Skip to content
State-Owned Enterprises (BUMN) Sector

Prompt Engineering Training for the State-Owned Enterprises (BUMN) Sector

Corporate secretary, corporate planning, and HR staff at SOEs perform high-volume knowledge work: official memos, policy summaries, RUPS materials, disposition classification. AI accelerates this, but director-level governance and AKHLAK demand quality consistency and auditable trails. A good prompt library becomes the standard across subsidiaries and the compliance evidence at the same time.

  • 4.9/5 · 287 reviews
  • 300+ clients nationwide
  • 2,000+ trainers
  • 1 to hundreds of participants
  • Standard · Premium · Private 1-on-1
Request ProposalChat on WhatsApp
format
In-house / online / hybrid
duration
3 intensive days per cohort
participants
10-30 per batch
language
Indonesian / English

2,000+ trainers and experts

Our group's teaching network, matched to what your team needs.

See class documentation

You see your trainer first

The trainer's profile, track record and field come with the proposal, before you decide.

4.9 out of 5

287 participant reviews.

300+ clients nationwide and 180,000+ private training sessions completed. Among them: MPR RI, Akademi Militer Magelang, DPRD Provinsi Jawa Timur.

One person to hundreds

Private coaching for one participant, a class for one team, or several cohorts at once.

Three budget levels

  • Standard: core class, material fitted to the participants’ work
  • Premium: senior trainer, deeper material design
  • Private 1-on-1: one participant, one trainer, schedule set by the participant
Tell us your budget
State-Owned Enterprises (BUMN) Sector Focus

Why Writing AI Prompts That Deliver is different in State-Owned Enterprises (BUMN)

Corporate secretary, corporate planning, and HR staff at SOEs perform high-volume knowledge work: official memos, policy summaries, RUPS materials, disposition classification. AI accelerates this, but director-level governance and AKHLAK demand quality consistency and auditable trails. A good prompt library becomes the standard across subsidiaries and the compliance evidence at the same time.

Sector KPIs
  • AI policy compliance across subsidiaries
    Recorded in quarterly governance reports to holding
  • SPI/BPK findings related to AI usage
    No material findings in subsequent period
  • Cycle time of RUPS material / management report preparation
    Materially down, review quality maintained
Relevant regulations & standards
  • PER-2/MBU/03/2023 SOE Corporate Governance Guidelines
  • AKHLAK Core Values BUMN
  • ISO/IEC 42001:2023 — AI Management System
  • UU PDP No. 27/2022
  • NIST AI RMF GenAI Profile (NIST AI 600-1)
Target roles in State-Owned Enterprises (BUMN)
  • Corporate Secretary
  • Director of HR & GA
  • Head of Internal Audit (SPI)
  • Head of Corporate Planning Division
  • Head of Digital Transformation Division
  • Head of Legal & Compliance Division
Outcomes commonly requested in State-Owned Enterprises (BUMN)
  • Corporate secretary & planning staff use a standard prompt library across subsidiaries
  • Structured output eases summarisation / compilation of RUPS materials and quarterly reports
  • Company data classification (public / internal / restricted / confidential) becomes the first step
  • Prompt usage trails recorded for SPI examination and BPK working papers
  • Internal prompt committee (corp-sec-IT-HR) approves prompts before release to population
State-Owned Enterprises (BUMN)-specific questions
How does this training support AKHLAK adoption in AI usage?
Each module carries a behaviour component: 'Amanah' for guarding company data, 'Kompeten' for prompt literacy & review, 'Loyal' for corporate policy compliance, 'Adaptif' for responsible adoption of new tools. Training is professional habit aligned with SOE values, equipped with prompt quality rubrics.
Can we use SOE procurement (RKAP, e-procurement)?
Yes. Neksus supports RKAP-friendly: NPWP-bearing proposal, PPN invoice, domicile certificate, SOE contract format, and can pass vendor verification / e-procurement. Schedule aligned with budget cycle.
Can it run as a holding programme across subsidiaries?
Yes, roadshow + recurring program format: onsite sessions in core cities, virtual sessions for distant subsidiaries, plus corp-sec-IT-HR champion network for institutionalisation.

