Skip to content

What we've built

Case studies

The systems we've built, what they do and how far along each one is. The waste tech business at the end is where it all started.

Client work

Built for other businesses

Healthcare · Ireland

Laptop and stethoscope on a desk

Where it stands: the dashboard is in use with the operations team. The report studio is being tested by clinicians on real cases, and each region's rules are being added as they're confirmed.

AI for healthcare practices →

A private diagnosis service that had outgrown its spreadsheets

This clinic carries out private diagnostic assessments for children and adults, as well as assessments for public referrals. It had grown quickly, mostly with contract clinicians. One person was scheduling dozens of clinicians across several spreadsheets, the same family was typed in again at every stage, and every report was written from scratch.

The biggest piece is a report studio. The clinician's session transcript goes in, and a draft report comes out on the clinic's own template. Recommendations are chosen from the clinic's library to suit the person's age, stage and findings, the wording is tidied, and slips such as a mismatched pronoun are corrected. Reports for public referrals follow each region's own rules and layout, and recommendations for a school come out as a separate letter. Anything the clinician needs to confirm is flagged, with the point in the transcript it came from.

Alongside it is the financial and operational dashboard, built on the clinical system they already use rather than replacing it. It follows every case from enquiry to delivered report, with child and adult assessments tracked separately because they run on different timelines. It sets revenue against clinician costs, shows each clinician's utilisation with targets scaled down for new starters, and puts a value on the hours that went unused.

Clinical records stay in the clinic's own system, and what each person sees depends on their role. A person's name is replaced with a code before any text goes to the AI, and a clinician reviews and edits every draft before it's downloaded.

Illustration with sample data.

Charities · Ireland

Hands joined together

Where it stands: phase one is built and their existing case records have been brought in. The next phase is being scoped with them now.

A children's charity where no child's name ever reaches an AI model

This charity works with young people referred to it by state agencies. Referrals, case files, session reports and funder returns all lived in shared folders and spreadsheets. Reconciling the monthly finances alone took about four staff hours a week.

Now every young person gets one profile the moment they're referred, and their file, sessions, reports and progress build up there. It runs inside the charity's own Google account, in the EU.

AI helps workers tidy session notes and draft reviews, but names and other identifying details are taken out before any text goes to the model and put back afterwards. The code is set up so nothing else is able to call the model directly. Every draft waits for a worker to approve it. Records can be archived but never deleted, because children's care records have to be kept.

Illustration with sample data.

Automotive · United States

Cars in a dealership workshop

Where it stands: connected to the dealership's live system, read-only.

Catching warranty claims the manufacturer would reject, before they're sent

A US car dealership group loses money every time a warranty claim comes back rejected. The rules sit in the manufacturer's long policy manual, and the person checking each repair order is doing it by eye.

Our system reads every warranty repair order from the dealership's management system as it's written. It runs seven checks, including billing rate, labour code against diagnosis, overlapping work and claimed hours against the dealership's own catalogue. It scores the technician's write-up and suggests a better one. Each suggestion cites the page of the manual it relies on.

It never submits anything to the manufacturer. The warranty administrator decides what changes and submits from the dealership's own system, and every change is logged.

Illustration with sample data.

Fire and security · Ireland

A fire alarm call point on a wall

Where it stands: live. Website enquiries and public tenders arrive in it automatically. Connecting quotes to their field-service system is in progress.

A sales system built around how a fire and security company actually sells

The sales team was split between two off-the-shelf tools, neither of which matched their process: enquiry, site survey, estimator, quote, acceptance. The owner had no single view of the pipeline.

We built a CRM that follows those exact steps, with a mobile view for engineers on site and a leaderboard that supports the new commission plan. Enquiries from the website form become leads on their own. Every night it pulls new public tenders in Ireland that match their categories, so nobody has to search for them. Their field-service system stays the record for jobs. The CRM sits in front of it rather than replacing it.

Illustration with sample data.

Event security · Ireland

A security steward at work

Where it stands: the staff app is live in the app stores. Attendance, costing and payroll preparation are built and being tested.

Staffing an event from the first shift offer to the invoice

An event security and stewarding company has to fill hundreds of shifts, prove who was on site and when, check that every licence is in date, and then bill the client and pay the staff. They were doing it across messages, spreadsheets and a system they kept patching.

Staff now get shift offers, rosters, zone chats and onboarding in one phone app. On the day they check in and out with their location, and only inside the shift's hours. The office sees attendance as it happens, and the system prepares costing documents in the client's own layout, payroll and invoices from the same records. Licences due to expire are flagged eight weeks ahead.

Illustration with sample data.

Construction · Ireland and the UK

A site manager checking drawings

Where it stands: the hub is built. The tender reader is being tested on real tenders and goes into use with their estimating team in November.

AI for construction companies →

A specialist contractor that was missing tenders in the volume of email

This specialist contractor works on sites across Ireland and the UK. The pipeline lived in spreadsheets, and much of it only in people's heads. An off-the-shelf CRM had been tried and dropped because it couldn't follow the way they bid. Tender deadlines were being missed simply because of how much email came in.

We built them a hub around their own process: an eighteen-stage pipeline from first enquiry to handover, in euro and sterling. When a contract is signed, the payment schedule is read out of the PDF in under 30 seconds and turned into milestone invoices with the right VAT for each country, and those feed a cashflow forecast. Every job they win gets a delivery workspace with the same thirteen sections, from the agreement and health and safety file to site photos and the handover pack.

The newest part reads tenders. A tender pack can run to dozens of documents and scanned drawings. The hub reads the whole pack and produces a Word summary and a draft bill of quantities for the estimators, then a second pass checks the points that matter most against the source documents, page by page. It never submits anything, and the contractor's own rates are never sent to the AI.

Illustration with sample data.

Waste tech · Ireland

A waste tech business: the AI hub where it started

It's a waste tech business running more than 30 vehicles across several sites. It was the first business we built AI into, and it's still our largest system: nine agents on one shared set of data. Between them, they brought its costs down by about a quarter.

BillingDebtorsCustomer queriesReconciliationReportingRoutesEmail triageQuotesMonth-endHubone set of data

Operations

Nine agents working from one set of data

Billing, debtors, customer queries, reconciliation and reporting each have their own agent, all reading the same data from Salesforce, Xero, the route system and the customer portal. Jobs that took two days every month now run in under an hour. Each new agent took less time to build than the one before, because it could reuse what was already there.

Finance

Matching Salesforce invoices to Xero payments

Salesforce and Xero don't talk to each other, so a senior accountant spent four days a month matching payments to invoices by hand. The agent now does the matching, splits part-payments across invoices in the right order, deals with Relevant Contracts Tax on construction work, writes the result back to both systems and puts anything it isn't sure of in front of a person.

Finance

Month-end close in hours, not days

Account tie-outs, matching between group companies and foreign currency revaluation used to take a senior accountant several days. They now run automatically and leave an audit trail behind them, and the accountant reviews the result.

The founder's inbox

200 emails a day, and the few that matter

Most of them never needed the person running the business. They were newsletters, notifications and things someone else could answer. Now the inbox is sorted before it's opened, replies are drafted in the founder's own style for approval, and messages from key customers always get through. That gave back about eight hours a week.

Thirty minutes, on Dublin or Atlanta time.