Human in the loop makes AI work for people

A quick insight: Human in the loop HR AI means using AI to reduce analysis and action-planning workloads while people retain responsibility for interpretation, decisions and conversations. In employee surveys, this balance matters because feedback shapes real decisions about people. This guide explores where AI adds speed, where human expertise adds value and how People Insight combines Prism with expert support to turn feedback into trusted action.

AI is changing how organisations work withsurvey results. It can analyse comments in seconds, identify important patterns and help managers turn broad intentions into practical actions.

The challenge is deciding how much responsibility to give the technology and where people need to stay actively involved.

That is where a human in the loop approach comes in. It keeps people at the heart of decisions while using AI to reduce manual work and make insight more accessible. Technology brings speed, consistency and scale. Humans bring organisational context, empathy and judgement.

Both are needed to make employee feedback meaningful and ensure the actions that follow are appropriate.

Across our benchmark data, 68% of employees say senior leaders make the effort to listen, yet only 59% believe action will be taken as a result of surveys. This gap shows why a balanced human in the loop model matters. Organisations do not simply need faster analysis. They need the confidence and support to turn insight into decisions employees can trust.

Related: 10 pros and cons of AI in the workplace

What human in the loop looks like in real work

Human in the loop means technology handles work that can be completed quickly and consistently, while people guide the process wherever context, judgement or sensitivity is required.

At People Insight, Prism is built into our employee survey platform to support organisations throughout the journey from listening to action.

It analyses qualitative and quantitative feedback together, helping leaders understand the themes, drivers and signals that matter. Prism Suggest recommends practical actions based on survey results, the manager’s level and the organisation’s priorities. Prism Improve helps managers strengthen their own ideas by turning them into clearer, more specific and measurable actions.

Prism Context makes this support more relevant by incorporating the organisation’s priorities, language, people initiatives, values and culture. Rather than producing generic recommendations, Prism can shape its interpretation and guidance around what matters to that organisation.

People remain in control throughout.

HR teams review organisation-wide findings and sensitive themes. Senior leaders agree strategic priorities. Managers discuss results with their teams, choose which actions are appropriate and adapt them to their local circumstances. People Insight consultants can provide an additional layer of expert interpretation, challenge and support.

This combination allows AI to make the process faster without allowing technology to make consequential people decisions by itself.

Research from MIT Sloan points to this complementary relationship: AI contributes scale and pattern recognition, while humans contribute context, creativity and ethical judgement.

Related: 10 AI prompts to get the most out of your employee surveys

Where agents shine and where people add value

AI is particularly useful when a task involves large quantities of information, repeatable analysis or structured guidance.

An organisation may have thousands of survey comments spread across departments, locations and employee groups. Reviewing all of them manually can take weeks and may produce inconsistent interpretations. Prism can analyse scores and comments together, identify recurring themes and help leaders see where attention is most needed.

It can also make action planning more accessible. Managers do not always struggle because they lack good intentions. They may be short on time, uncertain what action will make a difference or unsure how to turn a broad ambition into a well-defined commitment.

Prism Suggest and Prism Improve provide practical support at this point.

People add value when the meaning of the feedback is complex, sensitive or specific to the organisation.

A low score for trust may have a very different explanation in two departments. Comments about workload could relate to resources, unclear priorities, inefficient systems or a period of organisational change. An AI-supported interpretation can reveal the pattern, but people must consider what sits behind it and what response is realistic.

Human judgement is also essential when:

• deciding which priorities fit the wider organisational strategy
• interpreting sensitive or potentially distressing feedback
• considering the impact of an action on different employee groups
• communicating difficult findings
• involving employees in deciding what happens next
• checking whether a suggested action is genuinely achievable
• deciding when further investigation or specialist support is needed

Harvard Business Review has noted the limitations AI still faces when dealing with tacit social cues and ethical judgement. Human oversight is therefore especially important when technology is being used to support decisions affecting people at work.

Guardrails that protect employees and your organisation

Good governance turns human in the loop from a principle into a practical way of working.

Within an employee listening programme, this can include role-based permissions so managers only see the data they are authorised to access, minimum reporting thresholds that protect anonymity and clear rules establishing which decisions must always be made or approved by people.

Prism has been designed to support responsible use within complex organisations. Personally identifiable information is removed before AI analysis, safe-group reporting protects anonymity and role-based access helps organisations control who can see different levels of data.

Data is encrypted in transit and at rest, is not used to train external models and is supported by audit trails that improve accountability.

However, technical safeguards are only one part of responsible HR AI. Organisations also need clear internal expectations covering:

• what AI can and cannot be used for
• who is accountable for reviewing its outputs
• when human approval is required
• how employees will be told that AI is involved
• how potentially biased or unsuitable suggestions will be challenged
• how decisions and actions will be documented

The Information Commissioner’s Office identifies fairness, transparency and accountability as central principles when AI is used in decisions involving people. The UK government’s Algorithmic Transparency Recording Standard similarly encourages organisations to document what algorithmic systems do, why they are used and how risks are managed.

