The Institutional Memory Problem
As one Chief People Officer described it: "Every HR team I have ever worked with is simultaneously under-resourced and drowning in repetitive work. They spend the majority of their time answering the same questions — 'What is our parental leave policy?' 'How do I request a reference letter?' 'Where do I find the performance review template?' — and what remains goes toward high-judgment work that only humans can do: navigating a difficult termination, mediating a team conflict, assessing a candidate's cultural fit. The tragedy is that the first category is exhausting and adds no value, and the second is chronically under-invested because the first has consumed the time and energy it deserved."
Ayesha Raza started her new role as Senior Data Analyst at an EdTech company in Karachi three weeks ago. She had fifteen questions in her first week. Every single one had a written answer somewhere in the company's documentation. She could not find any of them. She asked her manager, Omar Farooq, who answered four and forwarded the rest to HR. HR answered them, one by one, over the course of four days. The answers were helpful. The wait was not. And Omar spent time he did not have finding answers he should not have needed to look for.
This is not a story about a disorganised company. It is a story about a structural problem that affects almost every organisation: the gap between knowledge that exists in principle and knowledge that is accessible in practice. Understanding this gap — and understanding what AI can and cannot do about it — is the foundation for everything that follows in this chapter.
Explicit and Tacit Knowledge in HR
Every organisation accumulates knowledge in two fundamentally different forms. The distinction matters because the two forms create different problems, and require different solutions.
Explicit knowledge is written down. It exists in employee handbooks, policy documents, onboarding checklists, compensation bands, process guides, and org charts. It is theoretically findable — the problem is that it is scattered across a shared drive, an HRIS system, a Confluence wiki, several PDFs that may or may not be current, and three email threads from 2023 that nobody archived properly. Even when employees find the right document, the language is written for legal defensibility, not comprehension. The knowledge exists. It is practically inaccessible.
Tacit knowledge is the knowledge that governs how things actually work — the unwritten understandings held in people's heads. "The policy says annual performance reviews, but the real cycle here is quarterly." "The travel expense policy says a daily limit, but the actual practice is that nobody claims above a lower figure because the last person who did got a difficult conversation from Finance." "The formal process for remote work requests requires two weeks' notice, but in practice you send a Slack message and it is approved within an hour." This knowledge is not in any document. It exists only in the heads of long-tenured employees — and when those employees leave, it leaves with them.
| Knowledge Type | Where It Lives | Problem It Creates | When It Disappears |
|---|---|---|---|
| Explicit | Documents, handbooks, wikis, HRIS | Hard to find, hard to understand without help | When documents become outdated or inaccessible |
| Tacit | In people's heads, informal norms, unwritten practices | Invisible until you need it; never formally documented | When the person who holds it leaves |
HR operations are degraded by both. Explicit knowledge is inaccessible, so employees ask HR instead of finding it themselves. Tacit knowledge is invisible, so organisations lose critical operational context every time a long-tenured employee departs.
The Three HR Functions That AI Transforms
Not all HR work is equally suited to AI assistance. The three functions below represent where AI creates the most leverage — not because they are the most important HR work, but because they are the most expensive in time and the least valuable in outcome.
Function 1: Information Routing
The majority of employee HR interactions are information requests. Employees need to know their leave entitlement, understand the expense process, find the performance review template, or clarify the remote work policy. These questions have written answers. The problem is finding them.
Information routing is the function that, industry research suggests, consumes the largest share of HR team time [VERIFY] — answering questions that have written answers, to employees who cannot find or understand the written answers themselves. It is not high-value work. It is not even particularly skilled work. It is the administrative overhead that crowds out the judgment-intensive work that genuinely needs experienced HR professionals.
An HR Knowledge Base Agent — a persistent AI agent connected to the organisation's policy documents, handbook, and FAQs — handles information routing automatically. It answers questions instantly, consistently, and accurately. It cites the policy source. It knows when to say "I do not know — speak to HR directly." It frees HR team members from the hundredth iteration of the same question to spend time on the question that actually needs human judgment.
