In a previous blog, you could read how we tested several use cases, and hopefully it sparked your interest to get started yourself. Before we walk you through the key steps to consider when introducing AI into your organization, we first want to focus on crafting instructions, commonly referred to as prompts. An AI model does exactly what you ask it to do, but not necessarily what you actually mean. The difference between a disappointing answer and a useful result therefore almost always comes down to the prompt.
In this article, we share ten practical tips for creating better prompts. These tips are illustrated using both a weak and a strong prompt, specifically in a NetSuite context.
Prompt frameworks
Before diving into the concrete tips, it’s best to first choose a prompt framework. This is a structured way of building prompts, ensuring the output is more consistent and better aligned with your needs.
Three commonly used prompt frameworks:
- RISEN / Role – Instructions – Steps – End Goal – Narrowing
Ideal for complex, multi-step tasks where sequence and scope are critical. - CO-STAR / Context – Objective – Style – Tone – Audience – Response
Strong for content creation, as it also takes tone and audience into account. - RACE / Role – Action – Context – Execute
The most accessible structure, suitable for almost any task.
The ten tips below are the building blocks within these prompt frameworks. Sometimes you’ll use all of them together, sometimes only a selection is sufficient. The prompt technique is not a checklist, but rather a tool to ensure you provide enough input to the AI model.
10 tips for better prompting
Tip 1: be clear and specific
Clearly define what you want, leaving no room for interpretation. The more vague your question, the more generic the answer. Specify the exact factors you want analyzed.
| Weak prompt | Explain the revenue changes. |
| Strong prompt | Explain why revenue changed last month, focusing on deferred revenue, recurring costs, and one-off sales. |
Tip 2: provide context
Add background information so the model understands the situation and objective. Context determines not only what the model answers, but also how. Without context you get a generic answer; with context you get something usable immediately.
| Weak prompt | Analyze this balance sheet. |
| Strong prompt | We are a Belgian manufacturing company with three subsidiaries. Our CFO noticed that intercompany balances do not match at the end of the quarter. Analyze this balance sheet and identify the sources of the discrepancies. |
Tip 3: define the audience
Specify who the output is for so that tone, terminology, and level of detail are appropriate. A summary for a CFO looks very different from one for a sales team.
| Weak prompt | Explain this sales report. |
| Strong prompt | Summarize this sales report for our regional sales team. Focus on the top 5 best-performing products and regions. Use simple language, avoid accounting jargon. |
Tip 4: use role prompting
Assign a role to steer the model’s perspective and expertise. A “financial controller” looks at different aspects than an “accountant” or an “auditor”.
| Weak prompt | Correct this journal entry. |
| Strong prompt | Take on the role of a senior financial controller: review these journal entries for accuracy, correct subsidiary allocation, and proper descriptions. Then provide the corrected version. |
Tip 5: add constraints
Set limits on length, tone, or exclusions. Without constraints you get an endless list; with constraints you get exactly what you need.
| Weak prompt | List our suppliers. |
| Strong prompt | List active suppliers with purchases above $10,000 this year. Exclude intercompany suppliers. Maximum 20 results, sorted by total spend in descending order. |
Tip 6: define the format
Specify in advance how you want the output, both in structure (table, list, steps) and file type (Excel, PPT, interactive dashboard). By defining columns, sorting, and format upfront, the result immediately matches your needs and requires no reformatting afterward.
| Weak prompt | Give me an inventory list. |
| Strong prompt | Export inventory levels per warehouse in Excel format with the columns: Item Code, Item Name, Warehouse, Quantity On Hand, Quantity Available, Reorder Point, Preferred Vendor. Sort by Warehouse ascending. |
Tip 7: let AI ask questions
Explicitly instruct the model to ask clarifying questions before starting, especially for complex tasks. This helps prevent missing information and reduces hallucinations or assumptions.
| Weak prompt | Create a sales order for customer ACME. |
| Strong prompt | Create a sales order based on this PDF file. Ask me clarifying questions to ensure all data is entered correctly and no important information is missing. |
Tip 8: use examples (few-shot prompting)
Provide example outputs so the model can replicate style and format. Two example rows are often enough to ensure consistency.
| Weak prompt | Create a closing checklist. |
| Strong prompt | Create a complete 10-step NetSuite month-end closing checklist in this style: 1. Reconcile bank accounts 2. Record depreciation. Example attached for reference. |
Tip 9: ask for reasoning
Let the model think step by step for complex or analytical tasks. This significantly reduces errors compared to jumping directly to a conclusion.
| Weak prompt | Why is cash lower? |
| Strong prompt | Step by step, analyze why the cash position decreased last month in NetSuite, focusing on the timing of accounts payable, accrued expenses, and customer payments. |
Tip 10: iterate step by step
You don’t need to capture everything in a single prompt. Start broad and refine with additional instructions unless you already know exactly what you want in which case you can be more specific from the start.
| Weak prompt | Improve this customer list. |
| Strong prompt |
Refine this customer list: remove inactive customers (no orders in 12 months), add a column for credit limit utilization in %, and flag customers with overdue invoices older than 60 days. |
Final note
The ten tips come down to one principle: the more you tell the model about what you expect, the less you need to correct afterwards. The sharper your brief, the more useful the result.
You don’t need to become an expert in prompting to make an impact. Our advice is to start with the first three tips: be specific, provide context, and define your audience. Together, these account for roughly 70% of the difference in output quality. The rest can be developed over time as you gain experience and your use cases become more complex. Prompting is ultimately not a theory to study, but a skill you build through practice.