AI proof of concept · Bangkok, Thailand

Test the idea.
See the evidence.

Find out what AI can do for one real business task. A focused pilot on your data gives you a working prototype, measured results and a clear next step.

Fixed scope. Fixed price. Typically 2–4 weeks, agreed after scoping.

From an idea to a decisionOne use case
01 / DefineYour task + your sample data

Agree on what a useful result looks like.

02 / TestA working AI prototype

Try real cases. Measure quality. Examine failures.

03 / ReviewEvidence for your next step

Results, production scope and projected running costs.

ProceedAdjustStop
Illustrative pilot process. The recommendation follows the results, including when they do not justify a production build.
Choose a useful first test

One task.
A result you can judge.

A good pilot starts with clear inputs, an expected output and someone who knows what “good” looks like. These are examples of tasks we can evaluate.

01

Read and extract

Turn invoices, forms or other documents into structured fields for your team to review.

What to measure

Field accuracy, exception rate and review effort.

02

Find and answer

Search approved manuals and procedures, with answers linked back to their sources.

What to measure

Answer quality, source relevance and unsupported answers.

03

Sort and draft

Classify incoming requests or prepare a first response for a person to approve.

What to measure

Classification accuracy and the editing each draft needs.

How the pilot works

A short build.
A structured evaluation.

We keep the scope focused enough to answer a useful question in weeks. You see the prototype working and review the results with the engineers who built it.

The timebox and price are agreed before the build, with data access and feedback planned into the schedule.

  1. Agree on one question

    Review the task, data readiness and business value. Define the inputs, expected outputs and acceptance criteria before development starts.

  2. Build on your real data

    Develop a working prototype in an agreed environment. Your process owner tries it on representative documents and cases, including difficult examples.

  3. Evaluate and decide

    Review quality, failures, response times and estimated running costs. Decide whether to proceed, adjust the approach or stop.

What you leave with

More than a demo.
A basis for a decision.

The deliverables connect technical feasibility to the next investment decision: what works, what needs attention and what production would involve.

01Try it

A working prototype

Software your team can use during the evaluation, built around the agreed task and sample data.

Why it matters

See how the approach behaves on the work your team actually does.

02Measure it

An evaluation report

Results against your acceptance criteria, including the cases that failed or required human review.

Why it matters

Understand where the feature helps and where its limits remain.

03Scope it

A production plan

Proposed architecture, integration points, data controls and projected operating costs, including model usage.

Why it matters

Know what still needs to be built before a production release.

04Decide

A clear recommendation

A recommendation to proceed, refine the scope or stop, supported by the results of the pilot.

Why it matters

Make the next investment decision with evidence you can review.

Before we build

Bring a process.
We’ll help shape the test.

Describe what comes in, what should come out and where your team spends time today. We will help turn that into a focused pilot.

Scope your first pilot

What we need to get started

  • One task and the outcome you want to improve
  • Representative sample data, shared after agreeing on handling
  • A process owner to evaluate results
  • Time for feedback and a final decision review
Before you start

Questions worth asking.

A useful pilot has clear boundaries, including what happens if the results fall short.

Talk through your use case
What do you need from us to run a pilot?

A representative sample of documents or records, a process owner who can judge the results, and time for regular feedback. We agree on access and data handling before you share anything. Samples can be anonymized where appropriate.

How long does an AI proof of concept take?

The typical build and evaluation window is 2–4 weeks for one focused use case. We agree on the schedule during scoping, after checking data readiness, access and availability for feedback. Broader requirements may need a smaller first pilot or a longer plan.

How is the price agreed?

We define the use case, deliverables, evaluation criteria and timebox first, then quote a fixed price for that scope. Any proposed changes are discussed before additional work. Production development is scoped separately after the results are available.

How do you protect confidential data during the pilot?

The pilot runs under NDA in an environment agreed with you. Before starting, we define the data to use, where it can be processed, who can access it and what is retained or deleted afterward. These requirements guide the model and infrastructure choices.

Is the prototype ready for production?

The pilot tests feasibility for the agreed task. A production release usually requires further integration, access controls, monitoring, error handling and deployment work. The production plan identifies those gaps so the next scope is explicit.

What if the results are not good enough?

That is a useful result too. We explain the limitations and whether different data, a narrower task or another approach is worth testing. If the case does not justify further investment, the recommendation can be to stop.

Start with a question worth answering

Which task would you
like to put to the test?

Tell us about the process and the result you need. We will come back with a proposed scope, fixed price and realistic schedule.

Discuss your AI pilot