Before investing in an AI application, choose one repeatable business task, check the data it needs and decide who reviews the result. Custom software provides the workflow and controls; AI should be added only where its output can be tested and handled safely when it is wrong.
A Sri Lankan SME does not need an ambitious AI transformation programme to begin. It needs a well-defined problem and a way to compare the proposed solution with the current process. This checklist covers the decisions to make before requesting an AI application development proposal.
Which Task Should You Automate First?
Choose a task that happens regularly, has an identifiable owner and produces an outcome you can review. Examples include sorting incoming enquiries, drafting replies from an approved knowledge base or suggesting categories for business documents. These are possible uses to evaluate, not guaranteed capabilities or results.
Describe the starting input, the steps staff follow and the final output. Record how long the work currently takes and where errors appear. If the underlying process changes every day or nobody owns it, agree on the process before introducing another tool.
Do You Need AI or Ordinary Automation?
Use explicit rules when the required behaviour is predictable: assigning a request by a selected category, checking that a field is complete or sending an approved reminder. AI may be worth testing when the input involves varied language or patterns that do not fit a small set of reliable rules.
| Task | Starting approach | Control to discuss |
|---|---|---|
| Known approval conditions | Rules-based workflow | Document exceptions and who may override them |
| Drafting from varied enquiries | AI-assisted draft with staff review | Check factual accuracy before sending |
| Updating stock or financial records | Validated software workflow | Require permissions, checks and an audit trail |
| Extracting document details | Test AI assistance against representative samples | Validate fields before accepting changes |
A custom software application can combine deterministic checks with AI assistance. Keeping critical business rules explicit makes it easier to investigate mistakes and explain what the system is allowed to do.
Is Your Data Ready for the Proposed Task?
Collect representative examples, including incomplete records, different document layouts and the languages your users actually use. A demonstration on a few clean examples is not enough to show that a system will work with everyday business data.
- Identify where each field comes from and who is responsible for keeping it accurate.
- Remove duplicates and resolve inconsistent customer, product and supplier identifiers.
- Confirm that the team is authorised to use the data for the proposed purpose.
- Decide what may be sent to a third-party service, stored or included in logs.
- Agree on access permissions, retention and a process for correcting records.
If the project connects to a POS system or ERP system, define which application owns each record. An AI-generated suggestion should not silently overwrite operational data. Our POS and ERP guide explains the surrounding workflow choices.
Who Reviews the Output and Handles Mistakes?
Decide which outputs can be shown as suggestions, which require approval and which should never trigger an action automatically. A drafted customer reply can wait for a member of staff. A request to change a price or financial record needs a separate, validated permission path.
Give reviewers the original input and an easy way to correct or reject the result. Plan a fallback when the service is unavailable, the input is unsuitable or the result fails validation. Keep records that help investigate an incident without exposing unnecessary sensitive information.
For broader context, the NIST AI Risk Management Framework provides voluntary guidance for considering trustworthiness in AI design, use and evaluation. It is a reference for planning, not a certification of a proposed system.
How Should You Test a Small Pilot?
- Set a baseline: record how the task is completed today.
- Define acceptance: agree on acceptable error types, review effort and completion time.
- Keep test examples separate: include cases not used while tuning the solution.
- Test exceptions: include missing data, misleading input and service failures.
- Compare total effort: count review and corrections as well as generation time.
- Make a decision: expand, revise or stop based on the observed result.
Do not set a promised productivity percentage before the pilot. The useful result is whether this task becomes easier to complete to an agreed standard, including the work people still need to do.
What Should Your Software Development Brief Include?
Include the workflow owner, sample inputs, desired outputs, existing systems, permitted actions and review rules. Ask the provider to separate setup, integrations, ongoing service usage, monitoring and support in the proposal. Agree on how data can be exported and what happens if a provider or model changes.
Choose the interface around the people doing the work. A browser-based staff portal may be sufficient; a field team might need a mobile app. Our website vs mobile app guide helps you decide how to deliver the workflow without building unnecessary channels.
Common Questions Before Starting an AI Project
Can AI connect to my existing business software?
It may be possible if suitable interfaces, permissions and data are available. Start by reviewing the existing system and testing a narrow integration. Do not assume a connection is available simply because a product has an AI feature.
Do I need AI for every automation?
No. A clear business rule or a standard software workflow may be more appropriate. Use AI where there is a specific reason to test it and a way to evaluate the output.
Can a small business start with one task?
Yes. A focused pilot makes the scope, review effort and costs easier to discuss. Expand only after you have evidence that the process meets your requirements.
Find a Practical First Automation Project
Tell us about your current process, the people who will use the system and the problem you want to solve. ForDo IT can help you define a practical scope for your business in Sri Lanka or abroad.
Discuss your project