Are our team members getting meaningfully different results from AI depending on how they ask questions?

If different people get wildly different results from the same AI tool, the gap is in prompt technique, not the tool. Sharing effective patterns across the team can double productivity without changing any technology.

Have we documented the prompts that work best for our most common tasks?

Undocumented prompts mean every person reinvents the wheel. A shared prompt library for common tasks turns individual discoveries into organizational capability.

What would it be worth if our entire team got results as good as our best AI user?

The gap between your best and average AI user is your prompt engineering opportunity. Closing it through training and shared templates delivers immediate ROI with zero additional tool cost.

Using AI in Your Work

Prompt engineering

By Mark Ziler · Last updated 2026-04-05

Prompt engineering is the craft of telling an AI exactly how to behave for a specific task. It is not coding — it is precise instruction-writing. When building an agent that answers questions about field services data, the prompt tells it: use the jobs table for individual service calls, use the financial table for branch-level P&L, never show raw database IDs, always format currency with two decimals, and when someone asks about "margin" they mean gross_margin_pct which is stored as a percentage not a decimal. Good prompt engineering is the difference between an AI that gives vague generic answers and one that sounds like it has worked at your company for years. It is invisible to the end user but it is where most of the intelligence lives.

Go deeper

Your field services company deploys an AI agent to answer operational questions. A dispatcher asks: 'What is the average job time?' The agent scans every table it can access and returns a number that averages all job types — emergency calls, installations, maintenance visits, inspections — into one meaningless number. The dispatcher wanted average time for residential maintenance calls only. The prompt engineering fix is not making the AI smarter. It is telling the AI upfront: 'When someone says job time without specifying a type, ask which job type they mean. Here are the categories and how they map to the database.'

The trap most companies fall into is treating prompt engineering as a one-time setup task. Your business rules change. You add a new service category, redefine how overtime is calculated, or restructure your regions. If the prompts do not get updated to reflect these changes, the AI gives stale answers using old business logic — and does it confidently. Prompt maintenance is an ongoing operational task, not a project with a finish date.

Questions to ask

Who in our organization is responsible for maintaining the AI's business rules and prompt definitions?
When was the last time someone tested the AI agent against a set of known-answer questions?
If we change a business rule next month (like how we calculate utilization), what is the process for updating the AI?