Short answer
In May 2026, a retail group deployed an AI that holds singular, memory-based conversations with each person. The system remembers past statements, job context, and local details, then surfaces hidden expertise across the workforce. A director of people development explained that managers filter information unevenly, so the tool talks directly to colleagues. It identifies who speaks Mandarin, knows SAP, has Manhattan experience, turning informal knowledge into searchable company intelligence without relying on manager recall.
Why do managers struggle to capture what employees really know?
Managers at different levels interview colleagues inconsistently. In May, a chief people officer told us that some directors help their teams well, others less so: the quality of support depends on who you report to. Information reaches HR through a manager filter, which means nuance and detail get lost. One director said she wanted to talk directly to the field rather than rely on secondhand summaries.
When a retail group tried to map skills, they found that asking managers produced incomplete answers. A manager might remember that someone on the team speaks a second language, but not which one, or that a colleague worked with a particular system years ago, but not the details. The group needed a way to ask every individual the same structured questions and store the answers in a form the business could query later.
How does a memory-based conversation differ from a survey?
A survey asks the same questions of everyone and discards context. A memory-based conversation adapts to the person: it knows their job title, location, business unit, and what they said in previous exchanges. In June, a people partner described testing a system that reminded a sales assistant, two weeks after an earlier check-in, whether a storefront task had been completed. If someone mentioned wanting to learn English in February, the system raised it again during the annual review preparation in May.
The difference matters because it turns a one-off form into a thread. A colleague who says they handle SAP but not PIM sees that detail reflected in later conversations. The system builds a profile over time, capturing not just static facts but changes in role, aspiration, and workload. One director noted that the more you use it, the more accurate the memory becomes.
What kind of knowledge does the system surface?
The system identifies skills, languages, system experience, and local process knowledge that would otherwise stay invisible. A director explained that if she needs someone who speaks Mandarin, knows SAP, and has Manhattan experience, she can query the entire workforce rather than ask around the coffee machine. The tool indexes every conversation, so a search returns names, context, and the exact exchange where the detail appeared.
It also surfaces process expertise. When a store in one country outperforms others on customer focus, the system compares conversations across regions, identifies what high performers do differently, and suggests which colleagues in the stronger market could coach teams elsewhere. In one example, the AI proposed operational shifts and named specific people to lead each workstream, all drawn from internal interviews rather than external benchmarks.
How do you prepare a conversation that works for both a shop assistant and a chief officer?
You configure tone, structure, and depth separately for each campaign. A director of people development showed us templates where the AI can be supportive or demanding, forensic or surface-level. For a chief executive preparing year-end reviews, the system was initially rigid; the executive said he would throw the laptop if it refused to move on when asked. The team adjusted the tone, and the system learned to respond to "next question" without resistance.
For store colleagues, the system uses a weekly check-in format that asks about tasks, sales targets, and immediate challenges. For senior managers, it walks through competency ratings, key performance indicators, and team dynamics. The same platform runs both, but the persona, methodology, and question flow change to match the audience. One director said the goal is to give everyone a structured conversation, not forms for hundreds of people.
What happens when a manager and a colleague prepare separately?
Both receive a PDF summarising their answers, and they meet with shared context already established. A director explained that the system runs mirror interviews: the manager reviews the colleague, the colleague reviews themselves, and the AI compares the two. If someone rates themselves highly on every dimension while the manager sees gaps, the discrepancy appears in a dashboard. The director can then ask why the differences exist and whether the colleague overestimates their position.
The approach reduces the risk that a manager avoids difficult conversations, then asks HR to remove someone without documentation. One director said that in France, employment law makes dismissal hard unless you have objective evidence. The system logs every exchange, attributes every grade to a specific quote, and builds a file that supports decisions without relying on memory or goodwill.
How do you turn thousands of conversations into actionable intelligence?
The back office analyses every interview against the company's competency framework. A director showed us a dashboard that assigns each person a position on eight dimensions, compares them to their business unit, and flags misalignments. If someone in payroll says they no longer reconcile accounts, the system suggests amending the job description and asks whether to apply the change to everyone with that title or just that individual.
Alerts highlight colleagues at risk of burnout, workload overload, or disengagement. One director explained that the system calculates an indicator of concern per person and per manager. If a manager's entire team shows declining engagement, HR sees the pattern and can intervene. The system does not make decisions; it surfaces signals that a human then investigates.
What does a knowledge query look like in practice?
A director asked the system which entity performed best on customer focus. The AI returned a ranking, then explained how the top performer could help the weakest. It proposed changes, identified the internal practices that drive the gap, and named colleagues in the stronger market who could act as champions. The entire answer drew on recorded conversations, not external research.
