Camila Insights
The AI agent is rarely the problem.
The real estate boom in northern Peru
Lambayeque is one of the regions showing accelerated growth in the development and sale of new real estate projects. Its coastal location, cuisine, productive land, and prices that remain attractive compared with other markets have drawn buyers and investors.
By August 2026, our internal research had identified more than 30 projects in the region. About 40% were apartments or houses between 40 m² and 90 m², usually starting near S/100,000 and reaching S/150,000. Another relevant category was lots for urban expansion, frequently priced between S/15,000 and S/20,000, with a significant supply of 90 m² lots.
This growth is also connected to programs such as Techo Propio and public works that create expectations of future appreciation. The market, however, is not made up only of buyers already living in Lambayeque. In the conversations we analyzed, we found people who had moved to Lima or other large cities and, after years of work, wanted to invest in their home region to build a house, country home, or future business.
This makes sales particularly sensitive to follow-up. Many prospects need the exact location, documentation, financing details, and a visit coordinated for when they return. Their intent may exist from the first contact, but the decision often matures over weeks or months.
In this context, real estate companies do not only need to generate more demand. They also need to respond quickly, preserve the context of every conversation, and help advisors identify which opportunities are truly ready to move forward.
Urgent problems
As in any growing market, development comes with operational challenges. At the beginning of the year, we thought lead generation was the main problem. After speaking with managers, marketing leaders, and real estate advisors, we discovered that much of the loss happened later: during qualification, distribution, follow-up, and closing.
Processes and commercial knowledge are not always documented.
A real estate company’s know-how is one of its main differentiators: how it profiles a buyer, which questions it asks, how it handles objections, when it schedules a visit, and what information it needs to recommend a project. Much of that knowledge, however, lives only in the experience of the longest-serving advisors. Without a playbook, reliable information source, and clear escalation rules, an AI agent must work from incomplete instructions. One of our first lessons was to organize the process: record stages, document FAQs, define minimum qualification criteria, and turn the team’s best practices into reusable skills and workflows.
Changing how people work creates more friction than installing the technology.
Some teams still organized prospects in basic spreadsheets without defined commercial stages. In others, a CRM existed, but not everyone checked the app, updated contacts, or knew how to make it part of their routine. Age is not the determining factor; digital familiarity, perceived usefulness, and support are. We chose small pilots with advisors willing to participate from the beginning and learned that individual training does not scale: adoption requires internal owners, follow-up, and frequent feedback.
Responding to everyone delays the advisor’s work.
Teams receive contacts with very different levels of intent, but do not always have clear prioritization rules. In two sector experiences, around 70% of contacts were estimated to be insufficiently qualified or not to require immediate attention. Meanwhile, advisors spend much of the day on visits, tours, meetings, and documents while WhatsApp keeps accumulating questions. The goal should not be to automate more conversations, but to help the team recognize which ones need immediate human attention.
Without evaluation and supervision, the agent repeats operational mistakes.
Giving feedback to an AI agent is as important as giving it to a new colleague. Consistent results require examples, quality criteria, current sources, and clear limits. In our first implementations, an incorrect result did not always mean the model was incapable: sometimes the CRM was outdated, inventory did not match the current offer, or nobody had defined a correct answer. Missing information should not automatically become a negative score. The agent should explain what it used and let the advisor correct the result.
These problems showed us that implementing AI agents is not just connecting a model to WhatsApp or a CRM. It also requires reviewing daily operations, organizing information, documenting knowledge, and supporting people through change. Technology expands the team’s capacity, but it does not replace the work of building a clear commercial operation.
Our proposal: Camila
These lessons led to Camila, an agentic application designed to complement the real estate advisor’s work. Our intention is not to replace advisors, but to protect the most human part of the sale: listening, advising, negotiating, and building trust.
Camila’s first version began as a workflow connected to WhatsApp and a CRM. It helped us validate concrete tasks: answering initial questions, collecting information, recording contacts, qualifying prospects, and transferring conversations to an advisor with more context. It also exposed an important limitation: when the customer did not understand who they were talking to, or the transition to a person felt abrupt, the experience lost continuity.
This led us to rethink the product. Camila 2.0 is being built around a simple idea: the advisor should not have to learn how to configure multiple agents and applications. They interact with one Camila while she coordinates the required capabilities behind the scenes.
Instead of assigning an opaque priority, we are developing explainable Lead Scoring that compares each contact with the relevant project. When data is missing, Camila should show what she needs and suggest the next question rather than invent a conclusion.
We also want Camila to help after the first contact: remember abandoned opportunities, prepare a daily summary, propose messages from the latest notes, record visits, and keep the next commercial action visible. The goal is not for advisors to talk less with customers, but to arrive better prepared for the conversations that matter.
- Business context: projects, inventory, commercial terms, FAQs, and playbooks.
- Commercial management: contacts, deals, tasks, visits, notes, and pipeline stages.
- Specialized agents: qualification, follow-up, CRM updates, handoff, and content generation.
- Advisor experience: a simple, mobile interface for work inside and outside the office.
Learn more about our vision for Camila and explore her AI capabilities.
Work is constantly changing
Building Camila forced us to question an early assumption: that developing a good agent would be enough for a company to start working differently. In practice, even capable technology can fail when installed on incomplete data, informal processes, and habits nobody is willing to change.
In one implementation, we paused part of the development and tested existing tools first. The sequence was more useful than continuing to accumulate features: adopt, train, integrate, document, and only then build on what we learned. Pausing the product did not mean abandoning the vision; it meant protecting it from our own assumptions.
Adoption does not happen on launch day. It happens when an advisor discovers that updating a contact takes less time, that several sheets are no longer needed to prepare an agenda, or that the context of a conversation is available before a call. Each small benefit helps replace an old habit with a new way of working.
That is why we believe the agent is rarely the only problem. The result depends on information quality, process clarity, ease of use, and feedback frequency. When a company changes tools without changing its operation, it digitizes the same disorder. When it first organizes its knowledge and supports its team, AI can become a true extension of their capabilities.
Our vision for Camila continues to evolve with every conversation and implementation. We do not want to build another chatbot that answers more messages. We want to build a colleague who understands real estate context, coordinates specialized agents, and gives advisors time back for what no automation should take away: being present for their customers.