Quick Answer

Deep prompt engineering training builds knowledge-worker prompt discipline using research patterns (Chain-of-Thought, Tree-of-Thoughts, ReAct, Self-Consistency), JSON-Schema structured output, an eval harness, and NIST AI RMF GenAI Profile (NIST AI 600-1) + UU PDP governance, so AI becomes an auditable work tool, not 'feels right' output.

Free Consultation

Without an eval harness, 'feels good' prompts differ from 'proven good' prompts

Many teams trust one successful demo example. A golden eval set + quality rubric + A/B prompt is the only way to prove a prompt change truly improves quality and avoids regression. This program places the eval harness as a foundation from day one.

Prompt patterns = academic literature + official guides

We reference official OpenAI Prompt Engineering & Anthropic Prompt Engineering guides, plus academic papers: Chain-of-Thought (Wei et al. 2022), Self-Consistency (Wang 2022), Tree-of-Thoughts (Yao 2023), ReAct (Yao 2022). Each pattern is applied with real criteria for when it helps and when it is noise.

Reasoning models change the rules

Reasoning models (OpenAI o-series, Claude extended thinking, Gemini Deep Think) often return better results with LESS explicit reasoning in the prompt. Manual CoT is often counter-productive here. A dedicated module covers when to choose reasoning models considering cost & latency.

Clients & Documentation

Some of the institutions we have trained

Photos from sessions we have delivered, each one named after the institution it belongs to.

See class documentation (123 photos)

Writing AI Prompts That Deliver

Deep prompt engineering training equips knowledge workers (analysts, planners, writers, researchers, legal, compliance) to turn LLM queries into reliable, structured, evaluable instructions, using current research patterns (Chain-of-Thought Wei et al. 2022, Tree-of-Thoughts Yao 2023, ReAct Yao 2022, Self-Consistency Wang 2022) and production practice from official OpenAI/Anthropic guides, with JSON-Schema structured output, evaluation harness, and governance mapped to NIST AI RMF GenAI Profile (NIST AI 600-1) and UU PDP No. 27/2022.

1Grounded in academic literature and official guides: Chain-of-Thought, Tree-of-Thoughts, ReAct, Self-Consistency, Least-to-Most, role/system prompting, OpenAI Prompt Engineering & Anthropic Prompt Engineering
2Structured output coverage: JSON Schema, function calling, tool use, constrained decoding, so output can be processed by systems without fragile parsing
3Governance mapped to NIST AI RMF (Govern/Map/Measure/Manage) + GenAI Profile (NIST AI 600-1), UU PDP, and OWASP LLM Top 10 2025 (LLM01 prompt injection)
4Measurable outputs: per-role prompt library, golden eval set, quality dashboard, and AI usage SOPs
5Practice uses participants' work context (analysis, drafting, summarisation, classification), never generic content
6Designed with ADDIE and Bloom-aligned learning objectives; Kirkpatrick L1-L4 evaluation
7References we work from: the Chain-of-Thought, Tree-of-Thoughts, ReAct, and Self-Consistency patterns, schema-bound structured output, an eval harness for testing quality, and the NIST AI 600-1 GenAI Profile for governance.

Measurable Outcomes

Expected Outcomes

Indicators mapped to Kirkpatrick L1-L4; qualitative targets set at TNA.

The prompting patterns are mastered
Participants pass assessment on when to use CoT, ToT, ReAct, Self-Consistency, and structured output. Judged through schema-bound prompting exercises in class (Kirkpatrick L2).
The prompt collection gets used
Majority of participants use the team prompt library in weekly workflow within 30 days. Seen in shared prompts the team reaches for repeatedly (Kirkpatrick L3).
Output quality rises and can be measured
Quality score on the team's golden eval set rises vs free-prompting baseline. Read from test set scores before and after the class (Kirkpatrick L4).
The AI usage rules are followed
All participants pass the anti-prompt-injection (OWASP LLM01) checklist and UU PDP data SOP. Judged through a quiz on data usage limits closing the class (Kirkpatrick L2).
A prompt library and test set owned by the company
Validated prompt library + golden eval set + quality dashboard owned by the team. The evidence is a prompt library and test set the team maintains (Kirkpatrick L3 transfer).
Monetary value of the working hours returned (optional)
Per-task-category time-saving calculation, isolated from other factors, when finance requests it. Hours returned become rupiah via Phillips L5 where required.

