Aicoursesforsg

Catalogue

Six programmes, each built around one task

Choose the programme that matches the work you already do. Foundations is the right first class if your team has not been trained on assistants. The other five programmes assume you can log in, paste a prompt, and read the output with a critical eye. Fees below are per person for an open class.

Contact and enrolment See the outcome records

How to choose

Start from the task you already perform. If the weekly load is an overflowing shared inbox, take Document and Inbox Automation. If the monthly load is a forecast that still takes a full afternoon, take Spreadsheet to Forecast. If the load is a board pack assembled from five exports, take Reporting and Drafting with Verification.

Team leads who already use an assistant and now need a shared pattern library should take Prompt Patterns for Team Leads. Owners of small firms who need a written rule for what staff may paste into a tool should take Safe Use and Internal Policy for SMEs. Anyone who has not yet been shown how models fail on confidential fields should take Foundations first; the later programmes move faster.

Group workshop

AI at Work: Foundations

Who it is for

Office staff, team coordinators and SME owners who write, check or file information and have not had structured training on commercial assistants. No programming background is required. You need a laptop, a live login to an assistant your workplace already licences, and a willingness to work on anonymised sample documents for two days.

What you will build

  • A personal operating note. Where the assistant sits in your week, which files never go in, and who you escalate to when the output looks fluent and wrong.
  • A fact-check pass. A short sequence for claims, dates, figures and names, with a forced pause before anything is sent.
  • A confidential-field map. A one-page list of data types in your role that stay out of the prompt, even when the vendor says the workspace is private.
  • A failure log. Four typical failure modes (invented citations, stale policy, over-confident totals, leaked internal phrasing) with an example from the sample pack.
  • A 30-minute practice drill. Timed on a 6-item sample so you can see what “checked” actually takes.

What we measure

Elapsed time to complete the 6-item sample, and the count of unchecked errors against an answer key. Setup, login and instructor explanation stay off the clock. The published site figures for inbox, forecast and reporting come from the specialist programmes below; Foundations uses its own classroom sample and is not mixed into those three records.

Hours: 14 h (2 days)
Format: Open class, Fusionopolis
Seats: 12
Fee: S$690
Prerequisites: None beyond a live assistant login and English working proficiency

Document and Inbox Automation

Who it is for

Staff who live in a shared inbox: customer operations, internal service desks, facilities, HR administration. You already reply to similar messages every day. You want a repeatable triage, a small set of reply templates, and a written rule for when a human must take the thread.

What you will build

  • A label set. Between five and eight labels that match how your desk actually routes work, tested on a 20-ticket anonymised sample.
  • A triage prompt file. The assistant proposes a label, a short rationale, and a draft reply. You keep the right to override every field.
  • Reply templates. Three to five shells with blanks the assistant must not fill from memory.
  • An escalation rule. The conditions that force a hand-off (legal threat, payment dispute, missing identity, anything the sample marks as out of scope).
  • A measurement sheet. Handling time and mislabel count, ready to reuse on a later sample at your desk.

What we measure

Median handling time per ticket and mislabel rate against a prepared key. On the March 2026 open class the median handling time fell 57% and the mislabel rate moved from 9.0% to 3.5% on 20 anonymised tickets. That record is written up in full on the inbox case page. A remote clinic within 30 days is included so you can bring questions from live mail after the class.

Hours: 7 h + follow-up clinic
Format: 1 day in Fusionopolis, clinic remote
Seats: 10
Fee: S$520
Prerequisites: Daily inbox work; Foundations or equivalent familiarity

Spreadsheet to Forecast

Who it is for

Operations, merchandising and finance staff who already maintain a monthly series in a spreadsheet and send a short commentary to a manager. You do not need a statistics degree. You do need to be comfortable with filters, simple formulae, and charting.

What you will build

  • A cleaning checklist. Missing weeks, duplicate SKUs, unit mixes, and calendar effects that quietly wreck a naïve model.
  • A basic forecast. A transparent method (moving average or simple seasonal naïve, chosen against the sample) that a colleague can audit without a specialist tool.
  • Sanity checks. Bounds, last-year comparison, and a forced comment when the assistant proposes a number outside the historical range.
  • A one-page narrative. What changed, what the model assumed, and what a reader should not conclude.
  • An error sheet. Mean absolute percentage error on a held-out quarter, computed the same way at baseline and after the rebuild.

What we measure

Forecast error on the held-out quarter, plus the time to produce the pack. On the May 2026 evening cohort, mean absolute percentage error moved from 18.4% to 11.2% on anonymised demo series. The full protocol, including what we refused to count, is on the forecast case page.

Hours: 9 h (3 evenings)
Format: Hybrid evenings
Seats: 12
Fee: S$780
Prerequisites: Working spreadsheet skills; access to the licensed assistant

Prompt Patterns for Team Leads

Who it is for

People who review other people’s assistant output: team leads in operations, customer teams, and small professional services firms. You already know the tool well enough to be dangerous. You need a shared library so that every member of the team is not inventing a new prompt on Monday morning.

What you will build

  • A pattern catalogue. Six to eight reusable shapes (extract, classify, compare, redraft, checklist, escalate) with a one-line purpose each.
  • A gold-standard example set. Three input–output pairs per pattern, taken from anonymised work, that the assistant must treat as the quality bar.
  • A review cycle. A weekly 25-minute slot in which the lead samples five outputs, scores them against the examples, and retires a pattern that has drifted.
  • A change log. Who edited a pattern, when, and why, so that a new joiner can see the current file.
  • A handover note for the rest of the team, written in plain English, with the files attached.

