Real projects, measurable results

Claude Code in the hands of a creative team

The situation

Two problems, both hitting the ceiling of the agency's growth. Turning a brief into finished copy took two and a half weeks on average, with writers spending hours on repetitive work, producing variants and reconciling versions. Capacity could not scale by hiring alone.

The less visible problem was that a great deal of creative thinking never left the table, because building it was expensive. Interactive microsites, personalised generators and custom campaign tools all needed a developer, weeks of time and real budget. The creative team had started censoring itself, no longer proposing ideas that would not fit the available development capacity.

The approach

We worked in two directions at once. First, speeding up the copy workflow with prompt engineering and a brand voice framework, where the agency's own tone was translated into instruction blocks embedded in every AI interaction. The output became consistent and immediately usable rather than erratic.

The bigger shift came from Claude Code. Writers, art directors and strategists learned to build working tools, pages and generators without coding knowledge. What used to be a development sprint became an afternoon.

The result

Copy capacity increased, but the deeper change was in what the agency could offer at all. The creative team can now propose and deliver solutions that previously only larger agencies could afford.

Rebuilding a contractor's back office with AI

The situation

The office was bleeding on three fronts. A mid sized competitive bid took three to five days to assemble, largely copied by hand from previous projects. Project data, contracts, subcontractor records and technical specifications lived scattered across different systems, spreadsheets and email. Invoicing and financial reporting were manual, slow and error prone, and no project manager could see their financial position in real time.

The approach

After the AI Transformation Day we worked on all three areas in parallel, building custom tools for the company's actual processes. An AI assisted bid preparation workflow draws on previous project data and the company's own templates, so the project manager only supplies what is specific to the job. Automated connectors bring project status, subcontractor delivery and contractual deadlines into a single live view, with data quality checks and anomaly flagging. On the finance side, automatic invoice generation from delivery data, financial summaries for project managers and leadership, and live settlement status reporting.

The result

The engineers went back to site. The administrative and finance team now solves problems instead of chasing data. Bids go out faster, invoices are more accurate and on time, and project managers see the financial position of their own projects in real time for the first time.

An internal AI knowledge base in a hotel

The situation

Most of a hotel's operation is knowledge intensive. Guests ask questions, colleagues ask each other, and new staff needed weeks of shadowing before they could work independently. The knowledge lived in people rather than in a system. With an experienced receptionist on holiday, a new colleague could not find the room service pricing, the late checkout rule, the spa closing time or the allergy options in the restaurant. Policies existed in different folders and nobody was certain which version was current.

The approach

We built a retrieval based internal knowledge base on the hotel's own documents, with a single authoritative source per topic and a clear ownership model for keeping it current. Every department can ask questions in natural language and get an answer grounded in the approved document, with the source shown.

The result

Onboarding time dropped sharply, and the answers guests receive became consistent across shifts and departments. Management stopped writing the same clarification emails to each department separately.

Microsoft Copilot rollout at a 200 person international robotics subsidiary

A structured Copilot adoption programme delivered in English, covering practical daily use across engineering and administrative functions, with role specific training rather than a generic product demonstration.

Copilot adoption and process optimisation at a logistics multinational

Copilot rollout combined with process review, so the tool landed on workflows that had been examined first. Training was built around the tasks the teams actually repeat.

Custom AI knowledge base and process automation at a 1000+ person enterprise

A tailored internal knowledge system with process automation on top, designed for an organisation large enough that document ownership and access control mattered as much as retrieval quality.

When translation loses the film, and a custom AI tool gives it back

A purpose built AI tool for subtitle and localisation work, where generic machine translation flattened tone and lost the intent of the original. The solution kept the human editor in control while removing the mechanical part of the job.