Practical guardrails like these strengthen confidence without preventing organisations from benefiting from the technology.

Practical flow from feedback to action

Start with a straightforward promise to employees: you will explain what you heard, agree a small number of meaningful priorities and report back on what happens next.

Your employee survey platform and Prism can make each stage easier, but people should remain accountable for the decisions and conversations that follow.

Prism Context: make AI relevant to your organisation

Generic advice is one of the biggest limitations of off-the-shelf AI.

An action that is useful in one organisation may be unrealistic, inappropriate or already underway in another. The language used in an action plan must also feel credible to the people expected to deliver it.

Prism Context gives Prism a clearer understanding of the organisation in which the feedback was collected. This can include:

• People Initiatives: programmes and priorities already in place, such as leadership development, wellbeing support, communications activity or changes to ways of working
• Values & Culture: the values, behaviours and cultural expectations the organisation wants to reinforce

This context helps Prism interpret employee feedback against real organisational priorities and produce guidance that is more relevant to leaders and managers.

Human input remains fundamental. HR and leadership teams decide what organisational context should be included, keep it accurate and assess whether the resulting interpretation reflects what is really happening.

Prism Suggest: identify practical next steps

Understanding a problem does not automatically tell a manager what to do about it.

Prism Suggest recommends actions for individual survey questions and priority areas. Suggestions reflect the user’s role and level, the organisation’s context and established employee experience best practice.

This means a team manager can receive guidance suited to changes within their direct control, while a senior leader can be shown actions appropriate for broader organisational priorities.

Managers can review each recommendation, decide whether it fits their circumstances and add it to their action plan when it is appropriate.

The suggestion is a starting point, not an instruction. Managers still need to consider local pressures, consult their teams and take ownership of what they commit to doing.

Prism Improve: turn good intentions into stronger actions

Managers often know what they want to improve but find it difficult to define exactly what they will do.

An action such as “communicate more” may reflect a genuine priority, but it does not explain what will change, who is responsible or how employees will know that progress has been made.

Prism Improve acts as an action-planning coach. When managers write their own actions, it helps refine them so they are more specific, measurable and time-bound.

For example:

Initial idea: Improve communication within the team.

Stronger action: Introduce a 15-minute fortnightly team update from September, covering current priorities, upcoming decisions and progress against actions raised in the employee survey.

The manager remains responsible for the action. Prism Improve provides structured support that helps turn an intention into something employees can see and assess.

Related: Turning survey comments into clear actions with AI

Human support

Human in the loop does not end when a manager reviews an AI-generated suggestion. People are needed throughout the survey process to challenge assumptions, interpret findings and support meaningful conversations.

People Insight combines its technology with employee survey consultancy, giving organisations access to experienced consultants who understand employee feedback, organisational behaviour and the practical realities of change.

Our consultants can help organisations:

• interpret survey findings within the wider organisational context
• distinguish meaningful patterns from short-term noise
• challenge conclusions that are not adequately supported by the data
• identify realistic priorities rather than trying to act on everything
• prepare leaders to communicate difficult or disappointing results
• help managers turn suggested actions into credible commitments
• establish an action-planning approach that is consistent without becoming rigid
• review progress and determine whether action is changing the employee experience

Expert support is especially valuable where findings are politically sensitive, vary significantly between groups or reveal issues that cannot be resolved through a simple team-level action.

Human support also includes the people inside the organisation. HR teams establish the listening strategy and governance. Senior leaders make decisions about organisation-wide priorities. Managers provide local context and host conversations. Employees help test whether proposed changes will address what they actually experience.

Clear survey communications help connect all of this. Employees need to understand what was heard, what has been prioritised and why some issues may take longer to address.

CIPD guidance encourages HR teams to establish clear policies, oversight and accountability when using AI in people-related work. A human in the loop approach supports this by combining automation with visible human responsibility.

What a safe, human-centred results conversation sounds like

Picture a manager sharing survey results after a demanding quarter.

Prism has helped them identify workload as an important theme and understand some of the issues sitting behind the score. Prism Suggest has offered several possible actions. The manager reviews them, considers what is within the team’s control and discusses the findings with HR before meeting employees.

They begin by thanking the team for taking part and acknowledging what the results show:

“Many of you mentioned workload pressures, particularly when deadlines overlap. The results suggest we need to improve how priorities are coordinated, rather than simply asking everyone to work more efficiently.”

They explain two proposals for the next eight weeks: shorter meetings with clearer outcomes and a shared calendar showing major deadlines and pressure points.

Rather than presenting these as a finished answer, they ask:

“Would these changes address what you are experiencing? What would make them more useful? Is there anything we should stop or adapt?”

