Case studies · Singapore SMEs
Real Singapore service businesses I’ve built custom systems for. Anonymous on request, but every number, every workflow, every artefact below is real. Pick the one that looks like yours.
Dozens of student applications per consultant, tracked across spreadsheets, shared inboxes, and PDF folders. I unified the work into one platform. The 3-hour weekly sync between consultants and the sales/CS manager is gone.
Multi-step compliance routing was held together by email and chasing. I rebuilt the chain as a versioned workflow with auto-escalation. Errors collapsed. The compliance officer stopped being a router.
Hundreds of bank statements per quarter, manually consolidated by analysts. I built a pipeline that ingests, reconciles, flags anomalies, and outputs a portfolio view by Monday morning.
Operational data from multiple retailer systems was reconciled by hand each week. I unified it into one BI surface with an AI agent that surfaces what changed and why. No more pivot-table archaeology.
Senior placements lived in inboxes, calendars, and partners’ heads. I built a pipeline view with movement tracking, follow-up alerts, and signal scoring. The Monday catch-up became a 5-minute scan.
Case study · Education consultingUniversity admissionsSingapore
A Singapore education consulting firm was tracking dozens of university applications per consultant across spreadsheets, shared inboxes, and PDF folders. I unified the work into one platform, in weeks, not months. The 3-hour weekly sync between consultants and the sales/CS manager is gone.
Before
Each consultant tracked their cohort in a spreadsheet they’d built themselves. Documents lived in shared drives nobody trusted. Status updates required messaging the consultant directly, which meant the sales and customer service manager spent half her week chasing answers parents had already asked her about.
New students entered the system through a Google Form that fed an email inbox. The inbox had no owner. Once a week, the team blocked out 3 hours to talk through every student in the cohort, line by line, so the manager could brief parents and the founder could see where revenue stood.
That meeting was the only time anyone had a complete picture. By Wednesday, the picture was already wrong.
“By Tuesday afternoon I knew where every student was. I just couldn’t tell anyone else without a 3-hour meeting.”
What I built
From kick-off to full deployment in weeks, not months. A custom build delivered in phases. Every component below was scoped in writing within 48 hours and locked before the build began.
Founders see the full cohort, revenue, capacity, and at-risk students. Consultants see only their assigned students. Sales/CS sees status across every consultant. One source of truth, three different views.
Every student tracked like a deal: stage, milestone, assigned consultant, parent contact, target universities, deadlines. Progress bars, status pills, and overdue flags surface what needs attention without being asked.
Each student logs in to upload essays, transcripts, recommendation drafts, and supporting documents. Submissions land directly in the tracker. Consultants stop chasing files; students stop emailing PDFs.
“Which students are at risk this week?” “Who hasn’t submitted their personal statement?” “What’s our enrollment forecast for spring intake?” The system answers in seconds, off live data.
Missing documents trigger reminders to the student, then escalate to the consultant, then to the manager, based on rules the firm set in scoping. The manager stops being the human escalation engine.
Timeline
Scope document delivered. Four functions, two follow-on phases, role-based access model, data sources, escalation rules, success metrics.
Working MVP. Operational dashboard live with the firm’s actual cohort data. Founder, manager, and one consultant testing in their real workflow.
Full build deployed. All four functions live, student portal opened, escalation centre active. The 3-hour weekly sync was cancelled the same week.
Follow-on phases. Two further phases added: parent reporting view and a counsellor handoff workflow. Each followed the same 48hr / day-7 / launch structure.
What this means
Every recurring meeting whose only purpose is “everyone update everyone” is a symptom of missing visibility. The fix isn’t a better meeting. It’s a system that makes the meeting unnecessary.
For this firm, that meant a custom build, no new admin hire, and a manager who finally got her Wednesday afternoons back. The work didn’t get smaller. The work that didn’t need to be done in a meeting stopped being done in a meeting.
Case study · Financial servicesCompliance workflowSingapore
A 22-person Singapore finance firm was running compliance approvals through email threads and spreadsheets. Version errors. Missed steps. Every escalation traced manually. I rebuilt the workflow with rule-based routing, timestamped steps, and a full audit trail that writes itself.
Before
The firm’s compliance team processed 30 to 50 trade authorisation requests a week. Each request started as a forwarded email, picked up a spreadsheet attachment somewhere along the chain, and bounced through compliance review, partner sign-off, and audit-log entry, manually, person by person, in whichever email thread someone happened to start.
Roughly 13% of submissions had a version-control error: a stale spreadsheet attached, an out-of-date compliance note copied in, or a missed sign-off step that surfaced only when an auditor flagged it weeks later. Every one of those errors meant a re-do. Every re-do meant the analyst, the compliance officer, and the senior partner all touched the request again.
The firm wasn’t short on talent. They were short on a system that knew what step came next without anyone having to remember.
