Superintelligence for Electricians: Service Calls, Quotes and Follow-Up

Superintelligence for electricians, in practice, is an AI agent with clear rules that logs service calls, asks the customer for the information and photos it needs, prepares a quote from templates your team has approved, and hands to a person anything that requires technical judgment. It organizes intake and reduces rework, but it does not diagnose faults, guarantee deadlines or decide on its own what is safe to work on.

Key takeaways

• A well-logged call includes the customer's identity, the address, the service type, the symptom in the customer's own words, and an urgency level classified by written criteria.
• Site photos help with triage, but they need clear instructions on what to photograph and what to keep out of frame, because of LGPD (Brazil's data protection law) and safety.
• A standard quote works with a fixed scope, written assumptions, explicit exclusions and human review before it is sent.
• Risk assessment, reports, warranties and confirmed deadlines stay with the responsible electrician. The agent only routes them.
• A pilot with full records, one agreed metric and a named owner should come before any expansion.

What an agent does well in an electrical operation

Electrical installation and maintenance companies receive requests through WhatsApp, phone calls, referrals and website forms, and information is often lost between the first message and the site visit. A useful agent at this stage has a narrow job: receive the request, confirm what the customer needs, record everything in one place, and give the customer a clear next step. On WhatsApp, the company must respect the 24-hour window rule, which defines when it can reply freely and when it must rely on an approved template. This is covered in the 24-hour window on the WhatsApp API.

It also helps to separate expectations from capability. What your company operates is a system with rules, limited data and permitted actions, not an all-knowing assistant. The difference is explained in Is ChatGPT Superintelligence? What It Is and What It Is Not.

How to log a service call with enough information

The service call is the unit of work in your operation. If it arrives incomplete, the technician goes back to the customer, the schedule breaks down and the quote is prepared blind. A minimum record usually includes:

• Customer name and a confirmed contact channel
• Full address, with access details when there is a gate, reception desk or commercial building entrance
• Service type, chosen from the categories your crew actually performs
• Symptom as described by the customer, without any diagnosis from the agent
• Urgency, classified from an approved list, such as a total power loss in part of the property or a smell of burning
• Availability window and preferred way to be contacted back
• Protocol number sent to the customer

Urgency is the most sensitive field. The model should not guess it. Write the list of signals that require immediate handoff to an on-call person, and make it clear that when in doubt, the call moves up in priority. Repair shops that already triage requests face the same challenge, and the electronics repair shop case shows how to separate triage, status and warranty.

Which site photos to request and how to guide the customer

Photos help estimate travel, materials and the type of intervention before the visit. The problem is asking for photos without guidance, which produces images of people, documents, mail, children, personal interiors or open panels that were never needed. Images of identifiable people and addresses are personal data under LGPD, so the company must define the purpose, who has access, how long the images are kept and how they are deleted. The national data protection authority (ANPD) is the official reference on the subject, and your company's legal advisor should validate the process.

A list of photo requests from the agent can be simple:

• Distribution panel from the front, with the cover closed, showing breaker labels if they are legible
• The work point from a distance and up close, framed without people
• The outlet, light fixture or entry point of the problem, with no documents or personal items in the frame
• A written description of what the photo shows, when the customer cannot describe it

One safety rule must appear in every photo request: do not open panels, do not remove covers and do not touch energized parts in order to take a picture. If there is any doubt about risk, the agent should tell the customer to wait for the visit and not use the point until it has been assessed.

How to build a standard quote without promising what you cannot deliver

A standard quote is a template with fixed fields and prices that come from your company's service table, approved by the technical and commercial leads. The agent fills the template with the call data, but it does not invent prices or negotiate discounts. The minimum fields are usually:

• Service and scope, described in plain language, with what is included
• Expected materials, with a note that final quantities depend on the visit or on confirmation on site
• Assumptions, such as access granted, working hours, height of the point and property conditions
• Explicit exclusions, such as repairs to third-party installations, civil work and services outside the registered categories
• Estimated timeline, presented as a condition and not as a guarantee
• Validity period of the proposal and how it is accepted
• A statement that the document was reviewed by a person before it was sent

The template protects both the company and the customer because it makes the terms visible. When the proposal depends on a site visit, it should say so. When materials vary, it should say how they will be confirmed. If you want to think about cost per service, the logic of unit economics can help you compare how much time and material each service type consumes, without turning that into a commercial promise.

Illustrative scenario: an outlet that heats up

Illustrative scenario, unrelated to any real customer: someone writes on WhatsApp that the kitchen outlet is getting hot and they smell burning. The agent does not suggest a repair, does not say the problem is simple and does not ask for a photo of the inside of the outlet. It records the symptom in the customer's words, advises them not to use the outlet, and to switch off the appliance at the source only if that is safe for the person on site. It then hands the call immediately to the on-call person at maximum priority. Afterwards, it records the reason for the handoff, the time and who took over.

If the same request arrived as 'I want to replace an outlet', the flow would be different: a request for a distant photo of the point, scheduling, and a quote with stated assumptions. The difference lies in the classification, and it must be written down before the system goes live.

Limits the agent must respect

Limits should be explicit instructions, not suggestions. An electrical service agent should not:

• Issue reports, technical opinions or statements of compliance
• Promise a deadline, final price or warranty that the crew has not confirmed
• Give instructions for work on energized installations or open panels
• Treat a report of risk as a routine request
• Accept or decline work outside the registered service area without human approval

Every handoff should create a decision record: what the customer said, which rule was applied, who it was transferred to and what remains open. That history is what allows the system to be reviewed later, without relying on the memory of whoever handled the call. The principle of keeping human decision points in place is reflected in Anthropic's guidance on building effective agents, which recommends simple systems, controlled steps and clear points of intervention.

How to start with a method

The safest way to begin follows the approach we use in our projects: map, limit, pilot and hand over. In practice, that means:

• Map how calls arrive today, who responds, where information gets lost and how long it takes from first contact to site visit
• Limit the service categories, the data collected, the photos allowed, the forbidden actions and the list of urgency signals
• Pilot one service type and one channel, with complete records and an agreed metric defined before launch, such as the share of calls with complete information or the time to the first human response
• Hand the operation to the crew lead with written rules, and after an observation period, decide whether to expand, adjust or stop

Tests with simulated scenarios should come before any real customer. The team needs to see the agent fail on risk cases, on requests outside the service area and on inappropriate photos before it can be trusted. For observability, the OpenTelemetry observability primer helps you think of records as an event trail that can be audited.

Frequently asked questions

Can the agent send a quote to the customer without review? We do not recommend automatic sending. The agent drafts the quote from an approved template, and a technician or manager reviews it before it goes out, especially when there is a visit, variable materials or risk involved. That review prevents promises your crew cannot keep.

Do you need to ask for a photo on every call? No. Ask for a photo when it changes the triage or the quote estimate, and explain what it is for. Panel photos should be taken without opening the equipment and without exposing people, documents or personal items.

Can the agent say whether a problem is dangerous? It can apply the list of urgency signals your team approved and route the call with priority. It should not give a technical diagnosis or state that something is safe. When in doubt, the rule is to hand it to a person and record the reason.

Where should you start without disrupting the team? Start with one service type and one channel, with complete records, written categories and an agreed metric before launch. Expanding after an observation period is usually safer than trying to cover the whole operation at once.

Sources

• Anthropic: Building effective agents
• OpenTelemetry: observability concepts
• ANPD: Autoridade Nacional de Proteção de Dados