Program Format

Program Format Options

Selected by participant population, task complexity, and institutionalisation ambition, finalised after TNA. The prompting patterns are mastered by one cohort inside those three days. Organizations wanting their prompt collection to become a shared asset carry on through a three month programme to build, test, and maintain a prompt library for each division.

1

Per-Role Prompt Workshop (1 day)

Workshop focused on one role (credit analyst, planner, customer service, etc). Build 8-15 main prompts for the participants' daily work with quick eval.

Best for: Specific teams wanting quick-win operational prompts
2

Deep Prompt Engineering Intensive (3-4 days)

Complete deep dive into advanced prompt patterns (CoT/ToT/ReAct/Self-Consistency), structured output, eval harness, security (prompt injection), and governance.

Best for: Power users, senior analysts, and AI team champions
3

Modular Prompt Bootcamp (6-8 sessions)

Weekly sessions with experimentation gaps: participants practice on the job, return with results, get reviewed, then move to more advanced patterns.

Best for: Knowledge workers with no operational break
4

Prompt Library Program (3 months)

Continuous program with office hours, prompt review club, and internal competition, until the team prompt library lives with owners and refresh SOP.

Best for: Organisations seeking institutionalisation beyond one-off events

Free Consultation

Build auditable prompt discipline

We map your team's roles, tasks, and AI stack, then build a proposal & budget estimate grounded in real need.

Request Proposal

Curriculum

Curriculum Framework

Designed with ADDIE; final modules curated based on TNA. Topics below are full coverage, activated partially based on participant population.

  • Tokenisation, context window, attention (without jargon)
  • Determinism, temperature, top-p, seed (effect on output)
  • Knowledge cutoff, hallucination, and when models 'invent confidently'
  • Model differences: GPT-5/4o, Claude Sonnet/Opus, Gemini Pro/Flash, Llama 3.x (effect on prompt style)
  • Reasoning models (o-series, Claude extended thinking): what changes in prompting

  • Four core blocks: role, context, instruction, output format
  • Few-shot prompting: when, how many examples, order, anti-bias examples
  • Role/system prompting per OpenAI and Anthropic guides
  • Providing explicit quality criteria (rubric) in the prompt
  • Handling long instructions: concise vs complete vs structured

  • Chain-of-Thought (Wei et al. 2022): when effective, when it becomes noise
  • Self-Consistency (Wang 2022): answer voting for mathematical/logical tasks
  • Tree-of-Thoughts (Yao 2023): multi-path solution search
  • Least-to-Most prompting for complex task decomposition
  • ReAct (Yao 2022): interleaving reasoning + tool use
  • Step-back prompting & meta-prompting for self-correction

  • Schema-bound JSON output (JSON Schema): a testable contract
  • Function calling & tool use (OpenAI, Anthropic, Gemini)
  • Constrained decoding & guided generation (Outlines, BAML)
  • Anti-parsing: extracting fields with regex vs structured output
  • Post-LLM validation: pydantic, zod, JSON Schema validator

  • Building a golden eval set with team subject matter experts
  • Quality rubrics: factuality, completeness, style, format adherence
  • LLM-as-judge: safe use (judge bias, calibration), when not to use
  • A/B prompt testing: how to test prompt changes before production
  • Prompt regression testing: CI/CD for the team prompt library
  • Prompt quality dashboard: Langfuse / PromptLayer / simple spreadsheet

  • OWASP LLM Top 10 2025, LLM01 prompt injection (direct & indirect)
  • Data forbidden from public models (UU PDP): per-role checklist
  • NIST AI RMF + GenAI Profile (NIST AI 600-1): Govern/Map/Measure/Manage
  • Prompt library governance: owner, versioning, deprecation, audit trail
  • 70-20-10 pattern: formal class + office hours + on-the-job adoption
  • Internal prompt committee: periodic review, taxonomy, retirement

Comparison

Choosing the Prompt Engineering Program Format

Concise decision matrix, finalised after training needs analysis.

