AI Product Manager

Sleek; Singapore; On-site; SGD 120,000 - 180,000; Full time

Job summary

Sleek is seeking an AI Product Manager to own the document-to-ledger pipeline end-to-end within SleekOne, spanning extraction, quality checking, matching and reconciliation. You will work daily with ML and backend engineers and practising accountants to drive measurable automation outcomes.

This is a hands-on, lean role with ownership over what gets built, how it is evaluated, and how it performs in production.

Qualifications

• Own the end-to-end document-to-ledger pipeline in SleekOne across extraction, quality checking, matching and reconciliation.
• Define evaluation criteria, datasets, and metric floors for model releases.
• Collaborate daily with ML engineers, backend engineers and practising accountants.

Responsibilities

• Define the roadmap for the four pipeline stages and automation impact.
• Write specifications with dataset, metric floors, and failure modes for changes.
• Build and maintain labelled evaluation sets with accounting team.

Skills

AI product management

Cross-functional leadership

ML collaboration

Data-driven decision making

Roadmapping

Evaluation criteria writing

Stakeholder management

Tools

SleekOne

Job description

Through proprietary software and AI, along with a focus on customer delight, Sleek makes the back-office easy for micro SMEs.

We give Entrepreneurs time back to focus on what they love doing - growing their business and being with customers. With a surging number of Entrepreneurs globally, we are innovating in a highly lucrative space.

We operate 3 business segments:

Corporate Secretary: Automating the company incorporation, secretarial, filing, Nominee Director, mailroom and immigration processes via custom online robots and SleekSign. We are the market leaders in Singapore with ~5% market share of all new business incorporations

Accounting & Bookkeeping: Redefining what it means to do Accounting, Bookkeeping, Tax and Payroll thanks to our proprietary SleekBooks ledger, AI tools and exceptional customer service

FinTech payments: Overcoming a key challenge for Entrepreneurs by offering digital banking services to new businesses

Sleek launched in 2017 and now has around 15,000 customers across our offices in Singapore, Hong Kong, Australia and the UK. We have around 600 staff with an intact startup mindset.

We have recently raised Series B financing off the back of >70% compound annual growth in Revenue over the last 5 years. Sleek has been recognised by The Financial Times, The Straits Times, Forbes and LinkedIn as one of the fastest growing companies in Asia.

Backed by world-class investors, we are on track to be one of the few cash flow positive, tech-enabled unicorns based out of Asia Pacific.

At Sleek, we are on a mission to streamline operations and elevate customer experience through intelligent automation and modern engineering.

About SleekOne and this role

SleekOne is Sleek’s AI-native accounting platform: a deterministic ledger wrapped in a layer of AI agents that collect, read, classify, match and post the financial documents of thousands of small businesses, with accountants reviewing exceptions rather than keying transactions. It is the single most consequential product bet in the company, owing to the fact that it replaces third-party ledger and extraction tooling across four markets and sets the operating leverage of our entire accounting business.

Mission: Own the document-to-ledger pipeline end to end, from the moment a receipt, invoice or bank statement enters the system to the moment a reconciled, correctly categorised transaction is posted, and raise its straight‑through rate while holding precision at the standard a regulated accounting firm requires.

We are looking for an AI Product Manager who treats model quality as a product problem rather than a research curiosity. This person owns the extraction, quality‑checking, matching and reconciliation agents at the heart of SleekOne, works daily with ML engineers, backend engineers and practising accountants, and is accountable for measurable automation outcomes rather than shipped features. This is a hands‑on role in a lean setup with real ownership over what gets built, how it is evaluated, and how it performs in production.

Primary accountability

You own the four stages of the pipeline that turns customer documents into a closed set of books, and the metrics that govern each stage.

• #1: Document extraction. Own the accuracy and coverage of structured data pulled from invoices, receipts, bank statements and supporting documents across formats, languages, currencies and scan quality in Singapore, Hong Kong, Australia and the UK. Define the extraction schema per document type, decide which fields must be correct for a transaction to be auto‑posted, and run the evaluation harness that benchmarks each model release against a labelled production set and a frontier baseline.
• #2: Quality checking. Own the classification and confidence layer that decides whether an extracted document is trusted, routed to an accountant, or sent back to the customer. Design the scoring and threshold logic so that the system auto‑publishes aggressively where it is right and declines where it is not, and build the feedback loop that converts accountant corrections into training signal and threshold adjustments.
• #3: Matching with bank statements. Own the logic that pairs documents to bank‑statement lines, including one-to-many and many-to-one matches, partial payments, foreign‑currency settlements, bank fees and timing differences. Define the match‑confidence model, the handling of unmatched items on both sides, and the customer and accountant experience for resolving them.
• #4: Reconciliation. Own straight‑through reconciliation to the ledger: categorisation, posting, and the month‑end close for each customer. Measure and move the reconciliation rate, precision of auto‑posted entries, rework rate and on‑time close, and own the exception queue that accountants work from when the system cannot decide.

Key outcomes

• Automation rate materially higher, precision intact: A rising share of customers whose documents flow from upload to posted, reconciled ledger entry with no human touch, with precision on auto‑posted and auto‑reconciled entries held above 99% and rework below 1%.
• An evaluation discipline the company trusts: Every extraction, classification and matching model release is benchmarked on a labelled production dataset against a defined floor and a frontier baseline before it ships; regressions are caught in evaluation, not by accountants.
• A closed learning loop from accountant corrections: Corrections made during review feed back into categorisation rules, confidence thresholds and training data on a defined cadence, so that the same error class is not corrected twice.
• Exception volume shrinking in both count and cost: Unmatched items, low‑confidence documents and failed reconciliations are triaged into named error classes with owners, and the accountant time per exception falls quarter on quarter.
• Engineering and accounting teams aligned on the problem: ML and backend engineers report clear problem statements, acceptance criteria expressed as metric thresholds, and well‑specified edge cases; accountants report that the system fails predictably and explains itself.

What you will do

• Define the roadmap for the extraction, quality, matching and reconciliation agents, sequenced by automation impact per unit of engineering effort and by the precision risk each change introduces.
• Write specifications that read as evaluation criteria: for each change, the dataset it will be judged on, the metric floor it must clear, and the failure modes it must not reintroduce.
• Build and maintain labelled evaluation sets with the accounting team, including