You’ve seen the AI demos, attended the webinars, and read the LinkedIn posts. AI is going to change everything. But then Monday morning arrives, you’re back at your desk, and one question remains: where do you actually start?
That was exactly our situation. At Dynappco, we work with our NetSuite ERP system every day. The promise of AI in ERP sounded compelling, and we felt it was time to experience the benefits firsthand: practical solutions to real business challenges that genuinely save time and improve efficiency.
In this article, we’ll share three use cases we tested, what we learned from them, and the patterns that started to emerge. For these experiments, we used Claude as our AI tool.
Starting Point
So where do you begin? The best use cases usually don’t come from a strategic roadmap or a trending LinkedIn post. They come from your own frustrations. Which tasks take up too much time every week? Where do the same mistakes keep happening? Which processes still feel like pure manual work?
We started with three simple criteria. Once we applied them, the first use cases quickly became obvious.
- Repetitive Daily or weekly manual tasks
- Rule-based Clear steps, logic, and validation rules already exist
- Low risk Mistakes are inconvenient, but not business-critical
Use Case 1: Lead generation in NetSuite via prompt
Problem
After a trade show or event, you end up with a stack of business cards. For each card, you manually create a lead record in NetSuite: company name, address, contact person, email address, phone number, etc. This takes five to ten minutes per card, and to be honest, that often means it either doesn’t get done or is delayed for weeks.
Solution
You take a photo of the business card and upload it to Claude. The AI model reads the text on the card, structures the data (name, job title, contact details), and through a NetSuite integration it automatically creates a lead record with the corresponding contact person underneath it.
For data that is not present on the business card, we use pre-prompts. For example, the correct subsidiary is automatically assigned based on the country, and a memo is added containing contextual information about the person and the event.
Prompt used and explanation:
I attended the NetSuite event and collected these business cards. Create a lead record in NetSuite based on the available information. Create a contact person under this lead. Retrieve the VAT/VIES number from reliable sources. Save the record as a draft.
This may look short and straightforward, but the real strength lies in the pre-prompting. For example, the subsidiary is selected automatically based on the country in the address. In addition, the lead record has already been pre-analysed, so there is no need to explicitly define every field.
To avoid hallucinations or incorrect data, the system is configured to request clarification whenever critical information is missing.
Result
This is a perfect example of a task that is small on an individual level, but highly time-consuming in aggregate. The AI processes multiple business cards in about thirty seconds, whereas doing it manually can take up to an hour. More importantly, it ensures the task actually gets done instead of business cards sitting in a pile indefinitely.
Important: Always create the record as a draft. A human should validate it before it is finally saved. AI does the heavy lifting; humans retain control.
Use Case 2: Automatically creating sales orders from email
Problem
Customers send purchase orders via email. Sometimes as PDF attachments, sometimes as plain text in the email body, and sometimes as Excel files. Someone has to open the email, interpret the data, look up the correct customer record in NetSuite, match the product items, and manually create a sales order. With just ten orders per day, this already adds up to half a working day.
Solution
We built an automated workflow (AI Agent) that connects three systems: Gmail (to read incoming emails), Claude AI (to extract and validate data), and NetSuite (to look up customers and items and create sales orders).
The flow works as follows:
- Incoming purchase emails are automatically labeled in Gmail (e.g. so-inbox).
- The AI agent reads all unprocessed emails from this inbox.
- For each email, the model extracts customer details, item codes, quantities, requested delivery date, and any remarks.
- The AI agent searches NetSuite for the customer using name or email address and validates item codes.
- If all data is complete, it creates a sales order with status “Pending Approval”, making it visible for an order processor to review and approve.
- If data is missing (e.g. unknown item codes or unclear descriptions), it either generates a draft response email with targeted questions for the customer or sends an internal review notification to a staff member to validate the order.
Result
The key strength is not just order creation, but the validation layer. The AI does not push incomplete orders into the system; it proactively requests missing information, preventing downstream errors.
