AI analysis troubleshooting

Most analyses come back clean. But every contractor will eventually hit a job where the photos came out dark, the dimensions look off, or a fixture got missed. This article covers the common issues and the fastest way to fix each one.

"Analysis failed" banner — what now?

If something goes wrong during analysis, you'll see a red banner inside the room with a short explanation and a ↻ Retry analysis button. Tap it to run the same analysis again. Most failures are transient (network blip, AI service briefly overloaded) and a retry fixes them.

[!steps]

1. Read the banner — it'll tell you what went wrong in plain English.
2. Tap ↻ Retry analysis.
3. Status flips back to Analyzing… and the room runs through the pipeline again.

If the retry also fails with the same message, the cause is probably something repeatable — often the room is carrying more photos than one analysis can chew through cleanly. Rooms work best with a handful of photos of the same area; if you've piled in a dozen, delete the weakest few or split the room into two smaller rooms and try again. If you keep hitting it, contact support.

[!note] BidCraft already retries a struggling analysis a couple of times on its own before it ever shows you a failure — so by the time you see that banner, the easy retries are spent. Refreshing the page won't restart it either; the room sits in the failed state with the Retry button waiting for you. That's intentional, so you don't come back to a project you thought was settled and find it spinning.

"Photo too dark or blurry"

BidCraft checks every photo before it goes through analysis and flags the ones it can't read well — too dark, too blurry, low contrast. It doesn't throw the photo out; it runs the analysis anyway and warns you that the numbers off that photo are shakier than usual. Treat the warning as a prompt to go back and get a better shot before you trust the measurements.

[!steps]

1. Read the warning — it'll say the photo is dark, blurry, low-contrast, or similar, and suggest the fix.
2. Retake the photo with the issue addressed.
3. Re-upload, delete the bad one, and re-analyze the room.

Common causes and fixes:

Wrong dimensions

The single most common issue. The AI looked at a photo and pulled a measurement that doesn't match reality. Maybe the shower came back as 4 feet wide when it's actually 3, or the floor sqft is 20% off.

You have two paths to fix this, and one is dramatically better than the other.

[!tip] Don't retake photos. Record a voice note with the correct number. Say "the shower is actually 3 feet wide, not 4". Re-analyze the room. The AI will use your stated dimension and the proposal will be correct. This takes 10 seconds.

Why voice notes win:

• They're the highest-trust dimension source in the whole app
• They take 5 seconds to record
• The AI updates all derived dimensions (sqft, perimeter, etc.) automatically
• You don't have to go back to the job site

If you can't record a voice note (you're back at the office), just override the dimension directly in the Measurements & Scope section. Same effect, slightly more manual.

Missing fixtures

The AI didn't see your vanity, or it thought the toilet was a bidet, or it counted one niche when there are two. Usually this means the fixture was at the edge of a photo, hidden behind something, or not in any photo at all.

[!steps]

1. Add a voice note describing what's missing: "there's a 36-inch vanity on the east wall, two niches in the shower — one big one at the bottom, one small one at chest height."
2. Re-analyze the room.
3. The AI picks up the additional context and adds the fixture.

If voice doesn't work, you can also upload a single photo focused on the missing fixture and let the analysis run again.

Video uploaded but no narration captured

When you upload a video that you narrated during filming, BidCraft pulls the audio off in the background and transcribes it into a voice note for you. Most of the time this just works — a minute or two after upload, you'll see a green ✓ Narration pill on the video and the transcript lands in your Notes for that room.

When the pill comes back as No audio or Failed, here's what to do:

[!steps]

1. Tap the video chip inside the room to expand the preview.
2. Watch a few seconds — confirm your voice is actually audible in the file (sometimes phone mics get covered, or the room was too noisy).
3. Tap Extract narration (for No audio) or Retry transcription (for Failed) in the preview panel.
4. Give it another minute. If it still comes back empty, just add a voice or text note directly describing the work.

The retry is safe to tap as often as you want — BidCraft dedupes against any existing transcript so you won't end up with duplicate notes.

Asbestos risk flagged

When BidCraft flags potential asbestos, that's based on visual cues — material patterns, age indicators, substrate type. It's conservative by design.

[!warning] Take asbestos flags seriously. Get a sample tested by a certified lab before any demo begins. Older floor tiles (especially 9x9 vinyl), mastic adhesives, and certain wall textures can contain asbestos. Removal requires licensed abatement and adds real cost to the job. Price accordingly — your proposal should include the abatement allowance or a clear contingency clause.

The flag means "investigate further" — not "definitely asbestos." But the cost of being wrong runs into five figures and serious legal exposure. Always test.

Condition scores look too high or too low

The 1-10 condition scale is calibrated, but it can occasionally read a surface differently than you would. If the AI scored a tub at 7 and you think it's a 3, you've got two options:

• Trust your eyes — you were on site, the AI saw two photos. Note the disagreement in your scope, but price for the condition you saw.
• Add more context — a voice note like "the tub is heavily chipped at the rim and stained, would not save it" gives the AI more to work with and the next analysis will reflect that.

The whole analysis just feels off

If multiple things look wrong — dimensions are weird, fixtures are missing, condition seems off — the usual cause is too few photos and no notes. The AI is working with thin data.

[!steps]

1. Take 3–5 more photos of the room from different angles.
2. Make sure at least two show opposite corners of the room.
3. Record one good voice note describing the work and the key measurements.
4. Re-analyze.

Almost every "the AI is bad" complaint resolves to "the AI had two phone photos and no context." Give it more to work with.

When to override vs. re-analyze

Next steps

If you've worked through this and the analysis still isn't right, the best path is to refine your inputs before going further. Head to Adding photos and videos for tips on capturing better source material, or Recording voice and text notes for the voice-note approach. If your fix worked, jump to Editing project dimensions to confirm the numbers locked in.