The evidence
How it performs
01 / Time back
More questions answered.
Less time waiting.
Screen a new submission. Add a question to your archive. Get through the work at the pace you need.
faster than Jev
depending on your workload
Short texts and straightforward questions move fastest. Longer documents and more complex decisions take more time.
See the accuracy behind it ↓Your workload
How much time could you save?
Illustration at 50× speed against the Jev comparison baseline. Your actual time depends on the workload.
Your pace. Your review rules.
Choose the effort.
Know when to trust the answer.
Low for speed. Medium to balance speed and accuracy. High when each decision deserves more care.
Every answer includes a reliability score. Move confident answers forward and keep uncertain decisions for review.
Accuracy comparisons focus on confident answers—the ones you’re more likely to act on. We set aside low-confidence answers from each service, using high effort for flexiclass. In the accuracy summaries, 100% means matching Jev’s accuracy.
02 / Pick from a list
The right category.
Ready for the next step.
Jev = 100%
Sort a claim, route a submission or label a document using your own categories.
Does confidence match the results?
03 / Yes or no
One condition.
A clear decision.
Jev = 100%
Does this document meet the condition?
Does the claim mention a prior injury? Is the submission in scope? Does the report describe a total loss?
Reliability helps you decide which answers can move forward. On the diagonal, confidence matches the results.
Does confidence match the results?
04 / Rate on a scale
More than yes or no.
How much? How close?
Jev = 100%
Rate sentiment or compare meaning. These plots show where the answer lands: on the diagonal, the average rating is right on target.
Results by task
| Task | Questions | Flexiclass · high | Jev |
|---|
Images · Beta
Photos and scans.
Measured on real questions.
Flexiclass works with text and images. Sort documents, recognise products and assess damage photos using your own categories.
Jev currently accepts text only. Image understanding is an additional Flexiclass capability, so these results show absolute accuracy, without a Jev comparison. Jev’s supported inputs.
Try an image →Image accuracy
Correct answers across 7,200 image questions.
9 tasks covering documents, products, buildings, diagrams and damage photos. Image results are separate from the text comparison.
Results by task
| Task | Questions | Flexiclass accuracy |
|---|---|---|
| Building floor counts | 1,042 | 43.0% |
| Building use | 374 | 40.9% |
| Car damage photos | 1,080 | 81.7% |
| Product categories | 2,000 | 73.7% |
| Hurricane damage from aerial images | 400 | 59.8% |
| Scanned document types | 480 | 65.8% |
| Questions about diagrams | 324 | 87.3% |
| UK house types | 300 | 64.7% |
| Wildfire damage to buildings | 1,200 | 58.8% |
| All image tasks | 7,200 | 65.2% |
Evidence · Beta
Performance comparison
coming soon.
Evidence extraction and verification are available to try.
Try evidence →Retrieval · Beta
Performance comparison
coming soon.
Ranking and retrieval-assisted classification are available to try.
Explore retrieval →Evaluate your own task
Try flexiclass on your own work.
Bring a few real examples. Write the questions your team actually needs answered. See what you can do with the results.
Try your text or images Integration guide →