“We didn’t have a process problem. We had an inbox problem pretending to be a process.”
What I built
A custom build, then two follow-on phases. The workflow stopped living in email and started living in a system that wrote its own audit trail.
Analysts file requests through a single form. Required fields enforce what used to be a memo’s worth of context. No more attaching last week’s spreadsheet by mistake.
Each request routes to the right reviewer based on type, threshold, and portfolio. The system knows the rules; people stop re-deciding them every time.
Sign-off pending past the SLA escalates to a backup. No one sits on a request because they’re on leave or out of office.
Every submission, review, sign-off, and escalation timestamps itself. The audit trail is a byproduct of the work, not a manual reconstruction afterward.
Live view of every request in flight. Stuck approvals, breached SLAs, and audit anomalies surface without anyone running a query.
Timeline
Scope document delivered. Routing rules captured, escalation thresholds confirmed, data sources mapped, audit-log format agreed.
Working MVP. Intake form live, three of five routing rules wired, dashboard reading from real submissions. Compliance lead testing in parallel with the old email flow.
Full build deployed. All routing rules live. Escalation active. Audit log generating automatically. Old email-and-spreadsheet flow retired.
Follow-on phases. Two further phases added: a quarterly compliance report generator and an external-auditor portal that reads the audit log directly.
What this means
Most “process problems” in service businesses are actually missing systems. The work itself is fine. The way the work is captured, routed, and recorded is held together by people remembering things.
For this firm, that meant a fast custom build, a 60% drop in coordination work, and an audit trail that’s never going to be the reason an audit fails. The team didn’t get smaller. The work that didn’t need a person stopped needing a person.
Case study · Family officeFinancial servicesSingapore
A Singapore family office held positions across multiple private banks, with each bank’s statement arriving in a different PDF format. The director needed a consolidated view; the team built it manually every week. I built an OCR + consolidation + risk-flagging system, fast. The director sees everything, live, without asking anyone.
Before
The family office held positions across four private banks. Each bank produced statements in a different PDF format, on a different cadence. The director wanted one consolidated view of the portfolio: total exposure, currency mix, asset class allocation, internal dependencies, and a flag on anything unusual.
The team built that view manually. Every week. Two analysts spent the better part of a day each pulling numbers, re-keying them into a master spreadsheet, and reconciling the inevitable mismatches. By the time the director saw the spreadsheet on Wednesday, it represented Monday’s reality.
For a family office moving real money in real time, “Monday’s reality on Wednesday” wasn’t visibility. It was a delay disguised as a report.
“I was making decisions on data I knew was a few days stale. I just had no way to fix that without hiring two more analysts.”
What I built
Fast, from kick-off to full deployment. A custom build with one follow-on phase. The OCR was what opened it up; the rest fell into place once the data was structured.
Statements from each bank parsed into a normalised position schema. Format differences absorbed; the director’s view doesn’t care which bank produced which PDF.
Cross-bank duplicates collapsed. Currency conversion applied at consistent reference rates. Asset class taxonomy reconciled across providers.
The system surfaces internal dependencies (positions correlated across banks), concentration risk, and unusual movement against a rolling baseline. The director gets flags, not a sea of numbers.
One view: total exposure, currency mix, allocation, top-N positions, recent movement. Drill in by bank, by asset class, by date. No spreadsheet involved.
Where positions can be consolidated for fee or risk reasons, the system surfaces suggestions with the rationale visible. The director decides; the system shows the work.
Timeline
Family office scope was tighter because the data sources were narrower. I compressed the schedule accordingly.
Scope document delivered. OCR template per bank confirmed, consolidation rules agreed, risk thresholds set, dashboard layout sketched.
Working MVP. Two of four banks parsing correctly. Consolidation logic running. Director reviewing real numbers in the dashboard.
Full build deployed. All four banks parsing. Consolidation layer reconciling. Risk flags live. Manual weekly compilation retired.
AI consolidation suggestions added as a follow-on phase once the director had used the base view for a while.
What this means
Manual compilation is fine when the data fits in one head. The moment it lives across multiple sources or formats, weekly compilation is a delay you’ve decided to accept. Most of the time, it’s a delay you can stop accepting in weeks, not months.
Case study · 3PL logisticsLast-mile deliverySingapore
A Singapore last-mile 3PL operator was running performance analysis across four major retailer accounts, each delivering data in a different Excel format on a different cadence. The GM wanted year-on-year, store-by-store, SKU-level views. The team produced them. Eventually. I rebuilt the data pipeline and dashboard. Now the GM asks; the system answers in seconds.
Before
The GM ran a 3PL operation moving parcels across four major retailer accounts. Each retailer pushed performance data (delivery rates, exceptions, SLA breaches, return volumes, SKU mix) in a different Excel template on a different schedule. One was daily. One was weekly. One was monthly. One was “when they remember.”
