Correlation One's workforce development practice for the AI era — programs with USAID, the U.S. Department of State, and government partners, priced on verified jobs and wages instead of people trained.
U.S. Department of Labor Registered Apprenticeships in data, cybersecurity, and AI roles.
Develop the technical skills that Fortune 500 enterprises and the public sector want.
Access highly skilled professionals trained for real-world enterprise needs — at no cost.
Access pre-trained candidates at zero cost.
Enterprise AI Enablement · Governance
Policy and oversight define what safe AI use is. Employees decide, thousands of times a day, whether it happens. We build governance into every enablement programme so the control exists where the decision is made, and we measure whether it holds.
What it is: an approach, embedded in every Correlation One enterprise programme and available as a standalone governance-to-practice engagement, that translates an organisation's AI policy into controls employees apply at the keyboard: a data classification ceiling that holds in every format, a purpose test, a tool selection rule for multi-tool environments, evaluation and agent-testing skill, tiered build permissions by persona, deliberately placed human review, and memory hygiene. Governance is the first session of each programme and the pre-send checklist in every session after it, and we measure it with pre/post capability data, agent pass rates against a testing framework, and usage data.
Because the gap in most enterprises is not the policy. It is the distance between the policy and what employees can do with it.
Risk, legal, and security teams have written the acceptable-use policies and approved the tools. What they cannot do is sit beside each employee deciding whether a file can be uploaded, which tool to open, or whether an output can be trusted. In our 2026 engagement data, the strongest predictor of realised AI return is whether governance discipline and workforce enablement run as one integrated system. Where they run as separate initiatives owned by different functions, the policy is read once and the behaviour never changes. Our AI Impact and Maturity diagnostic weights organisational maturity (governance, oversight, measurement) more heavily than any other dimension for the same reason: it determines whether experimentation converts into governed performance.
Risk, information security, AI, and HR leaders; the employees and leaders their policies cover
Embedded in every programme tier, or as a standalone governance-to-practice engagement (typically 3–4 weeks before a cohort)
Your data classification scheme, acceptable-use policy, approved tools and connectors, sharing settings, and risk appetite
Employee-facing data boundary and purpose test, tool selection rule, evaluation and agent-testing framework, persona permission tiers, champion playbook
Tool-agnostic; implemented on Copilot, Gemini, Claude, ChatGPT, and client-specific tools
Pre/post governance capability, agent pass rate against the testing framework, usage data, incidents at the keyboard
Each is a decision an employee makes, stated so it can be applied in seconds and practised on realistic cases before live use.
The highest classification permitted in the tool, applied in every format including screenshots and pasted extracts. One ceiling across all tools.
Permitted data can still be used in a prohibited way. Approved purposes are named; out-of-scope purposes (HR decisions, comparing colleagues, working around blocked systems) are named too.
In multi-tool environments, which tool starts which kind of task, so governance does not depend on an employee guessing which assistant risk prefers.
Knowing when an output is wrong in a way that matters. Taught as deliberately as generation; every output checked against sources before it leaves the session.
Personas (for example Business User, Citizen Developer, Power User) with escalating permissions earned through demonstrated, tested artifacts.
The step where a person decides, named explicitly per workflow, with the person trained to decide well. Includes agent testing: accuracy in scope, graceful refusal out of scope.
We take the policy as written and turn it into something employees can apply, test it on their real files, and then install it in the first session of the programme.
We work with risk, security, and legal to confirm the current classification scheme, approved purposes, enabled features (memory, sharing, connectors), and the questions employees are actually asking.
We write the employee-facing versions: the ceiling, the purpose test, the tool rule, and the review points, each with realistic worked examples from the organisation's own work.
We define the build-permission tiers, the evidence required to advance, and the feature roadmap (what is enabled now, what comes later, who approves).
We design the output-checking method and the agent-testing framework the organisation will use, and turn the checking method into a reusable instruction set where the tools support it.
Governance becomes Session 1 of the enablement programme and a three-question checklist (permission, location, reasoning) before every task in every later session.
Pre/post governance capability is reported with the programme's other outcomes; champions receive the playbook and become the first point of contact for governance questions in their teams.
A global airline group deploying Claude alongside Copilot.
Four classifications with Confidential as the ceiling and Strictly Confidential prohibited in every format. A purpose test practised on five realistic files before any live task. Projects and Skills enabled with sharing disabled for the first cohort; connectors limited to an approved list; the roadmap for widening access published. Memory hygiene taught in Session 1 with the rule that memory never expands what the organisation permits. Human review identified per workflow in the connectors session. A capstone in which every demonstrated use case must pass the organisation's own checks before it is presented. Governance is not a module; it is the structure of the programme.
Because human review is only a control if the human can tell what to look for, and in our training rooms that is the skill most consistently missing.
Employees arrive with a reasonably calibrated sense of AI risk. At a leading North American bank, pre-workshop confidence in safe and responsible use measured 3.9 out of 5, against 2.6 for prompting. The governance gap is competence, not awareness. We therefore spend less time on risk messaging employees have already absorbed and more on the skills that make safe use possible: framing a task, grounding it in approved sources, and checking the result. Leading adopters are formalising this at the agent level, and we help them: one airline group tests every agent on two dimensions, accuracy in scope and graceful refusal out of scope; a life insurer runs a formal challenge for its actuaries on interpretability, fairness, and selection bias.
On whether the controls hold, not on whether the policy was read.
They define what an organisation must be able to show. This is how the organisation makes it true at the keyboard.
Recognised AI risk-management frameworks, management-system standards, and the regulatory regimes now in force set requirements for documentation, oversight, testing, and accountability. We do not replace an organisation's compliance programme or its legal advice. We make the requirements operational in the workforce: the controls above map to the human-oversight, data-governance, and testing expectations those frameworks share, and the measurement produces the evidence a compliance programme needs.
Translating an organisation's AI policy into controls employees can apply at the moment of use (data boundaries, purpose tests, tool selection rules, evaluation skill, build permissions, human review), installing them through training, and measuring whether they hold.
Policy and committees define what safe use is and who oversees it. Enablement is how the policy becomes behaviour: employees practise the controls on realistic cases, champions carry them into teams, and capability is measured before and after.
We work from the organisation's existing policy and risk appetite, with its risk, legal, and security teams, and translate it into employee-facing controls. We do not provide legal advice or replace a compliance programme.
It is tool-agnostic and has been implemented on Microsoft 365 Copilot, Google Gemini, Claude, ChatGPT, and client-specific tools, including multi-tool environments with one data boundary across all of them.
Pre/post capability on the governance skills, the share of agents passing a testing framework before pilot, activation and 90-day usage from the tool's own data, and movement on the governance dimensions of the AI Impact and Maturity diagnostic.
A structured test applied before an agent reaches pilot. A common two-dimensional version tests in-scope requests for accuracy against sources and out-of-scope requests for graceful refusal, with documented results and a named owner.
Yes, as a governance-to-practice engagement of three to four weeks that produces the employee-facing controls, persona permission tiers, and evaluation framework. Most clients then install the outputs in an enablement programme so the controls are practised, not only published.
The AI Impact and Maturity diagnostic benchmarks governance alongside talent and momentum, and shows where policy and practice have come apart.