Work in practice

THE WORK, NOT THE PITCH.

Real client work, and what actually changed because of it. Cases where a stuck initiative became a changed way of working. The work, the result and what the client team owns now.

900-person cleaning company · Tenders & vending · Horizon 1 · AI Adoption

AI ADOPTION WHERE THE MARGIN IS, NOT THE INBOX.

A 900-person cleaning company, low-margin and volume-driven, where AI had no obvious place to start. So we didn't start with the inbox. We started with the two jobs that actually carry the margin: tenders and vending, analysed on their own data, with the team trained to run it themselves.

01Situation

The company had grown fast, its processes hadn't. Two jobs ate the most time and margin. Preparing one tender took an owner two to four days of manual calculation and German free-text answers, a few times a year. The coffee and snack machines were run by hand in a spreadsheet, refilled several times a day with no forecasting. The team was curious about AI, but had no idea where to start on the work that actually matters.

02What we did

No demo. We worked through their real tender set and vending data first, then ran a three-hour hands-on session for eight people, from owners to the vending manager, on their own work. They left with a repeatable, AI-supported way of doing both jobs: two analysis reports, two dashboards, an eight-step tender playbook with ready-to-use prompts, and a tender bid pre-filled with industry standards, down to the strategic choices only a person should make.

03What it delivered

Two expensive, undocumented processes became a repeatable approach the team was trained to run itself, with the person on the strategic call and AI doing the execution around it. Adoption landed where it rarely does first: an operational, low-margin business, on the jobs that make the money rather than the office tasks around them.

2–4 days per tenderThe manual work of one bid, resting on a single owner. We handed it back as a repeatable, AI-supported process, with the team trained to run it.

How we do this: Horizon 1 · AI Adoption →

European energy company · Pricing · Horizon 2 · Process Redesign

A PRICING TASK THAT TOOK DAYS NOW RUNS IN MINUTES.

A pricing analysis the team ran on repeat: their charging stations scored and priced by hand every round. The team rebuilt that logic as a model, with us in the work alongside them, and proved it produced the same prices, station for station.

01Situation

Every pricing round, each of their charging stations was scored on usage, local competition and price position, then classified into a pricing group and priced by hand before a price could be signed off. The logic lived in one team's spreadsheets.

02What we did

The pricing team rebuilt that logic as a model in code: every score, threshold and rule made explicit. They tested it against the stations the team had already priced, to confirm it produced the same result.

03What it delivered

Each pricing round took up to four days. It now runs in minutes, and the automated model reproduced the team's classification. The logic no longer lives in one spreadsheet: it is explicit, checkable and repeatable. The capacity it frees is the real prize: room to refine the model and move toward dynamic pricing: prices that respond to demand and competition as the market shifts from filling stations to competing on them.

Days → minutesThe automated model reproduced the team's classification.

How we do this: Horizon 2 · Process Redesign →

European energy company · Reimbursements · Horizon 2 · Process Redesign

THOUSANDS OF STUCK PAYMENTS, EACH BLOCKED FOR A DIFFERENT REASON.

Reimbursements owed for the power its charging points had delivered were piling up unpaid: thousands of them, each held up somewhere in the data across three separate systems.

01Situation

A stuck payment could be blocked by any of more than ten different faults: a missing contract, a mismatched date or a price that never loaded. The cause could sit in any of three systems. Working out why one payment was stuck took the team 15 to 30 minutes of cross-checking before anyone could act on it.

02What we did

The team that owns the process worked through the backlog with us, turning it from an undiagnosed pile into a sorted worklist: every payment grouped by what was blocking it, with the fix set out for the cases ready to clear.

03What it delivered

The cause now shows up front, so the team works down a ready list of fixes instead of chasing each payment across systems. The route to clearing the rest automatically is mapped, ready for them to run.

28%of cases needed several fixes applied in the right order: the reason a 3,187-station backlog never cleared by hand.

How we do this: Horizon 2 · Process Redesign →

950-person accountancy firm · Horizon 1 · AI Adoption + Horizon 2 · Process Redesign + Horizon 3 · New Business Models

A FIRM-WIDE AI TRANSFORMATION.

01The problem

Margin pressure leaves the firm no choice but to automate. But it bills by the hour: every hour AI saves is an hour it can no longer invoice, and salaries don't fall with the hours. The better the transformation works, the faster it eats the model that pays for it.

02How we're running it

Four parts, one flywheel. Not four workstreams running side by side.

Adoption, deepened per service line. We meet everybody where they are on the adoption ladder: a safe first try, then habitual use, then depth inside their own discipline, in clinics built on the workflows people actually run. Each service line names its own ambassador and two sherpas, and the firm makes them recognised positions with protected time. Adoption lives in the service line, not in the hub (the central transformation office).

The core work, rebuilt per service line. Every service line writes its own transformation roadmap with the hub, starting from the end-to-end process rather than a list of use cases. The board signs off on the direction, and each service line gets a small innovation team: five part-time roles with the time taken out of their other work, someone to run it, two builders, a data specialist and someone who knows the process end to end. They all follow one route from idea to production, through a single governance gate that decides on the spot. The hub owns the standard, the architecture and the reusable core, and hands the process over across three cycles: leading the first, coaching the second, stepped back by the third.

A new earning logic. One service line, ring-fenced, end to end, with a limited set of clients. Three candidates on the table: charging for the outcome, for access, or for the product. No model gets chosen before the pilot has run, and an explicit go/no-go comes before anything scales.

Leadership goes first. On every rung, before anyone below is asked to climb it. A change story per service line that names the job-loss fear instead of soothing it, and a dashboard that measures value and enjoyment of the work, not logins.

03What it delivered

Horizon 1 · AI adoption · measured. Nine in ten people across the firm now work with AI, and not shallowly. Nobody was ever ordered to. The programme is what moved it, from 105 volunteers in the first pilot to firm-wide use today. In that pilot, 79% said the work had become more enjoyable, and participants reported freeing 162 minutes a week on recurring work against a 45-minute target.

Horizon 2 · Process redesign · running. Every service line now has its own transformation plan, and teams across the firm are rebuilding the workflows they run every week. The hub owns the standard and the route to production; each service line has its own innovation team beside it. Governance is in place, and built to speed the work up: free experimentation in a sandbox, strict rules in production, one gate that decides on the spot. In a firm a regulator audits, that is what makes the pace defensible. The first redesigned workflows are live and carrying real volume: a fifth of all platform traffic now runs through assistants the firm's own people built, led by tools for audit files and year-end compilation.

Horizon 3 · New business models · under way. The billing question is not solved, and adoption will not solve it. The experiments are running now inside the ring-fenced service line, with the go/no-go still ahead. That is the decision that determines whether the transformation pays for itself.

162 min / person / weekself-reported in the first pilot, against a 45-minute target. 79% said the work became more enjoyable, and 96% stayed active, voluntarily. Firm-wide today: 869 active users against a headcount of 950, counted in the system.

How we do this: Horizon 1 · AI Adoption → Horizon 2 · Process Redesign → Horizon 3 · New Business Models →

Bring us the initiative that is stuck.

Your pilot proved the technology. Nobody funded the change around it. That is the part we do, and it starts from what you already built. We look at the people, the technology and the business system around it, and tell you what to move first.

Discuss what's blocking progress 30 minutes with a founder. No deck. Not ready to talk? Take the 5-minute AI Horizon Scan →

hello@h3.company