The manager then confirms who will own each action and gives the team a date for the next update.

AI made it faster to identify the theme and develop possible actions. The manager’s judgement, transparency and willingness to listen made the conversation credible.

Common risks when AI leads without people

The benefits of AI can quickly be undermined when organisations treat its outputs as decisions rather than support.

Below are some of the most important risks to consider.

Overreach from convenience

When a feature is quick and easy to use, it can gradually be applied to decisions it was never intended to make.

An AI tool that supports survey analysis should not automatically be used to assess individual performance, infer personal characteristics or make employment decisions.

Set clear boundaries around what the technology is authorised to do. Identify decisions that always require human judgement and make named individuals accountable for reviewing outputs before they influence action.

Lack of transparency

Employees are unlikely to trust a process if they do not understand how their feedback is being analysed or how priorities have been selected.

Organisations should be able to explain:

• where AI is used within the survey process
• what information it analyses
• what safeguards protect employee data
• who reviews the outputs
• how final decisions are made
• how employees can raise concerns

If a suggestion cannot be explained or defended, it should not be adopted simply because it was produced quickly.

Regulators in the UK and EU continue to stress the importance of transparency when AI may influence people’s working lives. Employees increasingly expect that transparency too.

Bias that shifts from people to patterns

AI can reflect biases within its source data or the way a task has been framed. Humans can introduce bias through assumptions, selective interpretation and organisational politics.

Keeping a human in the loop does not automatically eliminate bias. It creates an opportunity to identify and challenge it.

Managers and HR teams can reduce risk by asking structured questions:

• What evidence supports this interpretation?
• Are a small number of comments being given too much weight?
• Could different groups experience this issue differently?
• Are we interpreting the data in a way that confirms what we already believed?
• Does the proposed action create unintended consequences?
• Are anonymity thresholds strong enough to protect employees?

Sensitive demographic analysis should only take place where reporting thresholds allow it and where there is a legitimate reason to examine differences between groups.

As Harvard Business School research has shown, managing bias is an ongoing process rather than a one-off technical check.

Measuring whether the blend works

You do not need dozens of measures to understand whether your human in the loop approach is working.

A small number of practical indicators can show whether technology is saving time while people are still providing meaningful oversight:

• time from the survey closing to leaders receiving usable insight
• time from results being released to managers creating their first action
• percentage of actions with a clear owner and review date
• percentage of managers who have discussed results with their teams
• visibility of survey updates to employees
• quality of actions before and after support from Prism Improve
• completion or progress rates for agreed actions
• short pulse survey results on priority themes
• employee confidence that action will be taken
• governance checks covering transparency, fairness and human review

Speed should not be the only measure of success. A fast process that produces weak, generic or poorly communicated action will not strengthen employee trust.

The aim is to improve both efficiency and the quality of follow-through.

How to start human in the loop with confidence

Start with one survey cycle, a clear use case and defined human responsibilities.

Before launching, agree:

1. Which parts of the process Prism will support.
2. Which outputs HR or senior leaders will review.
3. What decisions managers can make independently.
4. Which sensitive findings require additional expertise.
5. How the use of AI will be explained to employees.
6. How actions and progress will be communicated.
7. How you will assess whether the approach has worked.

Use Prism Context to reflect your organisation’s current initiatives, values and priorities. Allow Prism Suggest to help managers explore practical next steps. Use Prism Improve to strengthen the actions managers create.

Then provide targeted guidance or coaching to the people who will host results conversations. Publish short, regular updates connecting actions to employee feedback and review whether employees are experiencing the intended improvement.

Once the process is working well, it can be extended to more teams, survey types or stages of the employee listening cycle.

CIPD’s work with employers highlights the importance of practical, policy-backed approaches to adopting AI in HR responsibly. Starting with a controlled and clearly governed use case allows organisations to learn without losing sight of the people affected.

Keeping people at the centre

AI can accelerate the journey from feedback to action. People ensure that journey remains fair, relevant and grounded in how the organisation actually works.

A strong human in the loop approach does not position people as a final safety check added after the technology has done its work. Human expertise shapes the context, reviews the interpretation, makes the decisions and leads the conversations throughout.

Prism helps organisations listen more sharply by making complex feedback easier to understand. Prism Context makes that understanding more relevant. Prism Suggest gives leaders and managers clearer options for what to do next. Prism Improve helps turn their ideas into stronger, more accountable actions.

People provide the judgement, empathy and ownership that turn those capabilities into meaningful change.

Together, they create faster insight without careless automation, more actionable feedback without generic recommendations and a survey process that builds trust through visible follow-through.

Ready to combine AI-supported insight with expert human guidance in your next survey? Talk to us about an employee survey that brings together Prism, consultancy and practical support for leaders and managers.