Risk Management | Hokstad Consulting

Risk Management

Blog posts in the Risk Management category

AI Pilot vs Production: How Decision Gates Differ

Pilot success isn’t permission to launch; satisfy four gates—test coverage, user exposure, support ownership and failure limits—before production.

Read more

RAG Abstention Thresholds: Balancing Errors and Refusals

Calibrate RAG abstention thresholds with retrieval, evidence coverage and uncertainty, using error costs to minimise business loss.

Read more

Grounded Generation Prompts: Handling Evidence Gaps

Four-step method to craft evidence-only prompts: cite sources, expose gaps, ask one clarifying question or abstain when unsupported.

Read more

Speech and Pointing vs Text for Agent Object Selection

Compare speech+pointing and text for object selection; use identify–confirm–act tests to measure accuracy, time, accessibility and recovery.

Read more

Stochastic Consensus & Multi-Agent Debate Compared

Start with stochastic consensus; escalate to multi-agent debate only when verified critique justifies the extra token cost, latency and risk.

Read more

5 Licence Checks Before Selling Fine-Tuned AI Models

Five licence checks to confirm commercial rights, distribution, attribution, limits and customer terms before selling fine-tuned AI models.

Read more

LLM SDK Upgrades: Breaking Change Checklist

Six-step checklist for LLM SDK upgrades: map breaking changes to apps and owners, validate requests/responses; rehearse rollout and rollback.

Read more

When Do AI Agents Need Human Approval Checkpoints?

Pause AI agents for human approval on financial transfers, sensitive-data exports and irreversible production changes; automate low-risk tasks.

Read more

LLM Retries vs Circuit Breakers: Handling API Errors

When to retry LLM API calls, when to open a circuit and how to switch fallbacks safely within shared deadlines and budgets.

Read more

Multi-Agent Consensus: Debate vs Voting

Compare parallel voting and limited debate for multi-agent decisions—use voting for fixed outputs and bounded debate for disputes.

Read more

AI Change Communication Plans: What to Tell and When

Staged messaging for AI pilots to rollout: owners, approvals, audience timing, evidence, and pause rules.

Read more

AI Agent Stagnation: How Progress Checks Catch Stalls

Measure verified results, enforce windowed progress checks and block further calls when AI agents show no measurable improvement.

Read more