With AI, anyone can build their own software. A prototype now takes a weekend. But a prototype is not a tool a team can rely on every day. What’s missing is someone to own the project for months, and that is what costs the most, far more than a subscription.
Prototyping has never been easier
On a Friday evening, an IT manager, or a curious sales director, opens a coding assistant. They describe the problem: read the purchase orders that arrive by email and turn them into a clean file, ready to import into the ERP. By Sunday, it works on the ten PDFs they pulled from their inbox.
That’s real, and it’s great news: AI has made experimenting almost free. On Monday, the question comes naturally: why pay a subscription for something we just built in two days?
What’s missing: someone who owns the project
Software used every day needs someone who decides, monitors, fixes and improves it. Not for a weekend: for months, then years.
- Deciding: which orders the tool handles, what to do with a price that doesn’t match the price list, who approves what, what goes straight into the ERP.
- Monitoring: knowing every day what went through, what failed, what is waiting for someone.
- Fixing: the customer who changes format, the replaced item, the ERP update that breaks the import, the AI model that suddenly reads a column differently.
- Improving: adding quotes after orders, a second site, a new mailbox, another ERP after an acquisition.
None of this shows in Monday’s demo. It adds up to days every month, and it usually falls on the company’s most valuable people: those who know both the business and the tools. When they change roles or leave, what remains is often a black box nobody dares to touch.
You don’t know what you need until you’ve tried something that works
There is a second, quieter trap. At first, the need looks simple: read an order and enter it. The details that make the difference only show up along the way:
- the customer who orders by the carton while the ERP counts units;
- the item reference written their own way, or no reference at all;
- the “cancel and replace” that arrives an hour after the first order;
- the exceeded credit limit that should never have gone through.
You only discover these details by using something that works, on real orders, with the real team. An in-house tool runs into them one by one, at your expense, while sales administration keeps entering everything by hand. A solution already in production at other companies has met them before: we describe a few in our articles on customer references and unit of measure conversion.
Core business: a rule AI doesn’t change
Plenty of people have said it before us: focus your efforts on what sets you apart from your competitors, on your core business, and hand the rest to those whose business it is.
For an electrical supplies distributor, a food wholesaler or a hygiene products manufacturer, what makes the difference is the product range, availability, advice and the quality of customer relationships. Not the code that reads a PDF.
AI doesn’t change this rule. It just makes it easier to forget, because the first step now costs almost nothing.
When building it yourself still makes sense
There are cases where building in-house holds up: a small, very uniform flow that a simple import can handle; a stable technical team that will have time for this over the long run; or an order intake process that is genuinely part of your competitive edge. And a prototype remains an excellent way to learn, as long as you don’t mistake it for a production tool. We cover these cases, and a method for deciding, in build or buy order automation.
Our take at Eledone
Owning the project is our job. We define the processing rules with you, monitor what the agent does, adapt it when your customers, catalogue or ERP change, and keep up with how AI models evolve. Your team stays in control where it matters: Eledone only shows them what needs human approval.
You can see for yourself with the online trial on one of your orders, or read the details on the order entry page.
Frequently asked questions
Can an AI-built prototype be enough?
To learn and to pin down your needs, yes. To process your customers’ orders every day, you need someone to monitor it, fix it and improve it. That is the time to cost out before deciding.
What does an in-house tool really cost?
The initial build weighs little. Most of the cost is in the upkeep: edge cases, ERP updates, changes in AI models, the person who has to stay available for all of it, and the time during which the team keeps entering orders by hand.
What stays in-house if you choose an existing solution?
The decisions. Your business rules, your priorities, approving the cases that need it and your customer relationships stay with you. What you hand over is the technology and its upkeep.
Want to see it on your own orders, without a project to carry?
Book a demo