Superintelligence for Solar Installers: Visits, Data and Proposals

Superintelligence for solar installers, in practice, means using AI to organize the pre-visit, gather consumption data, draft a proposal and follow up with the customer after it is sent. The real gain comes from a process with clear steps, verified data and defined limits, not from a system that promises savings without measuring the roof and the electricity bill. In this article, we show how to structure those three stages with a record of decisions and human review.

Key takeaways

• Superintelligence, in the technical sense, is a research topic, not a product an installer can buy. What exists today are language models and agents that carry out bounded tasks.
• The pre-visit should collect objective data (electricity bill, roof type, shading, utility company) and pass the rest to the technical lead.
• No agent should state expected savings, payback period or reduction figures before the technician has run the calculation with validated data.
• Electricity bills contain personal data. Ask only for what is needed, explain the purpose and define how long the information is kept.
• Proposal follow-up is the most useful part to automate, as long as the agent respects the channel, opt-in rules and the review of who can approve terms.

What the term means for an installer

Superintelligence, as a concept, describes hypothetical systems that would outperform people at almost every cognitive task. That debate exists, but it rarely helps a company decide what to buy, which process to change or who is responsible for what. When a vendor uses the term as a sales promise, the useful question is different: which task will be performed, with which data, and who approves the result.

For leadership, the conversation should stay concrete. An agent that answers customer questions on WhatsApp, collects visit information and generates a summary for the technician is an AI system with a limited scope. It can be evaluated, monitored and corrected. If you want to align vocabulary with your team before talking to vendors, the superintelligence glossary for executives helps separate marketing from operational capability.

Before the visit: what to collect and what not to promise

The pre-visit exists to qualify interest and prepare the technician, not to close the sale. A well-configured agent talks with the customer, understands the goal, organizes the information and schedules the visit. It should not estimate capacity, generation or savings, because those answers depend on measurements and assumptions that only the technician can validate.

A minimum intake list is usually enough to start:

• Name, city, property type and who makes the purchase decision
• Utility company and connection type, when the customer knows them
• Recent electricity bills, sent with explicit consent for this purpose
• Photos of the roof or installation site, with an approximate indication of orientation
• Signs of shading, planned construction or a change to the service entrance
• Reason for the contact and the timeline the customer wants

The golden rule is simple: the agent collects, organizes and asks for what is missing. When a customer asks how much they will save, the right answer explains that the calculation depends on the bill, the site assessment and the utility's conditions, and then schedules the technical step.

Consumption data: handling bills without risk

An electricity bill shows the name, address, consumer unit number and, in many cases, other identifiers. That makes it a document with personal data, and the LGPD (Brazil's data protection law) applies to how that data is processed. The practical guidance is to collect the minimum: the consumption history is usually enough for a pre-analysis, and the rest can be masked or not requested at all.

Before putting any bill into an automated flow, write down the purpose, who has access, where the files are stored and how long they will be kept. If the agent uses a model hosted outside Brazil, international data transfer also has to be assessed, and the article on international data transfer with AI covers that point in detail. The Brazilian National Data Protection Authority (ANPD) publishes guidance and is the official reference for following the topic.

There is also a technical risk. Automatic bill reading can misread a digit, pick the wrong month or confuse kWh with currency. A consumption figure read incorrectly produces an incorrect proposal. For that reason, the agent should show the extracted values to the customer or the team and ask for confirmation before any use.

From survey to technical calculation

The agent's job ends when the technician has a complete package to evaluate. That package includes the summarized conversation, the validated data, the organized photos and the open questions. The decisions on system size, configuration and commercial terms stay with the technical and sales team.

Illustrative scenario: a customer sends the last several bills and a few roof photos, but does not know the property's orientation. The agent confirms the consumption values with the customer, records that orientation and shading are pending and passes the case to the technician with those gaps clearly highlighted. The proposal is only released after the technician visits the site or validates the information with the analysis tools the company uses.

