Data and AI skills
The skills our games train and measure. They form a closed taxonomy, so your reports compare teams and games on the same basis.
Prompting
Giving context
State the role, the audience, the goal and the facts the model needs, so that it stops guessing.
Trained by:Double AgentPrompt Quest
Setting constraints
Set the length, format, tone and limits of the answer, and say what the model must not do.
Trained by:Double AgentPrompt Quest
Using examples
Show the model one or two examples of the expected answer rather than describing it at length.
Trained by:Double AgentPrompt Quest
Data quality
Spotting duplicates
Recognise records that describe the same thing twice, even when they are not written the same way.
Trained by:Data Cleaning Rush
Handling missing values
Notice empty or placeholder values and choose between completing, flagging or excluding them.
Trained by:Data Cleaning Rush
Fixing formats
Make dates, amounts, units and codes follow a single format so that they can be compared and computed.
Trained by:Data Cleaning Rush
Spotting outliers
Flag values that are implausible or far from the rest, and check them before correcting or keeping them.
Trained by:Data Cleaning Rush
Critical thinking about AI
Spotting hallucinations
Recognise invented facts, sources or figures stated with confidence in an AI answer.
Trained by:Double AgentHallucination Hunter
Detecting bias
Notice when an AI answer treats groups of people unfairly or presents a partial view as neutral.
Trained by:Hallucination Hunter
Knowing when to verify
Judge which AI answers need checking before use, and how thoroughly, according to the stakes.
Trained by:Prompt QuestHallucination Hunter
Using AI at work
Knowing when AI fits
Recognise the tasks where generative AI genuinely helps: drafting, rewording, summarising, exploring ideas.
Trained by:AI or Not AI?
Protecting personal data
Keep names, contact details and sensitive information about people out of AI tools that are not approved for them.
Trained by:AI or Not AI?Double Agent
Keeping a human accountable
Leave decisions that affect people, money or the law to a person who can explain and answer for them.
Trained by:AI or Not AI?Double Agent
Reviewing before publishing
Have a person read AI-generated content before it is published or sent outside the company.
Trained by:AI or Not AI?Double Agent
Choosing a script over a model
Recognise mechanical, repeatable tasks where a formula, a script or existing software beats a generative model.
Trained by:AI or Not AI?
Data leadership
Handling a data incident
Respond to a leak, an outage or a wrong figure in production: contain it, inform the right people, fix the cause.
Trained by:Who is the best CDO?
Complying with data protection
Keep projects and tools in line with data protection law: legal basis, minimisation, retention and people's rights.
Trained by:Who is the best CDO?
Steering data quality
Treat data quality as a steady effort with owners, measures and priorities, rather than a one-off clean-up.
Trained by:Who is the best CDO?
Investing for value
Choose the data and AI projects that bring measurable value for the business, and stop the ones that do not.
Trained by:Who is the best CDO?
Bringing people on board
Bring managers and teams on board so that data is used in everyday decisions, not only by specialists.
Trained by:Who is the best CDO?
Adopting AI responsibly
Roll out AI in the company with clear rules, the right safeguards and people who know how to use it.
Trained by:Who is the best CDO?
AI agents
Minimal data access
Give an AI agent access only to the data its task needs, at the narrowest scope that still works.
Trained by:Double Agent
Least-privilege tools
Give an agent the least power to act: read before write, draft before send, and no administrator rights.
Trained by:Double Agent
Untrusted content and injections
Treat what an agent reads (tickets, emails, web pages, CVs, forms) as data, never as instructions: it may hide an injection.
Trained by:Double Agent
Agent cost and latency
Size an agent to its budget and its expected response time: the biggest model and every tool are rarely needed.
Trained by:Double Agent