Most vendors talk about AI accuracy in the abstract: "hallucination rates," "grounding," benchmarks nobody can verify. We'd rather show the actual mechanics: the four places in a live Velaro deployment where the system is built to prefer "I don't know that field" or "a human should check this price" over a confident, wrong answer.
The field mapper stops the AI from guessing at your CRM's naming
Every CRM names things differently, one system's customer_email is another's ContactEmailAddress1. Velaro's field mapper is a per-site mapping table that translates Velaro's internal field names to whatever your connected system actually calls them, so the AI writes to the field that exists instead of inventing one that doesn't.
When a mapping is missing, the system falls back to a direct pass-through rather than silently dropping the value, and any required field that still can't be resolved raises a clear error instead of submitting a half-filled record. This isn't a new capability, it's had a round of real usability fixes recently: added schema discovery for Zoho CRM, Pipedrive, Monday.com, Guesty, Cloudbeds, and Uplisting, a fix for a required-field check that was flagging fields incorrectly on Monday.com, and a fix for a rename operation that could silently merge two distinct mappings into one.
Per-site mapping
Every site's field map is independent. Two customers on the same CRM can name fields completely differently.
Fails loud, not quiet
A required field with no mapping throws an error at write time instead of shipping an incomplete record to your CRM.
Pass-through default
Unmapped optional fields still flow through under their original name instead of getting dropped.
Generic AI node or dedicated integration node, how to choose
Every Velaro workflow that hands a conversation to AI uses the same underlying "AI Chatbot" node. The difference is how much of the setup you do by hand versus how much is pre-wired for a specific integration.
Neither one is the "advanced" option, they're the same engine with two different amounts of pre-configuration. The workflow builder nudges you toward the dedicated node when it detects you've hand-picked a full set of skills from a single integration, since that's almost always what the dedicated node already does out of the box.
Want to see the workflow builder in action?
Start Free Trial →Product comparisons that never invent a detail
When a customer asks the AI to compare two products, the comparison is built entirely from fields your store already has: title, vendor, tags, price range, inventory count, variant options, and a short excerpt of the real product description. The AI is not generating a description of the product from its own general knowledge; every field it presents is a direct read from your catalog.
Product images work the same way. Including an image in a comparison is an opt-in setting, off by default, and when it's on, the image is a direct link straight from your product catalog, never an AI-generated or AI-guessed image. If your catalog has no image for a product, the comparison simply omits one rather than filling the gap with something that isn't real.
The rule across every skill that touches your catalog is the same: show what's actually there, and show nothing rather than something invented.
Quotes the AI can build, but not always send on its own
Velaro's quoting skill can walk a customer through pricing in conversation and build a real, itemized quote, but it treats prices differently depending on where they came from.
Price comes from your catalog or index
The AI looked the item up against real, ingested pricing data. These line items can flow straight through to a shareable quote under your account's normal approval settings.
Price comes from an admin-configured override
You've explicitly told the system what to charge for this item. Same as above, it's a known, human-set value, so it follows your normal approval settings.
Price comes directly from the AI, with no catalog match
The model supplied a number itself rather than finding one. Any quote containing even one line item like this is forced into a pending-approval state, full stop, regardless of whether your account has approval requirements turned on for anything else.
That third path is the one that matters. A single AI-guessed line item on an otherwise real quote doesn't get to ride along and auto-send with the rest, the whole quote waits for a person before it can be shared with a customer.
The Bottom Line
None of these four guardrails depend on the AI model being smarter or better-prompted, they're structural checks in the code path itself: a field either has a real mapping or the write fails loudly, a comparison either has a real catalog value or it's left blank, and a price either came from a real source or the quote stops for a human. That's a deliberate design choice, not a limitation to work around.
Ready to see Velaro in action? Start a free trial, no credit card required.
Start Free Trial →Frequently Asked Questions
Does Velaro's AI ever write to a CRM field that doesn't exist?
No. The field mapper resolves every outbound value against your site's real, per-account field mapping before it writes anything. If a required field has no mapping and can't be resolved, the write fails with an error instead of silently sending an incomplete or mismatched record.
What's the difference between the generic AI Chatbot node and a dedicated integration node in a workflow?
Both use the same underlying AI engine. The generic node has a manual skill picker so you can combine skills from multiple integrations; a dedicated node (like a NetSuite-specific node) is pre-scoped to one integration's config and credentials, which means less manual setup when a workflow step is really just talking to one system.
Can the AI show a product image it generated itself during a comparison?
No. Product images in a comparison are an opt-in setting that, when enabled, links directly to the real image already in your product catalog. If your catalog has no image for an item, the comparison leaves it out rather than substituting a generated or approximate one.
Will Velaro's AI ever send a quote with a price it made up, without anyone reviewing it?
No. Any quote containing a line item priced directly by the AI, rather than pulled from your catalog or an admin-configured override, is automatically placed into a pending-approval state before it can be shared with the customer, regardless of your account's other approval settings.