Real work, shipped for our clients

Agents that pick up calls, recover every missed one, draft the replies and plan everyone's day. Humans approve — then spend their time on customers. Scroll down and watch the chaos get pulled into place.

01Law firmReal case, anonymized at the client's request

Nine AI workstreams in production: cases move forward while the lawyers are in court.

Family law · online case platform · several brands

BeforeA day that overflows.

The agents take overEverything flows into the pipeline. A human approves.

AfterEverything's in place, and it just runs.

  1. Calls, emails, documents
  2. Anonymization
  3. Classification
  4. Draft and checks
  5. Associate approvalA human approves
  6. Case file and log
  • 150+ missed calls per business day on the front-desk line, with no systematic callback.
  • Client documents checked by hand, section by section, behind an OCR that misread them.
  • Every message written from scratch, then copied again into the case activity log.
  • No big picture: activity had to be pieced together from exports.
  • Missed calls never returned
  • Unreliable OCR, rechecked by hand
  • Activity logged twice
  • Reporting via exports
CockpitLaw firmAll systems running
  • Automatic callback and message
  • Vision-checked documents + approval
  • A log that keeps itself
  • Cockpit + an agent that answers
  • Missed callsIn production
  • Automatic activity logIn production
  • Message structuringIn production
  • Reply suggestionsIn production
  • Document OCR & complianceIn production
  • Management cockpitIn production
9
AI workstreams in production
2,331
calls analyzed over 9 days to size the callback flow
3,500
anonymized exchanges studied to shape the replies
17
procedural statuses tracked in the cockpit

The firm didn't need a chatbot. It needed cases to keep moving without associates spending their days buried in email, documents and phone calls.

The context

A firm specializing in uncontested divorce that runs its clients on its own platform: every case moves through 17 procedural statuses, with two spouses, dozens of supporting documents and a constant stream of messages.

The approach

An audit of the platform, then a transformation plan in nine prioritized workstreams, shipped to production one by one — each plugged into the existing tool, never replacing it.

A day, before and after

  1. Before: 40 missed calls from yesterday, no idea which to return firstAfter: Every missed call has already had its callback and message
  2. Before: Checking every document by hand, section by sectionAfter: Documents are checked, corrections suggested
  3. Before: Writing every reply from scratchAfter: A context-aware draft is waiting — just review it
  4. Before: Copying emails into the case logAfter: The log kept itself
  5. Before: Piecing activity together from exportsAfter: Leadership asks, the agent answers

What runs today

  • In productionMissed calls

    Every missed call triggers a callback and a template message; the number is recognized, duplicates are filtered out, and the exchange is summarized in the case file.

  • In productionAutomatic activity log

    Every email sent is summarized and logged in the case file — no data entry, no duplicates.

  • In productionMessage structuring

    A raw or dictated note becomes a structured message, ready to review and send.

  • In productionReply suggestions

    A context-aware draft — history, case status, missing documents, firm rules — grounded in 3,500 anonymized exchanges. Nothing goes out without approval.

  • In productionDocument OCR & compliance

    Vision-model OCR, benchmarked against the old one: 9 sections checked, cross-checks between both spouses, corrections suggested.

  • In productionManagement cockpit

    Cases, agreements, email, telephony, productivity: a filterable dashboard, plus an agent that answers leadership's questions by querying the database read-only.

02Energy transitionReal case, anonymized at the client's request

Calls get picked up, follow-ups go out, and everyone's day is ready before they walk in.

Renewable energy company · two sister companies · sales and admin team

BeforeA day that overflows.

The agents take overEverything flows into the pipeline. A human approves.

AfterEverything's in place, and it just runs.

  1. Calls, campaigns, forms
  2. Intake agent
  3. Callbacks and sequences
  4. Days prepared
  5. Team approvalA human approves
  6. Customers and team
  • Calls missed during appointments, which had to be tracked down and returned by hand.
  • Online campaign leads in a spreadsheet — one tab per business line, status shown by row color.
  • The same texts and emails retyped twenty times a day, and follow-ups slipping by a week.
  • Days planned from memory, with no shared view of who was doing what.
  • Missed calls lost
  • One spreadsheet, many colors
  • Texts retyped by hand
  • Days planned from memory
CockpitEnergy transitionAll systems running
  • Every call recovered
  • Leads qualified automatically
  • Automatic sequences, approved
  • Day ready every morning
  • Missed calls recoveredIn production
  • Automatic call sequencesIn production
  • Daily planning agentIn production
  • Leads captured and qualifiedIn production
  • Managed follow-upsIn production
  • Messaging, reminders & reportingIn production
6
agents and automations in production
0
missed calls left without follow-up
7
follow-up stages run automatically
1
day ready every morning, for every team member

The phone rang during appointments, requests piled up in color-coded spreadsheets, and every morning started with the same question: who do I call first? Today, agents handle it — and the team talks to customers.