Function 2: Process Execution
HR processes generate an enormous volume of repetitive administrative writing. Offer letters share the same structure, with different names and numbers. Onboarding schedules follow the same programme, with different start dates and departments. Job descriptions use the same framework, for different roles. Reference letters follow the same format, with different achievements.
Every one of these is a document a skilled HR professional could write in 20 minutes — but should not, because the marginal value of the 200th offer letter is essentially zero and the 20 minutes spent on each is 20 minutes taken from genuinely high-value HR work.
AI does not improve these documents. They were already good. It produces them in two minutes instead of twenty, at the same quality, freeing the time for work that matters.
Function 3: Institutional Knowledge Capture
The most important — and most neglected — HR function in the AI era is knowledge capture: the systematic process of turning tacit organisational knowledge into explicit, searchable, updatable documentation before the people who hold it leave.
Every year, at every organisation, knowledge walks out the door with departing employees. A retiring subject-matter expert takes decades of domain knowledge with them. A departing manager's team context, decision-making frameworks, and key relationships are never recorded. An employee who has been at the company for seven years carries the unwritten rules that make the organisation function — and nobody thinks to ask them about it until it is too late.
AI makes the capture process substantially less burdensome. A structured knowledge interview, guided by an AI agent, can extract and document institutional knowledge in a fraction of the time it would take a human interviewer. The /knowledge skill in this chapter does exactly this.
What AI Changes — and Does Not Change
Understanding the boundary matters as much as understanding the opportunity.
| HR Activity | AI Can Handle | AI Cannot Handle |
|---|---|---|
| Holiday entitlement query | Answer instantly from policy | — |
| Offer letter for standard role | Draft in 2 minutes | Negotiate terms, read candidate psychology |
| Performance review structure | Generate questions, summarise feedback | Make the final performance judgement |
| Knowledge capture interview | Structure and document knowledge | Replace the relationship needed to surface it |
| Redundancy decision | — | Legal, ethical, and human dimensions |
| Grievance investigation | — | Impartiality, witness interviews, evidence weighing |
| Medical or mental health situation | — | Clinical judgement, empathy, duty of care |
| Team conflict mediation | — | Reading the room, relationship history, trust |
| Termination | — | Dignity, due process, human presence |
The 40% that only humans can do — difficult conversations, sensitive investigations, genuine talent judgements, ethical decisions — is unchanged. The goal is to free HR professionals from the 60% so they can do the 40% better.
Sensitivity Labels in HR Work
Before installing any plugins, it helps to understand the sensitivity framework that governs all HR AI outputs in this chapter. Every output carries one of three labels:
| Label | What It Covers | Handling |
|---|---|---|
| ROUTINE | Policy summaries, JDs, onboarding plans, general queries | Standard output — can be shared widely |
| CONFIDENTIAL | Offer letters, salary details, performance reviews, talent assessments, reference letters | Contains personal or confidential information — handle accordingly |
| SENSITIVE PERSONAL DATA | Medical, disciplinary, grievance, termination | NEVER auto-generate — always escalate to a named HR contact |
You will see these labels appear in every AI output throughout this chapter. They are not decoration — they are a reminder that HR data touches people's careers, compensation, and dignity. The label tells you what care to take with the output.
Exercise: Diagnostic — Your Organisation's Knowledge Inventory
Type: Diagnostic Assessment Time: 25 minutes Plugin command: None (concepts only) Goal: Map your organisation's recurring HR questions to the explicit/tacit framework and identify the highest-risk knowledge gaps
Step 1 — List the Top 10 Recurring Questions
Think about the questions that HR — or managers, or senior colleagues — are asked most frequently. If you work in HR, draw from your experience or your ticketing system. If you are studying HR operations, use the EdTech company in Karachi as your scenario: 250 employees, growing fast, policies scattered across several systems, three HR team members.
