The Alactic library
Practical guides for customer operations
From a better WhatsApp conversation to a clearer team handoff. Find a specific decision, a useful checklist or a workflow your business can put into practice.
1,030 articles
Decide Which Claims an AI Answer Must Support With Evidence
Review consequential AI claims without demanding citations for every sentence. Use a filled claim-review table to separate facts, calculations, recommendations, promises, and completed actions.
9 min readDecide Whether a Knowledge Update Is Ready to Publish
Use source checks, targeted retrieval comparisons, explicit issue decisions, and a tested recovery option to decide whether a knowledge update is ready to publish.
9 min readTest Whether Retrieval Distinguishes Similar Rules With Different Conditions
A plausible policy passage can still support the wrong decision. Use a filled contrastive test matrix to check whether changes in service, customer category, or operating context retrieve the right evidence.
10 min readDiagnose Whether a Retrieval Failure Comes From Missing or Hidden Knowledge
Before adding documents, trace a missed answer from its authoritative source to the evidence actually returned. Use a filled worksheet to distinguish missing knowledge from extraction, structure, vocabulary, filtering, and source-selection problems.
10 min readScore Retrieval Results for Supporting Evidence, Not Just Similar Wording
A retrieved passage can match the question yet miss its decisive exception. Use a requirement-level rubric to distinguish sufficient evidence from partial, misleading, and irrelevant results.
9 min readBuild a Retrieval Test Set From Real Business Information Needs
Test whether your knowledge collection supports real decisions, not just document-title searches. Use a filled example covering exceptions, dates, missing context, and unsupported questions.
9 min readSeparate Reference Facts From Embedded Instructions in Source Documents
A practical review method for separating useful reference material from copied prompts and embedded instructions, with a filled hypothetical checklist and a clear route for uncertain cases.
10 min readQuarantine Knowledge Conflicts That Cannot Yet Be Resolved
When approved sources disagree, isolate the disputed claims and dependent answers without disabling useful material. Use a conflict register to define owners, exclusions, and restoration evidence.
9 min readEstablish Source Precedence Before Conflicting Documents Reach Retrieval
When a handbook, temporary notice, and team note disagree, freshness is not enough. Use authority, scope, effective dates, and explicit supersession to decide which rule controls—and when to escalate.
10 min readSeparate Current Rules From Historical Rules in AI Knowledge
Publication dates do not tell an assistant when a rule applies. Use effective periods, explicit applicability conditions and date-sensitive retrieval tests to separate current operations from historical questions.
10 min readDecide What to Do With Knowledge That Has Missed Its Review Date
An overdue review date is a decision trigger, not proof that a source is wrong. Use risk, current evidence, and operational dependencies to decide what can stay available and what must stop informing AI answers.
10 min readIdentify Business Changes That Should Trigger Immediate Knowledge Review
A scheduled review is too late when a service, availability promise, or eligibility rule has already changed. Use a change-trigger map to identify affected sources, contain stale answers, and verify replacements.
10 min readSet Knowledge Review Intervals According to the Cost of Being Wrong
Stop reviewing every AI knowledge source on the same schedule. Use change likelihood, operational impact, usage, and dependable change signals to decide what needs attention first.
10 min readReplace an Approved Knowledge Source Without Leaving Two Active Rules
A new document does not retire its predecessor. Use a supersession record to control effective dates, remove obsolete passages from active retrieval, and verify that the replacement is the only current authority.
10 min readKeep Customer-Specific Exceptions Out of General AI Knowledge
A concession can be accurate for one customer and still be wrong as general guidance. Use a filled classification worksheet to separate standard rules, scoped exceptions, and historical material.
10 min readDefine What an Approved Source Is Actually Allowed to Answer
An approved product overview can explain features without authorising promises. Use a source authority card to define answerable topics, exclusions, audiences, and business contexts before an AI assistant relies on it.
10 min readAssign an Accountable Owner to Each Category of AI Knowledge
An uploader is not necessarily authorized to confirm accuracy. Use a domain ownership register to name accountable owners, reviewers, backups, and evidence requirements—and expose gaps before disputed content becomes an AI answer.
10 min readRemove Duplicate Knowledge Sources Without Erasing Useful Differences
Repeated documents can crowd retrieval, but similar wording does not mean identical scope. Use a decision register to separate copies from valid exceptions, choose canonical sources, and verify that cleanup preserves the right answers.
10 min readPrepare Tables So AI Retrieval Preserves Row and Column Meaning
Keep table values connected to their headers, units, conditions, and notes. Use a worked conversion and a filled reference artifact to verify that extracted facts retain their original meaning.
10 min readTurn Long Documents Into Retrievable Knowledge Without Losing Conditions
Structure retrieval passages around complete decisions, not arbitrary text lengths. Use a filled worksheet to keep rules, exceptions, restrictions, and source references connected.
10 min readDecide Which Business Sources Are Ready for an AI Knowledge Collection
A convenient file is not necessarily a reliable reference. Use a source admission checklist to separate approved material from documents that need repair or should stay outside the collection.
10 min readDecide When to Narrow an AI Assistant's Delegated Role
Repeated boundary confusion needs more than another reminder. Use a delegation-review worksheet to identify the disputed decision, narrow only the necessary authority, and set evidence-based conditions for reconsideration.
10 min readTest Whether People and AI Interpret Delegation Rules the Same Way
Use a blind calibration exercise to uncover disagreements about who may recommend, decide, and execute—then rewrite the policy and test fresh cases before expanding authority.
9 min readApprove Part of an AI Recommendation Without Implying Full Consent
Separate bundled recommendations into clear decisions, preserve conditions without inventing them, and record exactly what partial approval does—and does not—authorise.
9 min readTurn a guide into a working customer journey.
Explore the tools and services behind the operating process.