Another query asked which colleagues had mentioned a specific software skill in the past six months. The system returned names, the date of each mention, and the context. A comparison of regions on communication competency listed the distribution by department and suggested training topics based on the gaps. One director said the goal is to make the company interrogable: if you need a skill, a process insight, or a comparison, you ask, and the system answers from its accumulated knowledge.
How do you keep job descriptions current without a manual review cycle?
The system detects when someone's actual work diverges from their formal description. If a colleague says they no longer make photocopies, the AI flags the misalignment and offers to update the description. You can apply the change to that person alone or to everyone with the same title, and you can push the amendment back to your HR system or keep it internal.
One director explained that they generated initial descriptions using AI because complete, current files did not exist. As people complain that a task is wrong, the system refines the description. The approach accepts that job content changes faster than documentation cycles, and it uses ongoing conversations to close the gap.
Can you compare someone in Asia to someone in the US without local bias?
The system challenges vague or subjective input until you provide examples. A director demonstrated an evaluation where the manager rated someone poorly on organisational knowledge. The AI asked for a specific failure, refused to accept "he's useless," and kept pressing until the manager described a concrete incident. The director said the goal is to prevent a manager in one country from marking everyone perfect while a manager elsewhere grades harshly, making cross-region comparison meaningless.
The AI also injects process memory. If a store assistant says sales are below target, the system reminds them of the company's standard levers: offer the loyalty card to every customer, demonstrate hero products, greet visitors proactively. It adapts the reminder to the store location, so a flagship site hears different priorities than a small outlet. One director said the system ensures that the input quality is high enough to support fair comparison, rather than analysing poor data and calling it insight.
What skills do you need in HR to use this kind of tool?
You need to frame the right questions, interpret the signals the system surfaces, and decide when to act. A director said the AI does not replace the human relationship; it handles the process, the regular check-in, the administrative follow-up, and the manager focuses on the conversation that matters. Another director compared it to an exoskeleton: the tool does not make you stronger by itself, but it amplifies what you already do well.
You also need to configure tone, structure, and memory rules so the AI reflects your company's voice. One director spent months adjusting how the system challenges input, when it moves on, and how it reminds people of past statements. The work is not technical in the software sense, but it requires judgement about what kind of conversation will get honest answers from a shop floor colleague versus a senior executive.
How do you train people using the knowledge you have just captured?
The system suggests training topics based on the gaps it identifies, then generates the content. A director showed us a studio interface that produces a module in text, podcast, print, comic strip, or short video. The AI writes most of it, and you edit the rest. One example turned a customer-approach insight into a short clip, a two-page print handout, and a narrated capsule, all from the same source material.
The content draws on internal best practice rather than generic advice. If one region excels at converting browsers into buyers, the system analyses what those colleagues said in their interviews, extracts the method, and builds a training module around it. A director explained that the goal is to use the company's own genius to teach the company, not to buy off-the-shelf courses that ignore local context.
What stops this from becoming another unused HR system?
Launching a campaign takes a few clicks. A director demonstrated: choose a template, select the audience, set the start date, and the system writes the invitation email. If adoption is low, the barrier is not complexity. One director said that if using the tool gives you a headache, people will not return, so every interaction has to feel easy.
The other factor is relevance. If the system asks the same questions every time and ignores previous answers, people stop responding. Memory makes the conversation feel personal, and personal conversations get better answers. A director noted that the more you use it, the more the system knows, and the more useful it becomes. The loop reinforces itself, but only if the first experience is good enough to bring someone back.
A framework for deciding whether memory-based conversations fit your organisation
Before you deploy a system like this, ask four questions:
- Do managers currently capture employee knowledge unevenly, leaving gaps in your understanding of who knows what?
- Can you define the competencies, skills, or process insights you want to track, or do you need the system to help you discover them?
- Will your culture accept an AI asking follow-up questions until it gets a concrete answer, or will that feel intrusive?
- Do you have the HR capacity to interpret alerts, update job descriptions, and act on signals, or will the data sit unused?
If you answer yes to the first two and can address the second two, memory-based conversations will surface knowledge you cannot reach any other way. If you answer no to the first, a simpler survey will suffice.
Research published in 2025 showed that daily knowledge sharing expectations are positively related to employees' daily knowledge sharing, with the strongest effect size. Knowledge management describes the process of capturing and using organisational knowledge systematically, a challenge that grows as companies scale beyond what any single manager can hold. A study on knowledge management impact found a significant and positive impact of knowledge management processes and approaches on job satisfaction and work performance.