AspectPer-Role Prompt Workshop (1 day)Deep Prompt Intensive (3-4 days)Modular Prompt Bootcamp (6-8 sessions)Prompt Library Program (3 mo)
Primary goalQuick win 1 roleMastery of current patternsLearn alongside operationsInstitutionalise prompt library
Ideal participantsOne focused rolePower users & championsOperationally-loaded KWCross-role across the org
Eval depthConcise rubricGolden set + A/B promptStaged eval per sessionFull eval + prompt CI/CD
Training evaluation levelKirkpatrick L1-L2Kirkpatrick L1-L3Kirkpatrick L1-L3Kirkpatrick L1-L4 (+Phillips L5)
Best forQuick validation per roleBuild champion capabilityOperationally-loaded teamsOrg-wide adoption & impact target

For Whom

Who This Program Is For

Knowledge workers who want AI as a serious work tool beyond a demo toy. No coding prerequisite.

Business analysts, research, and planning

Build prompts that produce structured analyses, reliable summaries, and consistent classification.

Common challenges

  • Prompt results vary between people & between days
  • Hallucination on numbers & names goes undetected
  • No objective way to judge which prompt is better

Legal, compliance, and policy teams

Prepare prompts for drafting, regulation summarisation, risk classification, with accountable trails.

Common challenges

  • Concern about sensitive data leakage via prompts
  • No SOP restricting what can be typed into public AI
  • AI output hard to audit because it is not schema-bound

Marketing, communications, and service teams

Produce content variants, summaries, and customer responses with consistent brand voice.

Common challenges

  • Brand voice drifts between writers
  • Output is unstructured, hard to use in CMS / CRM
  • Product claims risk consumer regulation / BPOM violations without controls

L&D, HR, and AI champion teams

Build team prompt libraries & lead institutionalisation of measurable AI use.

Common challenges

  • AI training stops at 'what is ChatGPT', does not enter work
  • No way to measure adoption progress & use quality
  • Prompt library becomes a dead spreadsheet after the event ends

Industry Context

Industry Applications

Each industry has typical knowledge-work tasks. Prompt patterns are designed to follow them.

Banking & Financial Services

Credit-analyst & compliance prompt library: credit document summarisation, initial risk classification, committee memo drafts with structured output ingestible by the LOS system, all with anti-prompt-injection checklist and customer-data guardrails.

See in Banking & Financial Services context →
Technology & Startups

Engineer & PM prompt library: PRD drafting, post-mortem summarisation, unit-test generation, ticket classification, with JSON Schema structured output directly usable by internal systems and defense against prompt injection in new tools.

See in Technology & Startups context →
Retail & FMCG

Marketing & CX prompt library: multi-channel copy variant production, customer-review classification, complaint-response drafting, with structured output, brand-voice guard, and UU PK / BPOM product-claim checklist.

See in Retail & FMCG context →
State-Owned Enterprises (BUMN)

Corporate-secretary & planning prompt library: official memo drafting, policy summarisation, RUPS materials, disposition classification, with structured output, audit trails, and AKHLAK + UU PDP compliance.

Government & Public Sector

ASN prompt library: official letter drafts, policy summaries, disposition classification, public-service assistant, with structured output, information classification, and inspectorate audit trails per SPBE.

See in Government & Public Sector context →
Healthcare & Pharmaceuticals

Non-clinical prompt library for hospital & pharma teams: policy drafting, internal-literature summarisation, service-complaint classification, administration assistant, with strict boundaries: no direct clinical decisions and no medical records pasted to public models.

See in Healthcare & Pharmaceuticals context →

Delivery Method

Delivery

Hands-on practice using participants' work context. Every session produces real prompts usable the next day.

