There is a great deal of excitement about using AI to “build an app.”
That excitement is justified. Today’s AI tools can dramatically reduce the time required to move from an idea to a working prototype.
But building something entirely new is not always the best place to begin.
Most businesses already have valuable information stored in ERP systems, accounting applications, databases, spreadsheets, and other operational tools. The opportunity may not be to replace those systems. It may be to extend the value of the information they already contain.
Sometimes the best place to begin is a CSV file.
Start with a report you already trust
Imagine that your ERP system produces a sales report your organization uses regularly and trusts.
Can you export the underlying data behind that report for a selected date range as a CSV file?
If so, you may already have the starting point for a useful AI-assisted project.
A CSV file is not technologically exciting. That is part of its appeal. It is a simple, widely supported format that allows information to move from one system to another without requiring a live connection to the ERP system.
For an initial prototype, that simplicity can be an advantage.
Rather than beginning with APIs, ODBC connections, security credentials, or a complex systems-integration project, we can begin with a more practical question:
What could we learn—or build—from a carefully prepared export of information we already understand?
Learn the structure before using the data
A sales export might contain fields such as:
- Sales date
- Customer
- Salesperson
- Product
- Product category
- Quantity
- Unit price
- Discount
- Net sales
The objective at this stage is not necessarily to provide an AI tool with actual company transactions. It is to understand the structure of the information: what the fields represent, how they relate, and what kinds of business questions the data could help answer.
Even field names and business definitions should be reviewed before being shared with an external AI provider. Every organization should consider the sensitivity of its information and the privacy terms, security controls, and account settings of the tools it uses.
Fortunately, we do not need real sales transactions to explore the idea.
Replace confidential records with synthetic data
Once we understand the structure, an AI tool can help create synthetic data that resembles the type of information the system contains without reproducing actual customers, products, employees, or transactions.
Synthetic data should be completely fictional. It is different from simply changing customer names while retaining genuine transactions.
For this demonstration, we plan to use five synthetic CSV files:
- Consolidated sales transactions
- Customers
- Salespeople
- Products
- Product categories
The consolidated sales file will contain the fictional sales activity. The supporting files will provide useful names and classifications, allowing the dashboard to organize sales by customer, salesperson, product, category, and region.
The files can be connected through shared identifiers. A product identifier in the sales file, for example, corresponds to a product in the product file.
The user does not need to understand database design to benefit from the result.
Turn those files into a useful sales dashboard
Instead of connecting directly to an ERP system or database, the new version will begin entirely with CSV files containing synthetic sales information.
The dashboard will explore questions such as:
- How are total sales trending over time?
- Which customers generate the most sales?
- Which products and product categories are performing best?
- How do results compare across salespeople or regions?
- Can users filter the information and explore the supporting sales detail?
The result will be presented as a simple browser-based dashboard.
That is an important part of the experiment. We are not asking the user to install a new enterprise application or learn a complicated analytics platform. We are exploring how familiar business information can be transformed into something interactive and useful with today’s AI-assisted development tools.
Start simple, then validate the value
This first version is intentionally limited to sales.
We could add inventory, open orders, purchasing, forecasting, production planning, and many other capabilities. But adding those elements too early would make it harder to evaluate the central idea.
The initial objective is to answer a smaller question:
Can we use a familiar sales export, synthetic data, and AI-assisted development to create something that helps a business see its existing information differently?
If the prototype proves useful, complexity can be introduced later. The CSV files could be refreshed with approved exports. The dashboard could be expanded. A more automated or direct connection could eventually replace the manual process.
Those decisions should follow evidence of value—not precede it.
Extending what you already have
AI creates exciting opportunities to build new applications. It also creates an equally important opportunity to revisit the systems and information businesses already depend upon.
Your existing ERP system does not need to be replaced for you to begin experimenting.
Your first AI project does not need to be a sweeping digital transformation.
It might begin with a sales report you already trust, a carefully reviewed set of fields, and some completely fictional data.
Sometimes the best place to begin really is a CSV file.
What comes next
In the next part of this series, we’ll show the synthetic CSV files and the interactive sales dashboard created from them.
We’ll then go behind the finished product—sharing the prompt, the role Codex played, the iterations required, and where human business judgment remained essential.