Write your 10 questions as a numbered list. Each should be a genuine question that an employee would ask — "What is our remote working policy?" not "Remote working."
Step 2 — Classify Each Question
For each question, answer two things:
First: Is the answer explicit or tacit?
- Explicit — the answer exists in a written document somewhere, even if it is hard to find
- Tacit — the real answer is in someone's head and differs from (or is absent from) any written source
Second: Which function does answering it belong to?
- Information routing — someone needs to find and relay a written answer
- Process execution — someone needs to produce a document or run a process
- Knowledge capture — someone needs to extract and record something that only exists in someone's head
Build your inventory as a table:
| # | Question | Explicit or Tacit? | Function |
|---|---|---|---|
| 1 | What is our annual leave entitlement? | Explicit | Information routing |
| 2 | Who do I talk to if I have a problem with my manager? | Tacit | Knowledge capture |
| ... | ... | ... | ... |
Step 3 — Identify the Highest-Risk Tacit Knowledge
Look at every question you classified as tacit. For each one, ask: "If the person who holds this knowledge left tomorrow, what would break?"
Mark the three items with the highest potential impact with ★. These represent your organisation's most urgent knowledge capture priorities.
Deliverable: A completed 10-row inventory table with function and knowledge-type classifications, and three starred high-risk tacit knowledge items with a one-sentence explanation of why each matters.
Try With AI
Use these prompts in Cowork or your preferred AI assistant.
Reproduce: Apply the three-function framework to a fictional HR team.
I am going to describe an HR team's typical week. For each activity, classify it
as (A) information routing, (B) process execution, or (C) institutional knowledge
capture. Then identify which activities are candidates for AI automation, and
which require human judgement and should stay with the HR team.
Activities:
1. Answering "What is our sick pay policy?" for the fourth time this week
2. Writing an offer letter for a new marketing hire
3. Running a performance review meeting with a struggling employee
4. Documenting the team's informal promotion decision criteria before the VP leaves
5. Producing the monthly headcount report
6. Mediating a conflict between two team leads
7. Generating 20 onboarding schedule variants for the new intake
8. Deciding whether to terminate an employee for misconduct
9. Answering "How do I update my bank details for payroll?"
10. Capturing the finance team's knowledge about how the annual budget process
actually works (versus how the policy says it works)
What you are learning: The three-function framework is a classification tool — once you can correctly place any HR activity into one of the three, you can identify which activities AI should handle and which require a human.
Adapt: Apply the diagnostic to your own organisation.
I work in [describe your organisation: sector, size, team structure].
The HR team currently spends significant time on recurring requests and
administrative tasks.
Based on what you know about typical HR operations:
1. Generate a list of the 10 most likely recurring questions our HR team
receives
2. Classify each as explicit knowledge (written down somewhere) or
tacit knowledge (only in someone's head)
3. Classify each by function: information routing, process execution,
or knowledge capture
4. Identify the three highest-risk tacit knowledge items — what breaks
if the person who holds it leaves?
What you are learning: Applying the framework to your own context surfaces the specific overhead pattern in your organisation — which is different for a 50-person startup than for a 5,000-person enterprise.
Apply: Quantify the cost of information routing in your team.
I want to estimate the time cost of information routing in an HR team.
Assumptions:
- HR team of 3 people at a 250-person company
- They receive approximately 15 information routing queries per day
across the team
- Each query takes an average of 12 minutes to answer (finding the
policy, translating it, writing the response)
Calculate:
1. Hours per week spent on information routing across the team
2. Hours per year
3. As a percentage of a 40-hour week per person
Then: what would these team members be able to do with that time
if information routing were handled automatically? Give me three
specific examples of high-value HR work that is currently
under-invested because of information routing overhead.
What you are learning: The time-cost calculation makes the problem concrete — converting "a lot of time" into a specific number changes how organisations prioritise solutions.
Flashcards Study Aid
Continue to Lesson 2: Your HR Operations Stack →