How to Implement AI in an SME: A 90-Day Plan for Malaysian Businesses
A week-by-week 90-day plan to implement AI in a Malaysian SME: pick one use case, train the team, pilot with a KPI, and decide on production with real numbers.
Implement AI in an SME in three 30-day blocks. Days 1 to 30: choose one use case, measure the baseline and train the team. Days 31 to 60: run a pilot on real work with one owner and one KPI. Days 61 to 90: decide, harden and hand over. One use case done properly beats five started.
Part of the guide: AI Consulting in Malaysia: What It Costs and What It Delivers (2026)

On this page
- What does the 90-day plan look like?
- Weeks 1 to 2: how do you choose the use case?
- Weeks 3 to 4: what must be ready before the pilot?
- Weeks 5 to 8: how do you run the pilot?
- Weeks 9 to 10: how do you decide?
- Weeks 11 to 13: how do you make it last?
- What should you do next?
- Sources
- Frequently asked questions
A 90-day plan works for a small business because it forces one decision at a time. The goal is not to "have AI" but to change one repeated piece of work and prove it with a number. This plan is the one we use as a template at winsym.ai; adjust the timings to your size.
Prepared 22 October 2026. Figures are winsym.ai prices and the sources at the end.
What does the 90-day plan look like?
| Weeks | Block | Output |
|---|---|---|
| 1 to 2 | Choose | One use case, one owner, one KPI, baseline measured |
| 3 to 4 | Prepare and train | Data access checked, policy agreed, team trained on real tasks |
| 5 to 8 | Pilot | Live use by a small group, weekly log of time and errors |
| 9 to 10 | Decide | Pilot result against the KPI, go or stop |
| 11 to 13 | Harden and hand over | Review step, documentation, owner trained, next use case queued |

Weeks 1 to 2: how do you choose the use case?
Score candidates on volume, rule clarity, data access, cost of a mistake and time to result, then pick the best. Our scoring method is in how to pick your first AI use case. Good first candidates are repetitive and checkable: drafting replies to common enquiries, summarising meetings, extracting data from supplier documents, reporting. Avoid starting with anything where a wrong answer reaches a customer unreviewed.
Then measure the baseline. For one week, count hours spent, volume handled and errors. Without this number you cannot prove anything at day 90. Pick the single KPI: hours per week, response time, or error rate.
Name one owner. In a 30-person company this is usually an operations or sales manager with authority over the process.
Weeks 3 to 4: what must be ready before the pilot?
Three things: access, rules and skills.
- Access: check where the data lives. If it sits in a CRM or accounting system with an API, you are in good shape. If it lives in personal WhatsApp chats and spreadsheets, add cleanup to the plan.
- Rules: agree a one-page policy covering which data may enter AI tools, who reviews outputs and who may approve exceptions. Malaysia's Personal Data Protection Act 2010 applies once personal data is involved; see PDPA and AI integration.
- Skills: train the pilot group on their own tasks. A half-day in-house session starts from MYR 4,500 at winsym.ai and can be HRD Corp Claimable if you apply at least 14 days before it starts; see the HRD Corp claimable AI training guide. Book the date early because of that lead time.
If you want a second opinion on the use case before you train anyone, an AI Opportunity Audit (from MYR 3,500) does that in 3 to 5 working days for one workflow.

Weeks 5 to 8: how do you run the pilot?
Keep it small and real. Three to five users, live work, and a weekly ten-minute review.
- Each user logs time saved and errors found, in a shared sheet.
- The owner reviews a sample of outputs every week and records what went wrong.
- Fix the instructions, templates or data, not the people.
- Do not add new features during the pilot. Changes break the comparison.
A pilot with a floor works best: agree in advance the minimum result that justifies continuing, for example 5 hours a week saved across the group with no increase in errors.
Weeks 9 to 10: how do you decide?
Compare the pilot to the baseline. There are three honest outcomes:
| Result | Action |
|---|---|
| Hits the floor with fewer errors | Move to production and roll out |
| Close, with a clear fixable cause | Run four more weeks with the fix |
| Misses and no clear cause | Stop. Pick the next use case |
Stopping is a valid outcome. A measured stop costs a few weeks and teaches you which assumption was wrong. The expensive failure is the pilot that never ends and never gets measured; see why AI pilots fail before production.
Weeks 11 to 13: how do you make it last?
Harden the process so it survives holidays and staff changes:
- A review step for anything that reaches a customer.
- A short written procedure (one page) and a recorded walkthrough.
- A second person trained to run it.
- A monthly check of usage and accuracy.
If the pilot moved from a ready-made tool to a custom connection into your CRM or ERP, this is when a build starts. Custom builds begin at MYR 12,000 for one process; the cost logic is in how much it costs to implement AI. Then pick the second use case from your scored list.
What should you do next?
Spend the next hour on week 1: list five repeated tasks and count the weekly hours for each. Then book a free 30-minute call and we will help you pick the one worth piloting. The wider context is in AI consulting in Malaysia: costs and deliverables; for coaching the owner through the 90 days, see coaching.
Sources
Links checked November 2026.
- Red Hat, Red Hat and NAIO study of 133 Malaysian SMEs, press release, 30 October 2025 (study by Ecosystm; 36% piloting, 21% scaled beyond pilots, 60% cite lack of in-house technical expertise).
- AI Malaysia (National AI Office), national programme background.
- winsym.ai price list and audit scope, winsym-ai.my/audit, checked 22 October 2026.
- HRD Corp Employer's Circular 2/2026 (14-day lead time), as summarised in our HRD Corp guide, checked 22 October 2026.
The plan is our working template, not a guarantee of results. Timings depend on data access and team availability.
Frequently asked questions
Can an SME really implement AI in 90 days?
Yes, for one well-chosen use case. Ninety days is enough to measure a baseline, train staff, run a pilot on live work and decide. It is not enough to transform the whole company.
Who should own the AI project in a small company?
One named person with authority over the process, usually an operations or sales manager, not the IT contact. The owner decides what success means and who uses the tool.
How much time does the team need to give?
Plan on about two to four hours a week for the owner and one hour a week for each pilot user. The training session adds a half day once.
What if the pilot fails the KPI?
Stop or change the use case. A failed pilot with a measured result costs little and teaches you which task or data was the problem. Continuing without a number is the expensive option.