The complexity lies in edge cases: customers using their own internal item codes instead of yours, orders spread across multiple emails, or CC’d stakeholders also placing orders. For these scenarios, we implemented exception rules that trigger human review.
Despite these edge cases, the core process runs reliably and saves not just hours, but days of work time.
Important: This goes beyond a simple prompt. This is an AI Agent: an automated workflow combining multiple systems.
Use Case 3: Quarterly Business Review (QBR) preparation
In Use Case 1, we started from a simple prompt with predefined rules. In Use Case 2, we built an AI agent connecting multiple systems into a full workflow. In Use Case 3, we illustrate the difference between these approaches and when to use each.
Problem
Every quarter, a business review must be prepared: revenue per subsidiary, outstanding receivables, top customers, and deviations versus the previous quarter. This requires pulling data from NetSuite, exporting it to Excel, building charts, assembling everything into a PowerPoint deck, and writing the narrative.
Depending on the number of entities and data complexity, this process easily takes several working days.
Solution
There are two approaches to achieve this, each with different strengths.
Solution 1: Ad hoc prompt
The first approach is an ad hoc prompt for real-time analysis based on existing NetSuite reports. You reference report IDs directly in the prompt, ensuring full traceability of the data source and accuracy of the figures.
The AI then analyzes the data and generates either a presentation or an interactive dashboard. This approach is ideal for one-off analyses or when answering specific questions outside standard reporting cycles.
Example prompt:
You are a senior Financial Controller. Use the following NetSuite reports:
Profit and Loss (ID: 200)
Sales by Customer Summary (ID: 272)
A/R Aging Summary (ID: 274)Analyze revenue, top customers, and accounts receivable for Q3 and Q4 2025, including a comparison between both quarters. Provide the analysis per active subsidiary and on a consolidated basis.
As a data analyst and business consultant, highlight the key observations and attention points for 2026: financial and operational risks, growth opportunities, and actions to improve receivables management.
Solution 2: AI Agent
The second approach is a reusable agent that can be activated for any subsidiary or entity simply by instructing the AI to run the QBR Agent. Unlike the ad hoc prompt, you don’t need to restate requirements each time. The agent automatically goes through four phases in the background:
| Phase 1: Environment detection | Maps the full NetSuite configuration, including subsidiaries, currencies, fiscal calendar, and enabled features. |
| Phase 2: Dynamic query construction | Builds the appropriate queries based on that configuration, tailored to the specific NetSuite environment. |
| Phase 3: Data extraction & validation | Executes the queries and checks the results for internal consistency before any output is generated. |
| Phase 4: Output generation | Delivers the final result as either a presentation or an interactive dashboard. |
The key difference compared to the ad hoc prompt is that the agent first inspects the NetSuite environment for changes. It then re-generates and executes the relevant queries based on the current system state, ensuring up-to-date and accurate data.
This makes it robust even in heavily customized NetSuite environments, without requiring the user to know which fields, transaction types, or segments exist in advance.
Result
What previously took multiple working days of preparation is reduced to a 30–60 minute session, including human review of the output.
The ad hoc prompt remains the fastest option for one-off analyses, while the reusable agent delivers its value when the same review must be repeated quarterly or across multiple entities.
Important: The output is a starting point, not a final deliverable. While the data is validated through internal cross-checks, a human always reviews the findings before they are presented in a quarterly meeting. As with Use Cases 1 and 2: AI does the heavy lifting, humans retain control.
Closing
Three use cases, three levels of complexity, but the same underlying pattern: start from a concrete problem, test an AI-driven solution, and always keep human validation in the loop.
The biggest gains are not found in the spectacular, but in the repetitive work: entering business cards, processing orders, summarizing quarterly figures. Individually small tasks, but collectively they consume hours each week. This is exactly where AI already makes a measurable difference today.
In part 2, we will go deeper into how organizations transition from isolated prompts to integrated AI solutions. In part 3, you will learn how to design effective prompts and how they influence output quality.
Ready to start working on your own use cases? Contact our consultants to explore the possibilities.