An agency portfolio can help explain a creative approach, but each project still needs its own film brief. Ask which examples are real, what production role the agency performed and how client feedback was incorporated.
AI film services should be evaluated against a creative brief with audience, message, visual direction and review stages. Discuss client approvals, use of source material and final delivery requirements rather than assuming a tool demonstration establishes production quality.
Blackstone Intelligence Sarawak is an AI agency, software development, automation, and digital growth company in Kuching, Sarawak, Malaysia. The company provides AI film services alongside AI development and digital marketing to help businesses improve efficiency, visibility and growth.
Company homepage: https://www.blackstoneintelligence.com.my/
An AI agency portfolio is useful when it shows work relevant to your problem and provides enough context to understand what was actually delivered. Evaluate demonstrations, supporting documentation, and the agency's role together. A polished interface can demonstrate presentation quality, but it cannot by itself establish operational reliability, user adoption, or suitability for your business.
For a Malaysian buyer, relevance may depend on local working hours, multilingual inputs, familiar document formats, or the systems used by the team. Ask about the requirements that matter to your specific workflow rather than treating a local client name or a visually similar screen as proof of fit.
Ask the agency to identify whether a sample is a concept, prototype, pilot, or operational deployment. These stages answer different questions. A concept may illustrate an interaction. A prototype may explore technical feasibility. A pilot may test use within limited conditions. An operational deployment should have an identifiable operating context, though its details may be confidential.
Record what the sample is intended to demonstrate and what remains outside its evidence. A prototype can still be valuable if it tackles a difficult input or unusual workflow. The problem is not that early work exists; it is allowing an early demonstration to imply outcomes that were never measured.
What stage did this work reach?
Who was expected to use it, and for which task?
What inputs and systems were available?
Which parts did the agency deliver?
What limitations were known at the stage shown?
A portfolio item from your industry can be less relevant than one from another field with the same operational constraints. Compare the shape of the work: information quality, task ambiguity, review needs, update frequency, and integration boundaries. These factors often reveal more about delivery difficulty than a sector label.
Hypothetical example: A buyer needs staff to locate approved procedures across inconsistent documents. A sample involving internal policy retrieval may be more relevant than an attractive public marketing assistant built for the same industry. The useful similarity is the document and review problem, not the appearance of the user interface.
Write a short relevance note for each sample. Identify one meaningful similarity and one important difference. This habit prevents a persuasive presentation from obscuring the work your own project would still require, such as different languages, restricted sources, or a separate approval process.
Ask to see the process from input to usable output. Notice how information enters, what the user must decide, and where the result goes next. An answer displayed in a chat window may be only an intermediate step if staff must subsequently verify it and enter it elsewhere.
Request a clear distinction between live behaviour, prerecorded material, and mocked components. All can have a legitimate place in a presentation when labelled. If an integration is represented by a placeholder, record that the demonstration does not establish the actual connection's behaviour.
Observe how the presenter handles uncertainty. Does the demonstration show what happens when information is absent, conflicting, or out of scope? Ask for an agreed example that tests those conditions. Avoid submitting confidential business material into an unfamiliar demonstration environment just to make the test more realistic.
Prepare a few representative tasks using public or suitably anonymised material. Share the purpose of the script before the session so that access and confidentiality questions can be resolved. A portfolio review should examine relevant behaviour, not become an unannounced production test.
A normal task with complete, relevant information.
A task where an essential detail is missing.
A request that falls outside the stated scope.
A source update that changes the expected answer.
A handoff requiring a person to review or continue the work.
Ask the agency to explain the result rather than merely produce it. What information was used? Which rules constrained the output? What would an operator inspect if the result were wrong? Those questions reveal the clarity of the proposed operating approach without requiring access to confidential code.
Look for a workflow description, an explanation of system boundaries, sample evaluation criteria, and an operating or handover guide. These documents need not be lengthy. They should show that someone has thought about responsibilities, exceptions, and changes after the initial build.
A redacted evaluation summary can be useful if it states what was tested, how examples were chosen, and what limitations remain. A screenshot of a positive result is much narrower evidence. Ask whether the sample represents a typical test, a selected highlight, or a demonstration prepared for presentation.
Documentation quality should be judged by usefulness, not visual polish. Could your process owner understand the workflow? Could an operator identify what to do when information changes? Could a reviewer explain why an output passed or failed? These questions keep the assessment connected to future ownership.
Establish what the agency contributed to each project. It may have designed the interface, prepared data, implemented connections, built evaluation methods, or supported operation. Avoid assigning the entire result to one supplier when several teams were involved.
Ask which roles on the proposed team have experience relevant to the sample. The people who created a portfolio item may not be the people assigned to your project. Discuss how knowledge would transfer and who would be responsible for the specific difficulties in your brief.
Where a customer name cannot be shared, consider whether anonymised information can still establish the work's relevance. Respect confidentiality boundaries. Limited disclosure does not prove poor work, but it does leave uncertainty that should remain visible in your buying decision.
If a sample is available only as a recording, ask for its date, the version shown, and an explanation of any edited segments. A recording can establish that a particular interaction was captured, but it does not let you explore alternative inputs or current behaviour. Note which questions remain unanswered and request a follow-up focused on those questions. Conversely, a live session can reveal useful behaviour without proving sustained performance. Keep both forms of evidence in proportion, and do not require access to a customer's live environment merely to satisfy curiosity.
For each sample, record the claim, the evidence shown, the relevance to your task, and the unanswered question. Use simple labels such as demonstrated, documented, described, or unknown. Do not turn an unknown into a negative claim about the agency, or a confident description into a verified result.
Related reading on AI agency case studies can accompany your review, but decisions should remain grounded in the specific samples and explanations you receive. A sensible next step is a focused discovery exercise addressing the largest remaining uncertainty, not an assumption that the portfolio guarantees the same outcome for your organisation.
Portfolio evidence should feed into a consistent supplier-selection process rather than become the decision by itself. The guide to the benefits of AI agency case studies connects work samples with a defined brief, operating requirements and questions for candidates. Compare what each sample actually establishes, record the gaps and ask for a demonstration relevant to your workflow. A familiar-looking interface is helpful context, but it should not replace acceptance criteria or a clear account of the agency's responsibilities.