What does an agent-assisted Fishbowl implementation look like?
If you’re implementing Fishbowl, most of the time goes to loading data: units of measure, locations, vendors, customers, parts, opening inventory, and open orders. Each one is a CSV import with its own columns, and Fishbowl rejects a line over one wrong column. With an AI agent connected to Fishbowl through ILC MCP, your team can hand it a spreadsheet, check a preview, and approve the import in the chat. We’ve been running this on our demo Fishbowl and are looking for the first implementations to use it.
Where the time goes
Fishbowl publishes a suggested import order for a new database. Roughly, it comes in four groups:
- Setup data: units of measure and conversions, locations, payment terms, carriers, custom fields, QuickBooks classes, and tax rates.
- Master data: vendors, customers, customer groups, parts and products, pricing rules, the product tree, and default locations.
- Opening inventory: Add Inventory, with quantities and costs.
- Open orders: sales orders and purchase orders.
The work is mostly reformatting. Your vendor list comes out of the old system with different column names, the part list has blank cells, and the customer file uses payment terms that don’t exist in Fishbowl yet. Some imports also create records you didn’t ask for. The location import creates a missing location group, and the customer import can create a QuickBooks class from a typo. Every failed import means another round in Excel.
Planning
We start with a planning call. We go through the import order with your team, decide where each list comes from, and mark which ones Fishbowl’s Accounting Configuration Wizard can bring over from QuickBooks instead. The agent can check what is already in the database, such as how many units, carriers, or customers exist, so the plan starts from what’s already there.
We then turn on the import types you need, one at a time. Each one is off until an admin enables it, and they sit under separate permissions for setup data, master data, and inventory.
A loading session
A typical session for vendors looks like this:
- You upload the vendor export from your old system and ask the agent to load it.
- It maps your columns to Fishbowl’s vendor import columns and points out rows with missing required values.
- It shows a preview of what will be imported.
- You approve it, and the agent imports it under your own Fishbowl login, in batches of up to 500 rows.
If Fishbowl rejects a line, the agent shows you Fishbowl’s message, fixes the row with you, and tries again. It only sends the columns you set, because Fishbowl clears any field that’s sent empty.
Parts, products, customers, and vendors are create-only by default. If a row matches a record that already exists, the import refuses it and names the record, so running a file twice doesn’t overwrite anything. Updating existing records takes a separate switch, and the preview then lists what will change. Add Inventory has its own permission, since it changes stock and cost as soon as it runs.
Who does what
Your team does the loading, because they know the data. We set up the connection, write the import plan, turn on each import type when it’s needed, and review each group before the next one starts. Users and their rights are set up in Fishbowl directly, never through the agent.
What’s not there yet
Bills of materials, kits, and part, product, and vendor pricing (PPVP) files still go through Fishbowl’s own import screen. Orders get the strictest checks, since every customer, vendor, part, and carrier has to exist already. Other imports check their columns and then rely on Fishbowl’s own validation. The agent doesn’t check the import order on its own, so the plan sets it.
We don’t know yet how much time this saves on a real implementation, because none has used it yet. If you’re about to start one and would like to try it, we’d like to hear from you.
Related: