SKU Manager: Troubleshooting & FAQ

Quick answers to common questions and issues (build V3.5.1.0)

Start Here.

Most “it’s not working” reports come down to one of four things: the filter isn’t enabled, the AI server isn’t running, the model can’t read images, or billing isn’t active on the API account. The first section below walks them in order.

Nothing Showing in the AI Panel #

Work through these in order — each one is a real case we’ve seen:

Restart the whole application first. Thirty seconds, and it is genuinely the most common fix. Do this before starting a troubleshooting spiral.
Is the AI Sku Analysys panel visible? View → Panels if not.
Is your filter enabled? Open Home → Filters and check its checkbox — a freshly created filter can arrive disabled.
Is the filter’s action Apply AI Prompt, with your profile selected?
Is the AI server running? Data → AI Sku Analysis → Start AI Server. Red italic status text in the panel is normal — it’s not an error.
Is the feed Started, and does your search actually match the filter’s condition? The analysis only runs on listings the filter catches.

No Filter, No Analysis.

The setup wizard does not create the filter — it’s a separate step in Home → Filters. If you finished the wizard and nothing has ever run, this is almost certainly why. See Create the Filter in the Quick Start Guide.

Is My Filter Actually Triggering? #

Not sure whether your filter matches anything? Test it visually:

Load some search results first.
Go to Home → Filters and edit your filter.
Change the Action from Apply AI Prompt to Format Rows, pick a Back Color, and click OK.
Look at your results — the highlighted rows are exactly the listings the AI would analyze.
Change the action back to Apply AI Prompt when you’re done.

The Analysis Ignores the Photos #

If defect calls look blind to obvious photo damage, or the analysis never mentions what’s visible in the images, check your model first.

Model Choice is a Gate, Not a Preference.

Use gpt-5.4-mini — it reads images. If your install’s dropdown still offers gpt-4o-mini or an undated “GPT 4 mini”, avoid both: one cannot read images (every photo-based analysis silently comes back blind, with no error to tell you why), and the other doesn’t load at all.

Open Data → AI Sku Analysis → AI Settings, check the Model dropdown on the AI Configuration tab, and save with Start Service [Save Configuration] if you change it.

AI Column Problems #

Extra Columns I Never Asked For #

The column list is auto-detected from your prompt’s JSON and includes every key it finds — including junk. Open your profile → Fields & Columns → untick anything that isn’t a real column. Keep only the ones you defined under ## AI COLUMN CONFIGURATION.

A Column is Blank or Missing #

The column must be defined in the prompt’s ## AI COLUMN CONFIGURATION section with an exact name and exact values, and ticked under User-Defined Custom Columns (Results).
Give every column an Unknown fallback so it’s never blank, and keep values exact — Yes, not yes or Y. The grid sorts and filters on them verbatim.
After editing the prompt, reopen the profile so it re-reads the columns — your new column appears under Detected — not selected; tick it.

Template Showing Raw Code #

Check the Display Template contains valid HTML and was actually saved.
Check the template’s placeholder names match your prompt’s JSON fields — the assistant that wrote your prompt keeps the two in sync, so if you edited one by hand, that’s the first suspect.
Paste the whole template again from your master file — the code boxes can truncate what you paste, so never hand-edit inside the box.

AI Analysis Takes Too Long #

Analysis typically takes a few seconds per listing after it arrives — reading photos is the slow part. If it’s consistently slow:

Trim your prompt to what actually matters for the buy decision.
Image analysis costs the most time — add conditional logic so images are only analyzed when certain criteria are met.
Submit a feedback trace and take it to your AI project: “This is taking longer than expected. Can we reorganize the prompt for speed based on what’s most critical?”

API Provider Issues #

“Rate Limited” or Throttling Errors #

New API accounts are sometimes throttled at first — check your email for verification requests from the provider. Usually resolves within 24–48 hours.

Billing or Authentication Errors #

Verify billing is active on your API account — this is the #1 issue people hit.
Check the API key is complete, with no missing characters. Test it in the provider’s dashboard.
Check your usage in the provider’s dashboard after a few days to see actual spend.

A Change Doesn’t Take Effect #

Re-paste the whole prompt or template from your master text file — don’t hand-edit in the box.
Save with Start Service [Save Configuration] — that button is the save; there is no separate save button.
Test on fresh listings. Results already sitting in the grid were analyzed under your old settings.

Frequently Asked Questions #

General #

What Model Should I Use? #

Model	Price (in/out)	Notes
`gpt-5.4-mini`	$0.75 / $4.50	The recommended default — good quality, low cost, and proven to read images
`gpt-5.4-nano`	$0.20 / $1.25	The cheapest option — the dropdown lists it under “Recommended”
`gpt-5.4`	$2.50 / $15.00	Higher quality, higher cost
`gpt-5.5`	$5.00 / $30.00	The most capable and most expensive

You pick the model from a dropdown in the profile’s AI Configuration tab — there’s nothing to type or format. OpenAI is the path our guides walk, but the app also supports Google (Gemini), Anthropic (Claude) and xAI (Grok) — pick your provider, paste that provider’s API key, and each model dropdown marks its own Recommended entry (Gemini’s is gemini-2.5-flash-lite at $0.10/$0.40).

How Much Does it Cost to Run? #

Usually cents per listing. The model dropdown shows each model’s pricing, and your provider’s billing dashboard shows actual usage after a couple of days. One thing to watch: the first run analyzes every result already in your grid — cap it at Grid → Initial Results Max (e.g. 5) until you know things work.

Can the AI Do Calculations and Math? #

Yes. Pricing logic, profit calculations, quantity math and similar can all go in your prompt.

Working with Product Data #

How Many Products or Variants Can I include in My Prompt? #

Dozens to hundreds works well.
Thousands is too many — it slows processing and increases costs. For big lists, use Internal Endpoint mode instead: it matches listings against your own CSV locally, with no per-listing API cost.

Can I Use My Excel Price Sheet? #

Yes. When creating your prompt with Claude or ChatGPT, upload the spreadsheet, explain what each column contains, and the assistant builds the matching logic into your prompt. Works well for different models with different max prices, variants you buy vs. don’t, and condition-based pricing.

AI Columns vs. Display Template #

What’s the Difference? #

	Display Template	AI Columns
Where it shows	AI Sku Analysys panel	Results grid
Best for	Detailed analysis, explanations	Quick Yes/No decisions
Format	Full HTML with formatting	Short, exact values

Example: you only buy listings that accept returns. Add an AI Column called Can Return that outputs exactly Yes, No or Unknown — then sort or filter the grid on it and skip the No rows without opening a single listing.

Running Both Modes #

Can I use AI Analysis and Internal Endpoint at the same time? #

Yes. They are separate systems with separate servers — the Data tab has a Start button for each, and each renders into its own panel. A filter triggers AI analysis; Internal Endpoint needs no filter at all — your search terms are its scope.

Still Stuck? #

Check the basics: filter enabled? AI server running? Feed started? API billing active?
Use feedback traces: select the bad result, click Feedback on the AI panel, rate it and describe what’s wrong. That captures a complete diagnostic snapshot — the data sent, the exact prompt, the AI’s answer, and the images — which you (or support) can work from.
Contact us: Live Support at uBuyFirst.com, or su*****@*******st.com.