In-house onsite

Facilitator comes to the office; participants practice with team documents / data (in a safe environment). Best for 2-4 day intensives.

Live online

Interactive class via Zoom/Teams with per-role breakouts, screen-share prompt reviews, and session recordings. Best for modular bootcamps.

Hybrid

Onsite for intensive labs & prompt committee; online for office hours and ongoing prompt review club.

Schedule arranged around team operational calendar
Materials, initial prompt library, and eval template prepared by Neksus team
Safe practice environment if enterprise AI licenses are not yet available
Prompt-skills attendance certificates when requested
Post-training evaluation report for L&D team & leadership

Engagement Flow

Engagement Path

From need to live, measurable prompt library: qualitative duration, scaled to organisation size.

1

Training Needs Analysis (TNA)

Map roles, knowledge-work tasks, AI stack, data policy, and business targets. Output: needs profile + measurement baseline + initial prompt library scoping.

Initial phase
2

Program Design (ADDIE)

Draft learning objectives, per-role syllabus, initial golden eval set, and prompt library template participants will develop.

Before delivery
3

Delivery: Intensive Lab

Hands-on lab: prompt patterns, structured output, eval harness, security. Participants write real prompts for their own work and test against the golden set.

Program core
4

Prompt Committee & Governance

Form internal prompt committee, charter, per-prompt owners, versioning policy, and integration with team document / git system.

Post-intensive
5

Office Hours & Prompt Review Club

Coaching prompt usage in real work, weekly/biweekly prompt review club, and internal prompt competition for momentum.

Ongoing
6

Kirkpatrick Evaluation & Roadmap

L1-L4 measurement (reaction, learning, behaviour/adoption, results). Phillips ROI L5 on finance request. Prompt library maturity roadmap.

Post-program

Case Studies

Typical Outcome Patterns

Illustration of impact patterns from similar program structures; no named clients or promised numbers.

Analyst team at a financial institution with AI already used sporadically

Intervention

3-day intensive + initial 30-prompt library + 60-example golden eval set

Result

Output consistency improved; team has a live prompt library with owners; usage trails can be shown to SKAI

Marketing and CX team at a large retail organisation

Intervention

3-month continuous program + prompt review club + UU PK claim checklist

Result

Content variant throughput up, consumer-regulation risk incidents not rising; prompt library becomes the standard across regions

Secretariat & planning unit at SOE/ministry

Intervention

Guided workshop + internal prompt committee + UU PDP governance

Result

Document drafting faster with consistent quality; usage trails prepared for SPI / inspectorate examination

Procurement Info

Information for Procurement & Vendor Management

Materials for procurement, finance, legal, and information security teams.

Proposal

Structured proposal: measurable learning objectives, syllabus, governance mapping (NIST AI RMF/UU PDP/OWASP LLM), facilitator profile, schedule, TNA-based cost breakdown.

Pricing model

TNA-based: flat per program, per session, per participant, tiered, or custom. Estimate provided after TNA.

Payment & tax

Flexible terms (down payment + balance / per-batch terms); PPN tax invoice and PO document support available.

SOE / government process

Experienced with SOE/government procurement: vendor documents, e-procurement, HPS/bidding, compliance clauses.

Measurement

Kirkpatrick Level 1-3 evaluation report (attendance, assessment, practice results); Phillips ROI Level 5 on finance request.

Confidentiality & data security

NDA, confidentiality clauses, and practice that does not force confidential data into public models (aligned with UU PDP & OWASP LLM).

Material ownership

Prompt library & golden eval set built for the company become the company's property; training-material usage rights agreed in contract.

FAQ

Frequently Asked Questions

Regular ChatGPT classes stop at 'how to use'. This training builds prompt discipline: advanced reasoning patterns (CoT/ToT/ReAct/Self-Consistency), schema-bound structured output, eval harness, security (prompt injection), and governance (NIST AI RMF, UU PDP). Output: a live team prompt library with owners & versioning, going beyond a screenshot collection.

