AI Memory

Prosper learns your business. Forever.

Every correction teaches Prosper a pattern. Every pattern makes future categorization faster and more accurate. The longer you use Prosper, the more it knows about your specific business.

This is not a static rule system. It is a learning system that compounds value over time.

How AI Memory works

Step 1

You correct a transaction

"AMZN*2K7B3X" is Office Supplies, not inventory.

Step 2

Prosper remembers the pattern

Learns that AMZN* transactions for your business are office supplies.

Step 3

Future transactions auto-categorize

Next time AMZN appears, Prosper handles it automatically.

150+

Avg patterns learned

per active workspace

~3 mo

Time to 90% accuracy

with regular usage

87%

Correction recall

semantic match rate

Why this matters

Compounding accuracy

Month 1: 70% auto-categorized. Month 6: 90%+. Your patterns compound into a custom model.

Time savings that grow

Every correction saves you time on all future similar transactions. Early investment pays dividends.

Your knowledge, retained

Vendor mappings, category preferences, business context. All stored and used to improve suggestions.

Explainable, editable, deletable

See exactly what Prosper has learned. Edit patterns. Delete what does not work. You are in control.

Comparison

AI Memory vs. Static Rules

FeatureProsper AI MemoryQBO / Xero Rules
Learning methodSemantic AI with embeddingsExact-text rules
Pattern recognitionFuzzy matching (AMZN* → Amazon)Exact match only
CorrectionsLearned automaticallyManual rule creation
Cross-vendor patternsAI infers similar merchantsOne rule per merchant
Business contextUses your memories and historyNo context awareness
Improvement over timeCompounding accuracyStatic rules

Your learned patterns are an asset

After 6 months, Prosper knows your vendors, your category preferences, your edge cases. That knowledge took time to build. If you switch to another tool, you start from zero.

This is not lock-in. It is value. We show you exactly what Prosper has learned, and you can export your patterns anytime. But the intelligence is real, and it is yours.

How it works under the hood

  • Corrections are stored with semantic embeddings
  • Similar transactions recall past corrections (87% match rate)
  • LLM uses corrections as few-shot examples in prompts
  • High-confidence matches auto-confirm (configurable threshold)

What you can do

  • View all learned patterns in Settings → AI
  • Edit category assignments for any pattern
  • Delete patterns that are not working
  • Reset and re-learn from existing transactions

Ready to build your AI memory?

Start categorizing. Every correction makes Prosper smarter for your business.

Get started