What we have built, and what it did

A property management office of around ten people was messaged by tenants all day long. Every question had to be looked up and answered by hand. If it was a maintenance request, that meant retyping it into the ticketing system and contacting a contractor or repairer. We turned that into a digital front office, built on top of their own Salesforce environment.

Residents simply send a chat message, as they always did; the mailbox can be connected just as easily. The bot asks step by step for whatever is still missing instead of presenting a form, and turns those own words into a clean description and a priority estimate. Nobody has to write out or retype those messages any more. If it is about an appointment, it recognises that and immediately suggests a slot for the calendar.

Many questions are not reports at all. Those are answered straight from the company's own knowledge base, with no ticket and no waiting. If the report is complete, the case appears in Salesforce on its own, linked to tenant, contract, complex and property. When the status changes, the resident automatically gets a message.

If the bot is unsure, the question goes to a colleague with a summary. Their answer goes back into the knowledge base, so the system improves precisely on the questions it could not answer.

A multinational in inspection and certification checked vegetable oils using a process spread over three systems: measurements in Excel, certificates in Word, delivery as PDF, status updates by email. Every inspector built their own version of the same document, and errors only surfaced when the client called about them.

With Operium Harbor the whole job runs in one chain: the nomination comes in, the inspector records the values per tank on site, the lab delivers into the same file and the report is built from the source data instead of being retyped. The client follows along live.

The administrative time per inspected unit fell from roughly eight hours to roughly two. Not because less is checked, but because the inspector is no longer their own typist.

A construction company where the owner handled the administration on the side: quotes in Word, incoming invoices filed by hand in project folders, briefing the accountant afterwards. We built a digital colleague that takes over that work, reachable through a single chat window.

Quotes are drafted in conversation, including the PDF and the accompanying email. Incoming invoices are read and automatically filed under the right project; if the system cannot choose, it asks one short question and remembers the answer. If a client emails asking to drop by, it recognises that, checks whether it fits the calendar and puts the reply and the proposed appointment forward in one go. Nothing goes out until the owner approves.

For a construction company of fifteen to twenty people, that saves around forty hours of administration a month. The more people work there, the bigger the difference.

Built for a staffing agency. We may not name them, but we can describe the assignment: a system that finds the right companies by itself. You say who you want to speak to, it reads websites and profiles and puts the complete contact details in the list. Whatever it is not sure about, it leaves blank rather than making it up.

Then it learns to write the way the sender writes. Feed it your own emails, a book on sales, a video or a market report, and it picks up the tone, the way of persuading and the knowledge of the market. The more material, the sharper the emails.

It then sends the cold emails itself, spread over the week, and reads the replies that come in. It decides whether someone is interested, has dropped out, is on holiday or has unsubscribed, and acts accordingly: on a reply the sequence stops, on an out-of-office message it is rescheduled. After a phone call, the conversation is written out and a suitable reply is ready straight away, based on what was actually said. The sender reviews it and sends; the system keeps learning from every reply and every correction.