No. This training is for knowledge workers. Structured output, eval harness, and governance are explained in business-user language. For deep integration (e.g. function calling into internal systems), we have a separate engineering path that can be added as an advanced module.

Start with fundamentals: role-context-instruction-format-output + few-shot. Next, choose patterns by task: CoT for reasoning, structured output for system integration, ReAct for tasks needing tools, Self-Consistency for high-risk tasks. We provide a pattern-selection decision tree in the curriculum.

With a golden eval set + quality rubric (factuality, completeness, style, format adherence). Every prompt change is A/B tested against baseline on the golden set before production use. At scale, the quality dashboard monitors output drift over time.

Yes, explicitly. The pattern shifts: in reasoning models, explicit CoT prompting is often counter-productive (the model already 'thinks'); focus moves to providing clear evaluation criteria and output structure. Includes significant cost & latency considerations.

The OWASP LLM01 module teaches defense in depth: (1) user input sanitisation; (2) resilient system prompts (no raw user input paste); (3) structural output guards; (4) tool-call restriction; (5) red-team prompts for regression. For indirect injection via documents / web, there is an additional checklist.

An internal prompt committee (L&D + IT + domain owner) does periodic review: new prompts go through PR review, broken / out-of-date prompts are deprecated, and there is a per-prompt owner. The governance module contains complete governance + committee charter template.

Yes. TNA maps the team stack. Prompt patterns have nuance across models. We train the practical differences. For example: Anthropic prefers instructions in a long system prompt; OpenAI is flexible; Gemini has its own structured-output behaviour.

Yes. The continuous program provides office hours, prompt review club, and internal prompt competitions. The 70-20-10 pattern: most learning happens on the job, supported by periodic reviews so momentum is not lost after class.

Yes. The report covers attendance, assessment scores, practice results, per-participant golden eval scores, and recommendations for L&D and leadership, mapped to Kirkpatrick levels. Knowledge workers who finish their team prompt library can request an attendance certificate listing their competencies.

The first determinant is whether material is built per division, since prompting patterns for a legal team, a finance team, and a marketing team grow out of different kinds of work. Participant numbers, cohort count, participant seniority, and meeting count follow. When the class also builds a prompt library, drafting and testing sessions are added. The venue, access to the AI tools in use, and the host city finish the calculation. The figure lands once your training needs analysis is behind us.

One cohort holds 10 to 30 participants, which means twenty-odd people sit comfortably, and three full days covers the prompting patterns, schema-bound output, and testing the results. What makes the result last is the shared prompt collection: leave time after the class to tidy it and name whoever maintains it. When participants come from many divisions, one class still works, only the practice cases are better split by division.

They can be changed. A prompting class drifts into showing off tricks when whoever carries it never ties them back to the work in the room. Put a word to the programme coordinator, so a reserve validated prompt coach can finish the remaining sessions. The prompt library built in class, the test set, and per-participant progress notes rest inside our operations system, which lets the class resume from the last exercise.

Next Step

Build auditable prompt discipline

We map your team's roles, tasks, and AI stack, then build a proposal & budget estimate grounded in real need.

  • Proposal, syllabus, and governance mapping
  • Procurement-ready documents (company profile, NPWP, NDA, PPN invoice)
  • Kirkpatrick impact measurement (Phillips ROI on request)
Request a ProposalFree Consultation

Explore More

Related Topics

Pelatihan Ai Generatif KorporatPelatihan Rag Knowledge Base Llm AplikasiLiterasi Data Analitik BisnisKomunikasi Presentasi Eksekutif

Writing AI Prompts That Deliver training for your State-Owned Enterprises (BUMN) team

We map your team's roles, tasks, and AI stack, then build a proposal & budget estimate grounded in real need.

  • Proposal, syllabus, and governance mapping
  • Procurement-ready documents (company profile, NPWP, NDA, PPN invoice)
  • Kirkpatrick impact measurement (Phillips ROI on request)

Six fields to get started. Our corporate team reads and answers it.

Additional details (optional)

Add these if you already know them. Every field in this section can be left empty, and our team will cover them during the needs-analysis call.