates. The price clients pay has to cover what it costs to run.
A product nobody pays for is a project with a nice cover. It is the line I now repeat to anyone who shows me their new product.
How do you know if an internal problem is a product?
Before you give it a name, a logo and a pricing page, run it through this list. If you answer “no” to more than two questions, you probably have a good internal tool rather than a product, and that is fine too.
| Question | Why it matters |
|---|---|
| Does it hurt you every week? | If you do not feel it often, you will not improve it with discipline |
| Have you already solved it internally, and do you use it daily? | It is the only proof it works outside a slide deck |
| Do other companies have the same problem, in the same way? | If it is your own quirk, it is a custom project |
| Has someone outside already paid, even a little? | Free usage proves nothing; charging does |
| Does the price cover the cost of running it? | The voice dictation lesson |
| Can it run without your team hovering over it? | If every client needs your people, you are still selling hours |
| Who sells it and who supports it? | A product with no commercial owner stays on the website |
| Do you know what you will do if nobody pays in six months? | Deciding the exit point early protects you from sunk costs |
Two pieces of advice that do not fit in the table:
- Charge early. What taught us the most was charging, not launching. Free tiers fill up with users who never tell you the truth about the value of what you built.
- Do not take your paying service clients for granted. While the product matures, the service business is what keeps the company running. If a new product puts the care of your current clients at risk, it is not ready.
What am I not going to tell you?
I will not give you a revenue percentage by product. I do not yet have a clean measurement of how much each one contributes, and I will not make one up. Nor will I tell you the agency is over: we keep serving companies every month, and 117 different ones have bought from us so far in 2026.
What I can tell you is that the move from agency to martech was not a one-day decision. It was the sum of internal problems solved with enough discipline that other people wanted to use the solutions. If your problem is WhatsApp customer service, which is where Atendio started, the path is laid out in this guide to WhatsApp AI agents for business.
If you are thinking about turning an internal solution into a product and want a second opinion before investing, that is part of what I do as an AI consultant for companies.
Frequently asked questions
What is a martech company? A company that combines marketing with its own technology. In our case, besides providing marketing services, we build and sell software: a WhatsApp agent, a per-client memory platform, apps for online stores and tools for the medical sector.
How many in-house products does btodigital have? Eight as of October 2026: Atendio, Cerebro IA de btodigital, El mejor DOC, ShopiUP, Dictado Médico IA, the Medellín traffic simulator, btomant and AI video.
How do I know if an internal tool can be sold? If you use it every day, if other companies have the same problem, if someone outside has already paid for it and if it runs without your team on top of it. If it fails more than two of those conditions, keep it as an internal tool.
Do I have to stop selling services to sell products? No. In our case the service business is what made the products possible, because each one came from a problem we saw while serving clients. What matters is that the new product does not put the people already paying you at risk.
Do you need a large development team? Not anymore. I built a large share of our products with Claude Code. What you do need is someone who understands the problem deeply and has the judgment to know what not to build.
