Caribbean AI Association
Guh bak a di akaiv
Wait piepa dem

Practical AI fi human resources inna financial services

Harbour Think Tank, Caribbean AI Association, under the leadership of Adrian Dunkley   |   2026-11-10

Shaata edishan fi riida. Harbour Think Tank, Caribbean AI Association, anda Prezident Adrian Dunkley. Oktoba 2026. Revision 10 November 2026.

Dis short edishan go wid di full orijinal ina Inglish.

Shaata edishan fi riida. Harbour Think Tank, Caribbean AI Association, anda Prezident Adrian Dunkley. Oktoba 2026. Revision 10 November 2026.

Weh di test fi cover

Agree weh each measure mean first. Use approved documents fi test learning, preparation an follow-up. Keep di record read-only. Authorised people keep control a hiring, pay, leave an appraisal.

Ten use case fi test

01. Work goal. Use agreed priorities fi draft a goal fi each role. Di manager an team must agree how fi measure it, wid what resource an by when. Separate learning goal from performance target. Mark any missing baseline. Do not send drafts straight to staff.

02. Measure dictionary. Compare numerator, denominator an time period across reports. Di reporting owner resolve conflict an reproduce a result before sign-off. A second analyst reproduce di measure using only di written definition.

03. Weekly evidence. Separate submitted, accepted an pending work. Link each completion to proof. Operations correct any item weh show complete without evidence. Manager’s recorded acceptance is di test fi done.

04. Practice by role. Build a task from a procedure, like checking evidence fi open an account. A specialist approve di marking guide an test di skill without help. Use comparable tasks before an after practice. Satisfaction alone cannot measure skill.

05. New-starter questions. Answer from di current policy an show di passage. If it missing or unclear, pass it to HR. Run di same fifteen questions every time di prompt or library change.

06. Coaching talk. Start wid one goal, one observed example an di support already offered. Di manager check fact an fairness before agreeing action. Use GROW fi order di questions, then describe situation, behaviour an impact in di example.

07. Role description. Compare di listed duty wid di actual workflow an handovers. Managers settle overlap an work weh nobody own. Give each process step just one accountable role on di RACI chart.

08. Team capacity. Compare incoming work, available hours an possible queues. Di manager challenge assumptions against actual load an overtime. Vary assumptions an compare mean absolute percentage error where observed values allow that calculation.

09. Staff feedback. Remove identifiers, group themes an keep contrary comments. Two reviewers check di groups an name who fi take each action. Measure di two reviewers’ agreement wid Cohen’s kappa.

10. Development programme. Draft published entry criteria an learning activities. HR check access barriers. Di assistant must not rank staff by potential. Use di same criteria an evidence fi every applicant. A human panel log di reasons an check equal access before launch.

Di 90-day test

Day 1–15: pick one task. Name di owner an who can stop di test. Write down di approved source, access, time an error inna di work now. Day 16–45: test hard case an record weh disagree. Check every output an keep di test separate from live work.

Day 46–75: compare like task wid like task. Count di time fi prepare, check an fix. Day 76–90: look pon di serious error dem an decide fi widen, change or stop. Di think tank propose di workflow an priority dem. Dem never measure result from a company rollout. Di full English paper give di process, duty table, checks an source notes.

Suos dem

  1. [1] Caribbean AI Association. Think Tanks. Harbour remit: economy, work and enterprise. Accessed 26 October 2026.
  2. [2] Statistical Institute of Jamaica. Labour Force Data Explorer, Main Labour Force Indicators, both sexes, July 2025 and July 2026. Accessed 26 October 2026.
  3. [3] Gmyrek, P. and colleagues. 2025. Generative AI and Jobs: A Refined Global Index of Occupational Exposure. International Labour Organization Working Paper 140, 20 May.
  4. [4] Brynjolfsson, E., Li, D. and Raymond, L. 2025. Generative AI at Work. Quarterly Journal of Economics 140(2), 889-942.
  5. [5] Gifford, J. 2016. Could do better? Assessing what works in performance management. Research report, December. Chartered Institute of Personnel and Development. Overview:
  6. [6] Office of the Information Commissioner, Jamaica. Data Protection Standards. Accessed 26 October 2026.
  7. [7] Jamaica. Data Protection Act 2020, section 12, rights in relation to automated decision-taking. Official copy hosted by the Office of the Information Commissioner.
  8. [8] Agent Skills. Agent Skills Overview. Accessed 10 November 2026.