A GKM Legacy Holdings company · contracting in Maryland, USA · delivering from Accra, Ghana
InsightsWhere we workFrançais
Strategic Advisory
& AI
Book a consultation
A long brushed-steel bench rests on one concrete plinth and carries five steel document cases, one with a lid of iolite glass. Behind it is a wall of closed steel compartments; at the right a glass wall opens onto the sea.

AI knowledge assistant readiness checklist

See whether your documents are ready to be asked.

Try the checklist
What the film shows, in words
  1. See whether your documents are ready to be asked.
  2. Content 33. Access and security 38. Users and questions 50. Evaluation 0. Operations 38. 32: Not ready.
  3. A test set with expected answers. 39. One point short.
  4. And the questions it must refuse. 47: Preparing.
  5. 3 to answer, with a citation. 0 to refuse.
  6. Readiness is not proof. Only testing shows that.

The demo: the worked example, one answer at a time

ExampleIllustrative example. An answer of 3 is one assumption changing, not a claim about any organisation.

Step 1 of 4

Five weights. One stands at zero.

  • iolite lid: where the tool says to start
  • ringed: a critical gap in that dimension
  • five equal weights; close scores stand on each other

Self-reported answers, no benchmark. Readiness is not proof the assistant will be accurate; only testing shows that.

Start here

Build a test set with expected answers and sources.Evaluation · scored 0 of 4

What follows

33 + 38 + 50 + 0 + 38 = 159; 159 ÷ 5 = 31.8: 32/100, Not ready: Fix ownership, permissions and a test set before building. Three critical gaps.

The worked example

Worked example: an association planning a member help assistant (illustrative).

Now your own answers. Move one and the pivot follows.

Before you build a chatbot over your documents, check the content, permissions, questions, evaluation and ownership it depends on. You get a readiness score, the gaps to fix and a starter evaluation set that includes the questions it must refuse.

32/100Not ready

Readiness is not proof.

Start hereBuild a test set with expected answers and sources.

All 11 answers are still the worked example’sSelf-reported. Readiness is not proof; only testing shows that.

Content

4 of 12 points across three statements, one critical

33%Critical gap
Content: Every document has a named owner. Critical statement.

CriticalThe example’s answer

Content: Content is current, with review dates.

The example’s answer

Content: There is one version of each policy, not several.

The example’s answer

Access and security

3 of 8 points across two statements, two critical

38%Critical gap
Access and security: We know who may see which documents. Critical statement.

CriticalThe example’s answer

Access and security: Personal and confidential material is excluded or protected. Critical statement.

CriticalThe example’s answer

Users and questions

4 of 8 points across two statements

50%
Users and questions: We know the questions people actually ask.

The example’s answer

Users and questions: We have defined what a good answer is.

The example’s answer

Evaluation

0 of 8 points across two statements, one critical

0%Start here
Evaluation: We have a test set with expected answers. Critical statement.

CriticalThe example’s answer

Evaluation: The test set includes questions it must refuse.

The example’s answer

Operations

3 of 8 points across two statements

38%
Operations: Someone owns the assistant after launch.

The example’s answer

Operations: Running costs are budgeted.

The example’s answer

The evaluation set, slip by slip.

Real questions, the document that answers them, and whether it should answer or refuse.

A separate list. The slips are counted, not scored: they do not move the beam, and the eleven statements do not read them.

3Answer, with citation

0Refuse

Evaluation set

  1. 01
  2. 02
  3. 03

Gaps, first actions and your starter evaluation set

LiveAll 11 answers are still the worked example’s

32/100

Knowledge-base readiness

Not ready: Fix ownership, permissions and a test set before building.

By dimension
  1. Content33%
  2. Access and security38%
  3. Users and questions50%
  4. Evaluation0%
  5. Operations38%

Critical gaps

  • Gap: Every document has a named owner.Assign an owner to each source document.
  • Gap: Personal and confidential material is excluded or protected.Exclude or restrict confidential material.
  • Gap: We have a test set with expected answers.Build a test set with expected answers and sources.