The context

An energy transition company — heat pumps, solar PV, electrical work, refrigerants — with requests coming in nonstop from online campaigns, the phone and web forms. A small sales and admin team runs the entire chain.

The approach

Map the team's exact moves — pick up, take notes, follow up, schedule — then hand every repetitive one to an agent, connected to the company's phone system, email and calendar.

A day, before and after

  1. Before: Open the spreadsheet, figure out who to call backAfter: The day is ready: calls sorted, appointments, follow-ups
  2. Before: A call missed during an appointment, forgottenAfter: Contact recognized, callback scheduled, text sent
  3. Before: Retyping the same text for the tenth timeAfter: Sequences run — just approve
  4. Before: A follow-up slipping to next weekAfter: The follow-up went out at the right moment
  5. Before: Copying the day's activity into a fileAfter: The stats kept themselves

What runs today

  • In productionMissed calls recovered

    Every missed call is identified, matched to the right contact and followed by a scheduled callback and a message — no call falls through the cracks anymore.

  • In productionAutomatic call sequences

    The call queue builds itself: no answer, a text goes out and a callback is scheduled for the next day; on the second attempt, an email takes over.

  • In productionDaily planning agent

    Every morning, each team member finds their day ready: calls to make in order, appointments, tasks, follow-ups due — and a recap sent before 8am.

  • In productionLeads captured and qualified

    Campaign leads land in the cockpit on their own, duplicates filtered out, matched to the right business line — no more tabs and colors.

  • In productionManaged follow-ups

    A file sent creates its own follow-up; seven stages on a kanban board and a calendar, and the job date triggers the rest.

  • In productionMessaging, reminders & reporting

    Sent from the company mailbox, replies matched to the right contact, reminders on time, stats per business line — with roles that wall off sensitive data.

03B2B distributionIllustrative scenario — what we know how to deliver, told through a typical company

A team of six AI agents, one orchestrator, and humans who make the calls.

Technical equipment distributor · 60 employees · 3 branches

BeforeA day that overflows.

The agents take overEverything flows into the pipeline. A human approves.

AfterEverything's in place, and it just runs.

  1. Emails, forms, public tenders
  2. Orchestrator
  3. Six specialized agents
  4. Human approvalA human approves
  5. Customers and leadership
  • Incoming requests scattered across inboxes, the website form and the phone.
  • Quotes built by hand from the catalog and history, taking several days.
  • Quote and overdue-payment follow-ups that relied on everyone's memory.
  • Leadership seeing the week's numbers the following Monday.
  • Scattered requests
  • Quotes take days
  • Follow-ups from memory
  • Numbers the following Monday
CockpitB2B distributionAll systems running
  • One qualified point of entry
  • Pricing prepared, approved
  • Follow-ups at the right time
  • A brief every morning
  • Intake agentIn production
  • Quote agentIn production
  • Follow-up agentIn production
  • Order desk agentIn production
  • Tender watch agentIn production
  • Reporting agentIn production
6
specialized agents
1
orchestrator and one supervision cockpit
3
levels of autonomy: read, propose, execute
100%
of sensitive actions approved by a human

It wasn't a shortage of hands, it was a lack of continuity: requests came in everywhere, quotes sat waiting, and follow-ups depended on whoever remembered.

The context

A regional distributor selling to installers and local authorities. Emails, forms, public tenders and orders all go through a handful of key people — and one absence is enough to stall the chain.

The approach

Not one agent that does everything: six specialized roles, each with its own scope, tools and level of autonomy, coordinated by an orchestrator and supervised from a single cockpit.

A day, before and after

  1. Before: Leadership waits until Monday to see the numbersAfter: The morning brief has landed
  2. Before: Sorting overnight requests by handAfter: Requests are qualified and routed
  3. Before: A quote that's been waiting three daysAfter: Pricing is ready, the sales rep approves
  4. Before: The customer calls: where's my delivery?After: The customer was told before they called
  5. Before: Follow-ups depend on whoever remembersAfter: Today's follow-ups are out, approved

What runs today

  • In productionIntake agent

    Reads every incoming request, qualifies it (quote, after-sales, order, other) and routes it to the right role with a summary.

  • In productionQuote agent

    Prepares pricing from the catalog, negotiated prices and the customer's past quotes; a sales rep approves.

  • In productionFollow-up agent

    Tracks every quote and every due date, and suggests the follow-up at the right time, in the right tone.

  • In productionOrder desk agent

    Confirms orders, tracks deliveries, and lets the customer know before they call.

  • In productionTender watch agent

    Every morning, spots the public tenders that match the catalog and the coverage areas.

  • In productionReporting agent

    Sends leadership a morning brief: order book, pending quotes, alerts.

Your company could be the next case study.

A first 30-minute conversation to see where your team loses time, and what an agent could take off its plate. No commitment.