When your Fishbowl database does not spark joy
If your Fishbowl has been live for a few years and nobody trusts the numbers, the usual fix is to clean it up in place. That gets slow, because Fishbowl blocks a lot of changes once records have history. The other option is to start a clean database and bring over only the data you trust. With an AI agent connected to both systems through ILC MCP, the data work can happen in a single working session, and the whole rescue takes days instead of weeks. We are offering this to early-access customers now.
Before
These are the problems we see most often in a Fishbowl database that needs rescuing, and why they’re hard to fix where they are:
| Problem | Why it’s hard to fix in place |
|---|---|
| Parts on the wrong unit of measure, or conversions that don’t match how you buy and sell | Fishbowl won’t change a part’s UOM while it has open orders, and the history keeps the old conversions |
| On-hand quantities that don’t match the shelf | Years of adjustments, partial counts, and stock left in old locations |
| Average costs nobody believes | Every receipt and adjustment at a bad cost or conversion is part of today’s cost |
| Locations, parts, and customers nobody uses | Fishbowl won’t delete a customer with inventory history, so test and duplicate records stay |
| The same customer or vendor entered two or three ways | Orders are split across the copies, and merging them means rewriting history |
You can work through each of these in the live system, but every fix has to respect the history behind it. A new database starts without it.
The setup
You connect two ILC MCP connectors in the same Claude or ChatGPT chat:
- The old Fishbowl, with imports left off. The agent can query anything there and change nothing.
- A new, empty Fishbowl, with imports turned on one type at a time. Parts, products, customers, and vendors are create-only.
The old system keeps running until you switch over, and it stays available afterward for looking up history.
The session
In one chat, your team and the agent go through the old data and decide what comes across:
- What’s active. The agent lists the customers, vendors, and parts with activity in, say, the last two years. If a record hasn’t been used in two years, it doesn’t come across.
- Duplicates. It points out records that look like the same company or part, such as “Acme Inc” and “ACME, Inc.”, and you choose which name to keep.
- Units of measure. It compares each part’s UOM and conversions with how you’ve actually been buying and selling it, and you set the right ones once.
- Locations. You decide which locations the new warehouse needs, instead of copying every one.
Then the agent loads the new Fishbowl in import order: setup data, vendors, customers, then parts and products. Each import shows a preview, and someone on your team approves it before it goes in.
After
The new database starts with only the records you chose, on the units of measure you agreed on. On-hand quantities come from a fresh count, which the agent imports as opening inventory with a cost for each part. That count is the starting point for both quantity and average cost, without the old adjustments behind it.
Catching up
The old system keeps running while the new one is set up, so new customers, parts, and changes keep coming in. Before you switch over, a short catch-up session finds what was added in the old system since the first session and imports it into the new one, with the same previews. If records that already came across have changed, an admin turns on updates for that type, and the preview lists what will change. You can run a catch-up as often as you need until the switchover day.
What still takes time
The data work fits in a session. The rest of the rescue doesn’t:
- Counting the stock, which the opening inventory depends on.
- Connecting accounting and agreeing on opening balances with your accountant.
- Setting up users and their rights, which is done in Fishbowl directly.
- Bills of materials, kits, and part, product, and vendor pricing, which still go through Fishbowl’s own import screen.
- Training and the switchover day.
So the data work fits in one session, and the whole rescue takes a few days. If your Fishbowl is in this state and you’ve been putting off the cleanup, tell us what you’re dealing with.
Related: