How we work
We act as a technology partner to each client. The client keeps its judgement, its relationships, and its obligations, and we provide the hosted service and the engineering behind it.
The stages of an engagement
Software projects, AI configurations, and website work follow the same stages. The client decides at the end of each stage before go-live whether to continue, and each of those stages ends with a result the client can review. The subscription that follows runs for the term in the client's agreement.
- Discovery
We meet the people who own the problem and the people who will use the result, and we learn where time and risk concentrate. The client receives a written summary of what we recommend and the order in which we would do it.
- Scope and price
We agree a written scope, the data settings, and the price before work begins. Work that depends on the outcome of an earlier stage is identified and priced separately.
- Design and build
We design with the people who will use the software, build in short cycles, and show working software at the end of each cycle. For a website, the client approves the design and every page before launch.
- Test and measure
We test against acceptance criteria agreed in the scope. An AI configuration is also piloted on live work under controlled conditions, and we report the results against gates agreed beforehand, with their sample sizes.
- Run and support
We host and run what we build as a service, with support, monitoring, and updates as the client's needs and the rules change.
Terms of our work
Ownership and subscription
Our software and the websites we build are services that we host and operate, on subscription. ZCM owns the software it builds, including the work done for a particular client, and each client owns its data, its content, and its outputs.
Client data is not used to train models, and it is not used for any other client.
Measurement
Each AI product will be released for general use only after it passes defined gates, which cover:
- accuracy against labelled test sets
- repeatability across repeated runs
- response times
- behaviour under adversarial input
Each result will be reported with its sample size. Time saved on real work will be measured after release and reported in the same way.
Our delivery team
Product design, architecture, security, and client engagement are directed by our team in Hong Kong. Engineering and design work will be carried out with a vetted network of specialists across Asia, under our direction and subject to code review by our own team.
Their terms bind them to confidentiality and give ZCM ownership of their work. Engineering partners will work on code and test data, and they will not receive access to client data in production.
Where we work from
ZCM operates from Hong Kong. We work with clients in other jurisdictions by video conference, and we meet clients in person in Hong Kong.
Working within our clients' obligations
The firms and businesses we build for are often regulated. What we build is designed around the obligations they carry, and our terms say so.
Professional responsibility
Our products support practitioners in delivering their services. Responsibility for advice, and for any service that requires registration, remains with the practitioner who provides it.
Courts and tribunals
Where a court requires disclosure of the use of generative AI, the platform's records are designed to show which tool produced which content and how that content was checked.
Privacy
We will act as a service provider to each client under written terms covering storage location, access, retention, and breach notification, which support the client's own privacy obligations.
Records
Audit records and reports will be retained for the period the client sets, which allows the client to show its process if a decision is later questioned.
Talk to us about what you need
We meet clients and business partners by video conference or in person in Hong Kong.