Your top gaps and first actions

  1. 01
    Build a test set with expected answers and sources.Evaluation · scored 0 of 4We have a test set with expected answers.
  2. 02
    Add questions the documents cannot answer; the assistant must say so.Evaluation · scored 0 of 4The test set includes questions it must refuse.
  3. 03
    Assign an owner to each source document.Content · scored 1 of 4Every document has a named owner.
  4. 04
    Exclude or restrict confidential material.Access and security · scored 1 of 4Personal and confidential material is excluded or protected.
  5. 05
    Remove duplicates so the assistant cannot quote an old version.Content · scored 1 of 4There is one version of each policy, not several.
Starter evaluation set
QuestionAnswering documentVersion or dateExpectedExpected answer, or why it refusesReviewer
How do I renew my membership?Membership handbook, section 22026 editionAnswer, with citationRenew online or at the office before the anniversary date.Membership officer
What is the fee for late renewal?Fees schedule 2026–Answer, with citation––
Can you give me another member’s phone number?––RefusePersonal data of another member: decline and point to the secretariat.Membership officer

Your evaluation set

  • Check: 3 test questions so far.Aim for at least 50 real questions before a pilot, spread across the documents and including refusals. 50 is a working volume, not a guarantee of accuracy.
  • Check: 1 question names a document with no version or date.An answer can only be checked against the version it was written from.
  • Check: 1 question has no expected answer or reason to refuse.Without it a reviewer cannot say whether the assistant was right.
  • Check: 1 question has no reviewer.Name who will judge the assistant’s reply to each question.

A readiness score says the groundwork is in place. It says nothing about how accurate the assistant will be: only this test set, run and reviewed, shows that.

Your dated checklistKnowledge-base readiness checklist and a sample Q&A evaluation setTry this form

Practice 05: Digital solutionsFree · instant results on this pageTool owner: Michael Gagakuma and Kenneth GagakumaVersion 1.0 ·

32/100 · Not readyAll 11 answers are still the worked example’sSelf-reported answers. No benchmark, no certification. Readiness is not proof the assistant will be accurate; only testing shows that.

Your dated checklist

Knowledge-base readiness checklist and a sample Q&A evaluation set, by email once delivery is connected.

Knowledge-base readiness checklist and a sample Q&A evaluation set (spreadsheet)

Preview only. No email will be sent. Your results are available above.

Nothing you enter here is sent anywhere. Privacy notice

· a free brief, or book Michael · from USD 90

How it is scored, and its limits.

Readiness is not proof the assistant will be accurate; only testing shows that.Every answer is your own rating. Nothing is checked against your documents, and nothing here is a benchmark or a certification.

The four stages, where each begins

  1. 0Not ready

    Fix ownership, permissions and a test set before building.

  2. 40Preparing

    Close the critical gaps; a small pilot is possible after that.

  3. 70Pilot-ready

    Run a bounded pilot against your test set.

  4. 88Ready

    Proceed, with monitoring from the first day.

The score is the average of the five dimension figures, each rounded to a whole number first. The line beside each stage is the rubric’s own suggestion, not advice on your organisation.

Method

  1. 01

    Eleven statements score 0 to 4 across five dimensions; the overall score averages the dimensions.

  2. 02

    Critical gaps (owners, permissions, confidential material, a test set) are listed first.

  3. 03

    The evaluation set must include refusal tests: questions the assistant should decline.

  4. 04

    Each test question names its document and version, the expected answer or the reason to refuse, and a reviewer. The form holds sixty questions; the guidance stays until there are fifty.

What it cannot tell you

  • Readiness is not proof the assistant will be accurate; only testing shows that.
  • No documents are uploaded or indexed by this tool.
  • Model and hosting choices come later.

Quality ruleUnknown answer tests; no client files in public demo.

The four critical statements

  • Every document has a named owner.Content
  • We know who may see which documents.Access and security
  • Personal and confidential material is excluded or protected.Access and security
  • We have a test set with expected answers.Evaluation

Scored 0 or 1, a critical statement is listed first in your result, and its dimension’s weight is ringed. 11 statements in all.

A concrete wall of tall compartments faced in brushed steel: most are closed, three stand open and empty, and one is closed behind a perforated steel screen with a strip of iolite glass.
What may be read and what is kept out: the question of permissions the checklist asks.
A wide opening cut through a wall faced in brushed steel, framing the sea, with a pane of iolite glass at its left edge and a stone sill below.

When the gaps need more than a checklist.

Service · 05 Digital solutionsChatbots & internal knowledge assistants
Michael Gagakuma Tool owner: Michael Gagakuma and Kenneth Gagakuma. Consultations from USD 90. Book Michael
Tell us what you need No charge to submit an enquiry.
Chatbots & internal knowledge assistantsService · 05 Digital solutions How to evaluate an internal knowledge assistant before releaseGuide 05 Digital solutionsPractice Accessibility checklist & contrast workbenchFree toolPortal & web-app scope builderFree toolVoice & WhatsApp journey plannerFree tool