ReconLayer

Simplifying reconciliation and exception workflows for lean middle-office teams

I built ReconLayer to explore a problem I knew from years in trading operations: small teams still spend part of every morning comparing internal positions against broker files, finding breaks, and writing the same counterparty emails by hand.

Role

Solo Product Designer

Scope

Strategy · Research · Product Design · Build · Testing

Domain

Finance

Status

0 → 1

Duration

12 weeks

RECONLAYER SITS INSIDE THE WORKBOOK

The problem

A manual workflow with very little room for error

On a lean finance-operations desk, reconciliation often still starts with two files and a spreadsheet.


The analyst compares the firm’s internal book against a broker file, investigates anything that doesn’t match, and resolves the exceptions before a settlement or margin cutoff.


The problem isn’t just that this takes time. A missed trade or incorrectly classified break can affect a margin call or NAV — and the analyst is accountable for the decision.

MANUAL RECONCILIATION WORKBOOK - BEFORE RECONLAYER

Four decisions that shaped the product

01 Adoption

Layer on top. Don't rip and replace.

Research showed that enterprise vendors already cover most core reconciliation capabilities. A smaller standalone product would add exactly what lean teams don’t need: another system, login, and workflow to maintain.

I ruled out: a standalone reconciliation product, because adoption friction was bigger than the product opportunity.

02 System Boundary

Rules find the break. AI explains it.

Financial truth stays deterministic. Generation only begins after every break has been identified and classified.

I ruled out: LLM-based classification. The AI can help write the message; it doesn't decide what the numbers mean.

MODEL NEVER DECIDES WHETHER A BREAK EXISTS

03 Communication

Templates vs. AI drafting

Templates broke down as the mix of disputes changed. A constrained draft could adapt without changing the classification.

I ruled out: a large template library. Drafting was the one part of the workflow where generation reduced work without giving the model decision authority.

04 Domain Insight

A missing value isn’t a zero.

The most important exception was also the easiest to hide. Without a value on both sides, there is no variance to rank.

I ruled out: a single variance-ranked list. The missing side is itself the signal.

RECONLAYER PINS CLASSIFIED BREAKS

From two files to a worklist

Designed around the analyst’s next decision.

ReconLayer runs inside the analyst’s spreadsheet. It compares the internal and broker books, filters out differences below the tolerance, and separates the remaining exceptions into value breaks and population breaks.

What testing changed

I wanted to find the point where the user questioned the validity of the extension. That surfaced three things:

01 Terminology

“Value break” and “population break” are normal middle-office terms, but not everyone interpreted them correctly on first read.

Next change: clearer labels and inline explanations.

02 Trust

Even the testers who were comfortable with the AI generated draft said they would check the figures or broker format first.

"I'd at least still double-check the figures and breaks make sense."

Operations Analyst, Usability-Test Participant

03 Rule Logic

Once testers understood that the AI wasn't doing the classification, their next question became: How do I know the reconciliation logic itself is right?

Next change: show what items were flagged and why; bring to Compliance

What I'd change next

01 Make classification easier

Testing showed that the break labels were the first comprehension problem. I would clarify the terminology and expose more of the reasoning behind each classification.

02 Add audit view

The next trust problem isn't the AI — it's the rules engine. Analysts need to be able to inspect why a break was flagged and verify the calculation themselves.

03 Normalize in environment

That means flexible broker-file headers, broker-specific email formatting, and eventually a compatible extension across any workbook.

Go Deeper

ReconLayer is a Google Sheets add-on, so it runs inside your own Google account rather than as a public demo. Sign-in to launch the sidebar.