The work rarely stops at sending. When a customer asks a question or changes the specification, the agent prepares a draft reply or a revised quote. That version goes through the same approval as the first one, and the earlier version stays on file. The agent does not pressure customers. It keeps the overview that would otherwise slip between two mailboxes.
Quoting software with AI or an AI agent?
If your whole quote comes together in one program, the AI features in your quoting software are often enough. If the work runs across several systems, an AI agent helps more. Many quoting tools already offer text suggestions, rewriting, or a draft from a template. But the request often sits in your mailbox and the customer details in your CRM. The prices live in a price list, and the quote itself in your quoting software. The agent gathers what it needs, prepares the draft in your existing software, and keeps an eye on the expiration date. Your quoting software still handles the pricing calculations. So start with where the groundwork costs time today.
How can you automate quotes without losing control?
Let the agent prepare and let a person decide: that is our starting point when automating quotes. You approve prices, discounts, and terms before anything goes out. An incorrect amount in a polished quote can outweigh the benefit of preparing it faster.
The work of our own Hermes installation at Voltti also starts with a defined task. The agent carries it out and records what happened. In our CMS, it sometimes edits and publishes directly; we check the result afterward and make corrections. For quotes, we choose a different boundary: a person approves the price and terms before sending.
The controls behind that boundary are straightforward. Do not give the agent a sending path that bypasses approval, retain each version of a quote, and require a person to review an unusual price. The workflow must enforce those rules; a written instruction alone is not an access control. The OWASP guidance on excessive agency in LLM applications puts it this way: "Utilise human-in-the-loop control to require a human to approve high-impact actions before they are taken." In other words, a person approves high-impact actions before they happen. That lets you delegate preparation while you keep control of the decisions that matter.
How do you start small with quote automation?
Start with one type of request that comes back often enough, not your entire quoting process. Pick the workflow that frustrates your team most, often standard requests that already follow a set pattern. Document the pricing rules and gather around ten recent quotes as references. Then spend a few weeks having the agent create drafts that a person always checks.
Decide before the pilot how you will judge it. Note how long a standard quote takes today and how often someone has to correct a draft. Track the same two things while the agent prepares drafts, then compare after a few weeks. If someone still has to rewrite most drafts, fix the pricing rules or templates first. Expand to follow-ups and other request types only when the results hold up.
We explain that approach, one workflow before expanding, in automating business processes step by step. Our page on AI automation for your business explains which tasks suit fixed steps and which need a flexible approach.
Want to see what this could look like for your team? Voltti's approach explains how a pilot works: one task, a defined period, and a clear point where you have the final say. For the broader setup, explore Voltti's AI agent solution. Discuss your pilot and we can review your quoting workflow together, for businesses in Belgium and the Netherlands.
