How Archer's machine learning improves your mapping over time

How it works

Archer's mapping AI has been pre-trained on over 130,000 unique financial line items. When you parse a new T12, it uses this training to predict where each line item should go in your Chart of Accounts.

When the prediction is wrong and you correct it — say you move "answering service" from Admin to Contract Services — that correction is stored. Next time Archer sees "answering service" on any T12 you parse, it knows to map it to Contract Services for your team.

This learning is specific to your team. If another Archer client maps "answering service" to Admin, that's their preference — it doesn't affect yours.

What trains the system

• Changing a T12 line item mapping in the dropdown (either on the web app or in Excel)
• Changing a rent roll charge code mapping
• Hitting Save to Archer after making changes in Excel

What does NOT train the system

• Using "Subtotal Ignore" — this is a one-time exclusion flag that the system deliberately doesn't learn from (you might ignore something on one deal but want it on the next)
• Changes to your own model tabs (pro forma, assumptions) — these are your analysis, not mapping preferences
• Changes to floor plan names or renovation status — these are property-specific, not patterns to learn

How fast does it learn?

Most clients see significant improvement after 5-10 deals. By that point, the system has seen enough of your corrections to predict 90%+ of line items correctly. Unusual or highly specific line items (like "convergent billing fee") may take a few more appearances before the system locks in your preference.

After your first few deals, you can also ask the Archer support team to accelerate the learning. They can take all of your corrections to date and match them against Archer's full database to pre-train your system on line items you haven't seen yet but likely will.