A simple decision guide helps the team set these boundaries:

• If roof or bill data is missing, the agent asks for what is missing and does not fill the gap with an estimate.
• If the customer asks about savings, the agent explains what the calculation considers and schedules the technical step.
• If the customer asks for a discount, a special deadline or terms outside the standard offer, the case goes to an authorized person.
• If the customer expresses a complaint, a safety concern or a cancellation request, the agent transfers the conversation to a human.
• If the conversation involves sensitive data beyond what was planned, the flow stops and the team is alerted.

Proposal and follow-up without pressure

Once the proposal is approved and sent, the stage most likely to be lost begins: the customer thinks it over, compares options, asks for changes and goes quiet. Follow-up is the right task for an agent, because it requires consistency and records, not commercial judgment. The agent can remind the customer about the proposal, answer questions about the document, ask whether the customer wants to book a conversation and update the status in the company's system.

On WhatsApp, this follow-up depends on clear rules. Messages sent within the customer service window are treated differently from outbound messages sent outside it. Review the article on the 24-hour window and the article on outbound messages, templates and opt-in before scheduling any reminder. Every follow-up also needs to reach the CRM, and the guide to updating your CRM with AI after a sales meeting shows how to structure that record.

One additional caution: the agent should never offer terms the company has not approved. If the customer received a proposal with an expiry date, the agent states the deadline as it appears in the document, without creating urgency of its own.

How to run a pilot

The safest way to start is with a real but small process. Our method follows four verbs: map, limit, pilot and hand off. This path does not depend on any specific technology and works for any installer, from a single-person residential business to a team with several field crews.

• Map the current flow: where leads come from, who does the pre-visit, who calculates, who sends the proposal and who handles follow-up.
• Limit data and actions: define the fields collected, what the agent can look up and what it can never change, such as prices and commercial terms.
• Pilot one flow: choose one property type and one channel, with logs of every conversation and an agreed operational metric, such as the time until the technician receives a complete case.
• Hand off and decide: review errors with the team, adjust the rules and decide whether the pilot will be expanded, adjusted or stopped.

The metric must be operational. Response time, the share of complete cases and the number of follow-ups recorded are good examples. Avoid measuring only first-month sales, because they depend on factors the agent does not control.

Security, oversight and common mistakes

A customer may paste unusual instructions into a message, ask the agent to ignore its rules or try to extract internal information. This kind of attack is called prompt injection, and OWASP documents prevention measures that should be part of the agent's design. The main one is limiting what the agent can do: it can read the product catalog but cannot change price tables, and it can create a visit record but cannot approve a proposal.

The NIST AI Risk Management Framework offers a structure for thinking about AI governance, measurement and risk management, and it is a good reference for leadership to follow the topic without technical jargon. In daily operations, what protects the company most is readable logs, periodic review of conversations and a clear owner for each rule.

Common mistakes include letting the agent estimate savings to sound more pleasant, keeping bills indefinitely, mixing data from different customers in the same context and having no exit path to a human. Each of these is solved by a process decision before any tool decision.

Frequently asked questions

Is superintelligence a product an installer can buy? No. The term describes a research concept about hypothetical systems far more capable than people. In practice, an installer contracts AI agents and automations with a defined scope, such as pre-visits, data organization and follow-up, and these need limits, logs and human supervision.

Can the agent tell the customer how much they will save? Not before the technical calculation. Savings depend on the validated bill, the installation site, orientation, shading and the utility's conditions. The agent can explain which data feed that calculation and schedule the analysis with the responsible team.

Do you need to ask for the customer's bill? Only what is necessary for the pre-analysis, with explicit consent and a clear purpose. Identifiers such as the CPF (individual taxpayer number) and other personal details can be dropped or masked when they are not used. Define the retention period and who has access before starting the flow, and check the ANPD guidance.

Where should a pilot start safely? Start by mapping the current process and choosing one property type and one channel, such as WhatsApp. Define required fields, prohibited actions and an operational metric agreed with the team. At the end of the test period, decide based on the logs whether to expand, adjust or stop.

Sources

• NIST AI Risk Management Framework
• ANPD: Autoridade Nacional de Proteção de Dados
• OWASP: LLM Prompt Injection Prevention Cheat Sheet