AI for electrical contractors: why it needs a company brain — and what that means for a business in practice

A company brain is a business's accumulated knowledge in a form an AI can work from: the customers, properties and installations, the finished quotes and Rapporte (signed records of work done on site), the business's own article catalogue with its own prices, its own order types and fields. Without this memory, an AI writes text that sounds like a textbook and guesses prices nobody can invoice. With it, the AI suggests what the business actually did last time — which is why the question to ask of an AI for electricians is not "how clever is the model?" but "what does it work from?".

What a company brain is

A company brain is the sum of the data a business holds about its own work, stored so that software can find it and use it on the next order. The term comes from the world of AI vendors, and it describes something an electrical contractor has had for a long time — only scattered: in the boss's head, in the installation's binder, in the office's Excel sheet and in three years' worth of Rapporte.

For an electrical contractor, that memory is made of five things:

1. Customers, properties and installations — who the customer is, which buildings they own and which installation sits in which of them. The installation is the memory, not the electrician: whoever works on a KNX installation five years from now wants to know what was configured last time.
2. Finished documents — the quotes and Rapporte the business actually wrote. They show how this business phrases and structures things, and which line items it usually sets for which type of order.
3. Its own article catalogue — materials and services at the prices the business actually charges, not list prices from the internet.
4. Its own structure — order types, statuses, custom fields. An order called "Standby" has to land, in this business, where standby orders land.
5. The deadlines — the inspection period of every installation, the inspection body's defect list with its deadline for remedying the defects, the due date of every invoice.

This is the same data that makes up an ERP and a CRM for electricians — described in the article "ERP and CRM for electricians". A company brain is not an additional system, but what emerges when that data sits in one place and the AI is allowed to access it.

What an AI does without that memory

An AI that does not know a business is a very well-read intern on their first day. It writes a flawless introduction for a quote that sounds like every other business. It suggests line items from a textbook, not the ones this business sets for a flat renovation. And it guesses prices — or leaves the price out, which forces the office to look up every line.

The problem is not the model. The same model writes a usable quote as soon as it has seen three finished quotes of the same order type from this business and takes its prices from the business's own article catalogue. The difference between a useful AI for electricians and an irritating one therefore lies almost entirely in what it works from.

That leads to an uncomfortable order of things: first the memory, then the AI. A business whose installations are recorded nowhere and whose quotes sit in Word files on three computers has nothing an AI could learn from. The work of bringing customers, properties, installations and documents into one place is not the boring part before the AI — it is the AI, four-fifths of it.

How elektro.box uses the company brain

We describe here what the software does today, as stated on the page "What the AI does here" — and we keep its two modes of operation apart, because the difference between them is the whole point.

In the inbox, the AI acts. An incoming message — email, WhatsApp, web form — becomes an order: with a subject, a short description for the electrician, the order type chosen from the business's own types, the custom fields, the site address and the person on site. Customer, contact and property are created along with it if they do not exist yet. This is where the company brain first shows: the order lands in this business's first status, with its order types, not in a generic drafts folder. If a detail is missing, the software asks the customer — autonomously by default, and switchable to require approval.

On the document, it suggests. Line items for quotes and Rapporte come from the order text, the messages attached to it and up to three finished documents of the same order type — so that the suggestions sound like this firm and not like a textbook. The prices come from the business's own article catalogue and are calculated; none is invented. Every suggestion lands in a window with a checkbox per line, and only what is ticked gets added. Whatever is already on the document stays there.

On top of that comes what the software reads from documents: a photo of a delivery note becomes material line items with supplier and delivery note number. A photo of a distribution board becomes the circuits — designation, type, rated current, the RCD's rated residual current. An inspection body's defect report becomes one line per defect, and the defect texts are taken over verbatim, because what the inspection body wrote has to appear exactly that way later. Each of these readings adds to the company brain: the circuits belong to the installation, the delivery note to the order, the defect to its deadline.

The rules a company brain needs

A memory an AI is allowed to write into needs rules — otherwise it gets watered down. Four of them are built into elektro.box, and they are a useful yardstick for any AI a business introduces:

1. What exists is never overwritten. On a customer who already exists, the email address and billing address are not touched. A property is only created when the street and postcode are there; a company is inferred from a business domain, and explicitly not from a free email address.
2. What is unclear is shown rather than guessed. An order whose text is not enough is flagged for review. A deadline in the defect report that is not clearly stated comes back empty.
3. Reversible things immediately, everything else with a click. The assistant logs an hour or adds a measurement while you type. Merging, deleting and changing many records at once build an approval card, and without that click nothing happens.
4. Every action appears in the activity feed — and the feed distinguishes whether a person acted or the software. That is the real answer to the question "what did the thing do?".

And a fifth, which sounds less technical: where it can be done without a model, it is done without one. Whether a document is a supplier document is decided by fixed rules. An SIA 451 file and a Steiger file are read line by line, the same way every time. A path that ends the same way every time is worth more than one that is right most of the time.

What this means for a business in practice

Anyone evaluating an AI for an electrical contractor — elektro.box's or another — should ask four things before watching a demo with somebody else's data:

1. What does it write from? From our finished documents and our article catalogue, or from what the model knows in general? The answer decides whether the suggestions sound like your own firm.
2. What does it do without asking? Which actions run on their own, which wait for a tick, and can that be changed? A business needs to hear this from the software, not from the customer who got an automatic email.
3. What happens to what we already have? Is an existing customer overwritten, an existing quote line replaced? "Never" is the only acceptable answer.
4. Where do we see what it did? A feed that distinguishes person from software — or nothing at all.

And one recommendation: bring to the demo an email as it actually reaches you and a document as it looks at your business. What the AI makes of them says more than any slide. Which areas the elektro.box company brain covers today is listed on the Features page.

Frequently asked questions

What is a company brain? A business's accumulated knowledge — customers, properties, installations, finished documents, its own article catalogue, its own structure and its deadlines — in one place, stored so that an AI can work from it instead of from a textbook.

Does an electrical contractor need its own AI for this? No. It first needs its data in one place. The same model writes a usable quote as soon as it has seen three finished quotes of the same order type from the business and takes its prices from the business's own article catalogue — and an unusable one if not.

Does the AI invent prices? Not at elektro.box: the prices on a quote come from the business's own article catalogue and are calculated, and the example documents the AI learns from never contain prices. With any other AI, that is the first question a business should ask.

How much time does an AI save an electrical contractor? There is no figure here, because we have not measured one. What there is, is a calculator into which you enter your own hours and your own rate, to see for yourself whether it pays off: office workload calculator.

---

The product information about elektro.box in this article comes from the documentation on "What the AI does here" and Features, as of 15 September 2026. The article makes no statement about what other vendors can or cannot do.