Who is ReconLayer for, what single problem does it solve, and why did reframing it as an internal layer make it buildable?

Discovery proved where a small builder could stand. Define turns that into the people it's for, the one problem worth solving, and the reframe that made it buildable.

How I grounded this

Rather than rely on one desk's view, I ran five structured interviews across four firms — peers from across the institutional buy-side, in equivalent middle-office roles, all anonymized. Cross-firm was deliberate: I wanted the pattern, not one shop's habits — which is also why the personas below are composites, not portraits of individuals.

Two people, two stakes

Sarah — the tactical operator. Senior ops analyst, ~4 years in, reconciling 150+ positions against a hard morning cutoff. Wants speed and error-proofing on repetitive work. Friction: mismatched file formats, manual transcription of trade records into dispute emails, and hunting phantom one-sided trades.

David — the strategic buyer. Risk-conscious COO, ex-trader, owns the budget, the prime-broker relationships, and the ODD story he tells allocators. Wants auditability, security, and cost control. Friction: enterprise vendors that ingest data to their own cloud, and the key-person risk baked into one analyst's undocumented macros.

The problem statement

A middle-office analyst on a lean, budget-constrained desk must reconcile internal and external books and resolve the breaks before a hard cutoff — but does it by hand in a spreadsheet, because every tool that automates the work is priced, scoped, and sold for large institutions. The cost of that manual gap is measured in operational errors, not just minutes.

What the research changed — the dissenting views

The interviews didn't just confirm the plan. The sharpest input pushed against it, and that resistance shaped the design more than the agreement did:

• "Don't hide the logic in a script I can't audit." The manager's fear wasn't AI — it was moving matching logic out of transparent Excel cells into an opaque sidebar. This forced a visible, rule-based audit trail, not just a list of breaks.
• "A Google Sheet reads as a retail tool to an ODD auditor." The COO's concern shaped the target security posture: matching stays local on the firm's own machines, and the only data that leaves — minimal break summaries for the draft — goes to a zero-retention endpoint. (The prototype uses a personal account and the free Gemini API on synthetic data; the enterprise posture is future state, not built.)
• "If the AI swaps 'we owe you' for 'you owe us,' that's millions." The derivatives analyst's distrust of AI-authored correspondence drew the hardest line in the product: the AI drafts only, never sends, and a prominent copy-to-clipboard gate keeps the human in control.
• "If you build it and then leave, who maintains it?" The key-person-developer risk argued for standardized, documented, boundaried code — not another home-grown black box that becomes the next fragile macro.

That last cluster is the whole point: the tool's guardrails exist because users said, plainly, that they wouldn't trust it otherwise.

The reframe that unlocked it