The AI conversation in Sri Lanka is mostly about tools. The useful version is about hours: which repeated task eats your team's week, and can a machine do the first 80% of it reliably. That is the lens we use on every AI automation project we take on.
Five automations that usually pay back fast
- Inquiry triage: read incoming WhatsApp, form and email messages, classify intent, route to the right person with a draft reply.
- Quotation drafting: turn a request into a priced draft from your own rate card for a human to approve.
- Document intake: pull fields out of invoices, purchase orders or ID documents into your system.
- Reporting: assemble the weekly numbers and write the summary paragraph.
- Follow-up: sequence reminders so no lead goes cold because someone forgot.
Rules or AI?
If the input is structured and the logic is fixed, use rules — they are cheaper, faster and predictable. Use AI where the input is messy: free-text messages, scanned documents, photos, mixed Sinhala, Tamil and English.
Most working systems combine both: AI reads and drafts, rules decide and route, humans approve anything with money or reputation attached.
What it costs
Two costs: the build, and the model usage. Usage for a typical SME workflow is small — often a few thousand rupees a month — because volumes are modest. The build depends on how many systems it must touch.
How to run a pilot
- Pick one task and measure its current cost in hours per week.
- Define what good looks like before building.
- Keep a human approving the output in phase one.
- Log every run so you can audit failures honestly.
- Expand only after the numbers hold for a month.