What we measure

Time for a lead to review five sample outputs, and the count of missed defects against a key. We do not publish a site-wide average for this programme; sample sizes so far are smaller than the three featured records. You will still leave with your own before-and-after pair from the day.

Hours: 7 h
Format: Open class, Fusionopolis
Seats: 10
Fee: S$560
Prerequisites: You already review team work; Foundations or equivalent

Reporting and Drafting with Verification

Who it is for

Staff who assemble a monthly or quarterly pack from exports, emails and a previous file: secretariat, operations, finance support, and SME owners. You are judged on whether a director can trust the page, which is a slower problem than whether a paragraph reads fluently.

What you will build

  • A draft skeleton. Headings that match the pack you actually send, with placeholders the assistant must not invent.
  • A source-marking rule. Every figure carries a pointer to the export or cell that produced it.
  • A fact-check checklist. Period, unit, comparison base, and a forced “unknown” when the source is missing.
  • A redraft pass. The assistant shortens prose after the numbers are locked, never before.
  • A second-marker sheet. A colleague (in class, another learner) counts gaps the way we do in the published record.

What we measure

Median time to a complete first draft with source marks, and the count of factual gaps found by a second marker. On the June 2026 class the median moved from 74 minutes to 31 minutes on an anonymised pack. The write-up is on the reporting case page.

Hours: 7 h
Format: Open class, Fusionopolis
Seats: 12
Fee: S$540
Prerequisites: You already draft or compile a recurring pack

Safe Use and Internal Policy for SMEs

Who it is for

Owners and operations managers of small firms who have allowed staff to use a commercial assistant and now need a written rule. The session covers general principles of handling personal data and vendor questions. Instructors do not provide legal advice, and the draft you leave with is a training artefact.

What you will build

  • An allowed / forbidden paste list. Personal identifiers, financial credentials, health information, and unpublished commercial terms stay out; operational how-to text can go in once stripped.
  • A vendor question sheet. Retention, training on customer content, residency of logs, and who can see a workspace. You take this to your supplier; we do not answer it for them.
  • A draft internal policy. Two to three pages of general principles, with blanks for your tools and roles, written so a non-lawyer can apply it on a Tuesday afternoon.
  • A staff briefing script. Ten minutes, to be read in a team meeting, covering examples from the sample pack.

What we measure

Time to classify 12 sample pastes as allowed, stripped, or forbidden, and the error count against a key. We do not treat this as a substitute for advice from a qualified professional on your specific processing. The 3.5-hour format is deliberate: it is long enough to produce a draft and short enough that an owner can attend without a two-day absence.

Hours: 3.5 h
Format: Open class, Fusionopolis
Seats: 12
Fee: S$320
Prerequisites: Authority to set a team rule, or a mandate to draft one

Delivery formats

Open classes run in our Fusionopolis training room at 1 Fusionopolis Place, #08-11, Galaxis. Tables are set for laptops. Class size is capped at 12 (10 for inbox and team-lead programmes, where each learner needs more live review). Coffee is available on the floor; lunch on full-day classes is 60 minutes and is off the stopwatch.

On-site delivery is available for private cohorts at your premises in Singapore, provided there is a room with power, a board, and a network that will allow the licensed assistant to run. We bring the sample pack and the measurement sheets. Your IT team remains responsible for access. Travel outside the island is quoted separately and is uncommon in 2026.

Hybrid evenings are the default for Spreadsheet to Forecast: two evenings in the room and one remote, or the reverse, depending on the month. Remote learners still complete the same timed passes; a second marker is assigned in the room or on the call. We will not run a fully self-paced version. The measurement requires a clock and a second person on the day.

Fees and what is included

The listed fee is per person, in Singapore dollars, for one open-class seat. It includes the contact hours, printed and digital notes, the prompt file used in class, the checklist, the measurement sheet with your own baseline and after readings, and a 30-day clinic (30 minutes, remote) for questions that appear once you are back at your desk.

The fee does not include a licence for the assistant. You attend with the product your organisation already pays for. It does not include production engineering, custom model work, or a guaranteed saving. Payment is by invoice, due before the first session unless a private-cohort schedule is agreed in writing.

If we cancel a class, the fee is refunded in full. If you cancel more than ten working days before the first session, the fee is refunded minus a S$80 administration charge. Inside ten working days we will move you to a later 2026 date where seats remain, or retain the fee if no date is usable. The clinic lapses 30 days after the last contact hour.

Booking a private cohort

A private cohort starts at six participants from the same organisation, with a practical upper bound of 12 so that the stopwatch protocol still holds. Dates are agreed at least three weeks ahead. We need the name of the task, confirmation that an anonymised sample can be used, and a named coordinator who will collect logins and dietary notes.

Prepare a stripped sample before the scoping call if you can: twenty tickets, a year of a monthly series, or a previous pack with names and account numbers removed. If stripping is slow, we can run the published demo pack and still take a baseline; the number will then describe the demo pack, which is the same limitation as the public outcome records.

Write to [email protected] or use the contact form. Telephone +65 6774 3820, Monday to Friday, 09:00–18:30.