Skip to main content
Th1nkDev
AjustOne — 1
All Projects
LivelabsCollaborations2025AjustOne

AjustOne

AI-assisted bank reconciliation solution for the CEMAC region: automated transaction matching, exception management, and reporting.

1 / 9

Project Overview

Your bank applies a fee schedule. No one checks that it honors it.

Every company signs a set of banking conditions: a fee grid, tolerances, value-date commitments, a VAT regime. Then the contract goes back in a drawer, and the statements keep arriving, month after month, with no one equipped to hold one against the other. Reconciliation happens however it can — in a spreadsheet, by hand, on amounts alone — and what has merely been counted gets declared "balanced." The gaps that survive that exercise are not details: an off-grid fee, a miscalculated VAT, a value date beyond SLA are net banking income leaks that repeat on every single transaction and that nobody ever sees go by.

AjustOne is a multi-tenant SaaS platform that turns this control into an engine. It confronts internal ledger entries with bank statement movements over a date range, and produces not a binary verdict but an actionable typology: unsettled instruction, unexpected entry, amount mismatch, value date out of SLA, fees above or below the schedule, incorrect VAT, probable duplicate. Eight families of exceptions — R1 through R8 — that can be assigned, investigated, disputed with the bank, and closed.

What makes it distinctive is what drives those rules. The thresholds are not arbitrary settings: they are derived from the account's actual banking conditions — amount tolerances, SLA windows, fee grid, VAT regime — with the PDF contract parsed at import and bound to the account. The negotiated schedule stops being a dead document and becomes the executable reference against which every statement line is judged.

A detected gap is not an endpoint but the start of a lifecycle. Each exception is qualified (risk score, recurrence, presumed cause), assigned to a role, investigated in a side-by-side ledger/statement view with its supporting evidence, remediated — internal correction or bank claim — tracked under SLA with escalation, then closed under maker-checker: whoever proposes is never whoever approves. A post-mortem traces root causes, so the anomaly stops recurring.

What leaves the platform is built to stand up to scrutiny. Reconciliation certificates, net banking income health, exception aging pyramid, auto-match rate, resolution times — all backed by a chained, immutable audit trail: who, what, when, why. Every import, every manual match, every sensitive unmatch, every approval leaves a record neither the internal auditor nor the statutory auditor has to reconstruct.

The product is designed for the reality of the markets it serves, the CEMAC zone in particular: where the statement arrives as a PDF rather than through an API, where MT940/942 and camt.053/054 channels coexist with in-house exports and spreadsheets, and where banking reference data is anything but standardized. That heterogeneity is absorbed at the entry point, not pushed onto the user.

The result: a treasury function that stops absorbing its statements and starts auditing them — and that can, for the first time, prove what it claims.

Project Details
Year2025
Categorylabs
Project typeCollaborations
ClientAjustOne
StatusLive
Stack15 technologies
Project Narrative
01

The Problem

01 / 03

Bank reconciliation remains largely manual across the CEMAC region. A monthly close typically requires several person-days of repetitive work, where mistakes are costly: in reconciliation, a false positive is far more damaging than a false negative, as it incorrectly validates a match between transactions.

02

Our Solution

02 / 03

The reconciliation engine combines deterministic rules with an AI-assisted layer for approximate transaction description matching, using a carefully calibrated confidence threshold that prioritizes escalation to a human operator over uncertain automatic validation. Every decision is fully logged to ensure complete auditability.

03

Impact

03 / 03

Target metrics: percentage of transactions reconciled automatically · hours saved per financial close · percentage of unmatched items resolved without manual intervention · false positive rate.

Tech Stack

Technologies we used

15 technologies carefully chosen for this project.

Nuxt Vue 3 TypeScript Node.js PostgreSQL IA appliquée Tairo Python Data Engineering LLM MCP PDF Manipulation Advanced IA Machine Learning Big DATA

Our Process

How we built it,
step by step.

Every project at Th1nkDev follows the same structured methodology — from discovery to launch and beyond, with full transparency at every step.

Explore our services
01

Discover

Deep-dive into your context, goals and constraints.

02

Design

Architecture, UX flows, prototypes — before code.

03

Build

Clean, tested, documented production code.

04

Launch

Zero-surprise go-live and ongoing support.

Ready to get started?

Your next digital product
Start here.

Let's talk about your project. No commitment required — just an honest conversation about what's possible.