To answer something like “How are we doing on Retailer A’s frozen-goods SKUs in the West region year-on-year?” the team had to compile from four sources, normalise the categories, line up the dates, and rebuild the pivot. Two days, sometimes three. By the time the answer landed, the GM had already moved on.
The data was there. It was the compilation that was killing visibility.
“By the time I had the answer, the question had aged out.”
What I built
One custom build, then two follow-on phases. The AI agent came in the second phase, added once the data layer was solid.
Each retailer's Excel template parsed automatically. Format differences absorbed at ingestion. SKU and category taxonomy reconciled across feeds.
Year-on-year, month-on-month, store-vs-store, SKU-level. The GM filters by any dimension and gets answers immediately.
Late deliveries, missed pickups, return spikes flagged automatically. The ops team sees the same data the GM sees, scoped to their region.
Trained on 3PL logistics patterns. Surfaces recommendations from live data: where to rebalance capacity, which SKU mixes are degrading, which retailers are trending unfavourably. The GM gets prompts, not just numbers.
Each retailer's performance against their SLA exposed back to them via a controlled view. Disputes drop when both sides see the same numbers.
Timeline
Scope document delivered. Retailer feed templates documented, taxonomy reconciliation rules agreed, dashboard dimensions confirmed, SLA thresholds set.
Working MVP. First two retailer feeds parsing, dashboard rendering live data. GM testing real questions in real workflow.
Full build deployed. All four feeds ingesting on schedule. Dashboard live across the team. Excel compilation retired.
AI analysis agent added in a follow-on phase. Trained on the firm’s actual patterns, not a generic LLM dropped in front of a database.
What this means
Most operations teams accept multi-source manual compilation as the cost of doing business with multiple partners. It isn’t. The compilation is a piece of software you haven’t built yet. The longer you live with it, the more decisions you make on stale answers.
Case study · Executive recruitmentSenior placementsSingapore
A Singapore senior-placements firm was doing manual market mapping for every search: eight hours of LinkedIn, references, and judgment per candidate, much of it repeating work done six months earlier on a similar search. I built an automated sourcing pipeline with intelligent filtering. The candidate pool grew 2 to 3×. Past hiring data is now searchable and compounds with every new search.
Before
The firm placed senior executives (VPs, country heads, C-suite) across financial services and professional services in the Singapore region. Each search began with a blank document and eight hours of manual market mapping per consultant: LinkedIn searches, talent stack reviews, reference cross-checks, judgment calls about who to surface.
The team was good at this. The problem was that the work didn’t compound. The market map for a CFO search in financial services in 2024 didn’t make the next CFO search easier in 2025. The notes lived in a Word file in someone’s Drive folder. The candidates lived in a CRM nobody loved. The judgment calls (the actual value the firm sold) lived nowhere.
So the firm sold the same eight hours of judgment over and over. The candidates seen in past searches drifted. The shortlist for a similar role next year wasn’t materially better than the one a year before.
“We were great at executive search. We were terrible at remembering we were great at it.”
What I built
A custom build plus two follow-on phases. The data sources and filtering logic needed careful scoping with the partners.
Given a search brief, the system pulls candidates from the firm’s data sources, filters by role / industry / seniority / geography, and produces an initial long-list within hours instead of days.
Beyond keyword matches: trajectory, tenure patterns, industry adjacency, role progression. The system narrows hundreds of candidates to dozens worth a consultant’s time.
Every search, every shortlist, every consultant note retained and searchable. The next CFO search inherits from the last CFO search. The judgment calls finally compound.
One view per active search: the long-list, the shortlist, the rejected (with reasons), the active conversations. Consultants stop juggling Word docs and Drive folders.
Anonymised view for the hiring company: where the search is, how the long-list is shaping up, what’s next. Status updates stop being a phone call.
Timeline
Scope document delivered. Filtering rules locked, data retention model agreed, consultant workflow mapped, client view defined.
Working MVP. Sourcing pipeline pulling candidates for one live search. Consultants reviewing real long-lists in the new workspace.
Core build deployed. Sourcing, filtering, workspace, retention layer all live. A follow-on phase scoped for the data-history backfill (turning the firm’s Word docs into searchable structured records).
Full system live. Past hiring history backfilled and searchable. The next search starts from the firm’s accumulated knowledge, not a blank page.
What this means
Every professional services firm sells expertise. Most of them lose 80% of that expertise the moment a project ends, locked in someone’s notes, someone’s head, or a Word doc nobody opens again. Building a system where the judgment compounds is the difference between getting better as a firm and just getting older.
Get started
No pitch, no proposal. Book a call, or tell me the one thing eating your week and I'll say if it's worth one. I read every one myself, and take three projects a month.
ben@3nm.io · Singapore · No pitch, no proposal.
