“AI assistant” gets thrown around so loosely these days that it’s fair to be a little skeptical when you hear it attached to your ERP. Is it actually useful, or is it a chatbot bolted onto a login screen so someone can put “AI-powered” in a pitch deck?
Here’s the honest answer: a well-built AI assistant sitting on top of your inventory and order data can do real, specific work, not “vibes-based” work, but the kind of thing you’d otherwise hand to a person with a spreadsheet and twenty free minutes. The trick is understanding what it’s actually good at, so you’re not disappointed when you ask it something it was never built to handle.
Let’s get into what that looks like in practice.
The Three Buckets Most Requests Fall Into
Almost everything you’d want to ask an ERP-connected assistant lands in one of three categories: finding things, building things, and explaining things. Once you think of it that way, it’s a lot easier to know what’s worth asking.
1. Finding Things
This is the “go dig through the data so I don’t have to” category. Instead of opening a report, setting filters, and squinting at a spreadsheet, you just ask.
Things like:
- “What’s below safety stock right now?”
- “Which SKUs haven’t sold in the last 90 days?”
- “Show me every open order for a specific vendor.”
None of this is magic, it’s the same data that’s already sitting in your system. What changes is the retrieval speed. Instead of clicking through three menus, you get an answer in the time it takes to type a sentence. If you’ve ever spent ten minutes of a meeting waiting for someone to “pull that up real quick,” you already know what this saves.
2. Building Things
This is where it gets genuinely more useful than a search bar. A capable assistant doesn’t just find the data, it can turn that data into an actual output you were going to create anyway.
The most common version of this: reordering. Instead of manually building a purchase order line by line, you can say something like “reorder everything that’s below safety stock” and get a draft order back, organized and ready to review instead of dumped in a random list. You’re still the one who approves it. The assistant just did the tedious middle part.
The same idea applies to documents. Supplier invoices, packing lists, quotes; normally someone has to sit there and manually key that information into line items. A document-aware assistant can read what’s on the page and turn it into structured data without a human doing the typing. It’s the digital equivalent of having someone transcribe your mail instead of doing it yourself at 6pm.
3. Explaining Things
This is the “what am I actually looking at” category, and it’s the one people underestimate the most.
Instead of building a custom report from scratch (which, depending on your ERP, might mean learning a query language or bugging whoever on your team knows SQL), you can just ask a direct question: “How much did we spend with this vendor last quarter?” or “Which products had the most returns this year?” The assistant builds the report on the fly, based on a plain-English question instead of a technical one.
This matters more than it sounds like on paper. A lot of useful reporting never gets built, not because the data isn’t there, but because nobody had the time (or the SQL skills) to build the report that would’ve surfaced it.
Lowering that barrier means more questions actually get asked — and answered — instead of quietly going “we should really look into that someday.”
What It’s Not Good At (Yet)
In the interest of not overselling this: an AI assistant connected to your ERP isn’t a replacement for judgment. It can find the SKUs below safety stock, but it doesn’t know that your best vendor is on vacation next week, or that you just had a conversation with a customer that changes the forecast. It drafts the order — it doesn’t decide, unprompted, that now’s a bad time to place it.
Think of it less like an autonomous employee and more like a very fast, very literal assistant who’s read the entire database but hasn’t sat in on your Monday meetings. That combination of fast retrieval, zero office politics context, is exactly why the “review before you approve” step matters. It’s doing the first draft, not the final call.
Why This Matters More for Businesses on Fishbowl (or Similar ERPs)
If you’re running an ERP like Fishbowl, there’s a decent chance you already know the gap this fills. Fishbowl is a genuinely powerful system, but a lot of its power lives behind menus, filters, and reports that take some training to use efficiently. That’s not a knock on Fishbowl, most full-featured ERPs work this way, because there’s a lot of data to organize and no single screen can show all of it at once.
An AI assistant layered on top doesn’t change what the ERP can do. It changes how many steps it takes to get to it. That’s a small-sounding difference until you add up how many times a week someone on your team needed one specific number and had to go hunting for it.
So, What Should You Actually Try Asking It?
If you’ve got access to one of these tools and aren’t sure where to start, a decent rule of thumb: start with whatever question you already ask a coworker most often. “What’s low on stock?” “What’s still outstanding with this vendor?” “Can someone turn this invoice into a PO?” If it’s a question you’re used to routing through a person, it’s probably a good first test for the assistant.
You’ll figure out fast where the useful edge is — and where you still want a human double-checking the details before anything ships out the door.
FAQ
Can an AI assistant place orders on its own, without approval?
In a well-designed setup, no — it drafts based on your data, but a person reviews and approves before anything is finalized. That review step is the point, not a limitation.
Do I need to know SQL or a query language to use one of these tools?
No, that’s the main advantage. You ask in plain language, and the assistant builds the report or pulls the data behind the scenes.
Does an AI assistant replace the reports I already have set up?
Not necessarily — it’s more useful for the questions you haven’t built a report for yet, especially one-off or unusual questions that wouldn’t justify building a permanent report.
Can it read documents like invoices, not just answer questions about existing data?
Yes, if the tool supports document parsing — you can hand it a supplier invoice or packing list and get structured line items back instead of typing them in manually.
Is this only useful for large companies with a lot of data?
Not really — smaller teams often benefit more, since there’s usually nobody whose full-time job is reporting. The assistant ends up filling a gap that would otherwise just go unfilled.
