Frequently asked questions about Agentic AI engagements.

Agentic AI is a software system capable of making decisions and executing sequences of actions to accomplish business objectives — not merely responding to messages or running fixed scripts. Chatbots answer messages one at a time. Automation tools execute predefined steps. AI agents receive a goal, choose the steps, call tools or internal systems, and adjust their strategy when conditions change.

For organizations, the practical difference is workflow coverage: a chatbot completes one interaction; an agent completes one end-to-end process such as document review, ticket triage, or contract matching.

A prototype demonstrates that an approach can work conceptually. A production-grade implementation is designed to run daily under real load, with real internal system integrations, an accountable audit trail, and error controls that do not suspend business processes.

We focus on production-grade. Every agent we deliver to clients comes with observability, fallback paths, operational boundaries, and runbook documentation so the client's team can monitor and intervene without being fully dependent on the vendor.

Our primary focus is Indonesia's mid-market to large enterprises that already have mature internal processes and sufficient operational load to justify AI agent-based automation. Mid-market organizations with operational teams of dozens to hundreds of staff are our core ICP, because they typically have defined processes, reasonably structured data, and a Sponsor at the director or VP level who can make technology decisions.

For smaller organizations or those without baseline processes, we typically recommend the Diagnostic Workshop first before any larger commitment.

The readiness indicators we consistently look for: (1) a workflow with sufficiently high volume and repetition that ROI can be calculated; (2) the input data for that workflow is documented or accessible through existing systems; (3) a Sponsor at director level is willing to partner on the process; (4) the operational team is willing to be part of the outcome evaluation.

The Diagnostic Workshop is designed to validate these indicators in a structured way, before any larger implementation commitment.

The Diagnostic Workshop is a one-day on-site or online engagement producing three primary deliverables: (1) a map of the three most relevant AI agent implementation opportunities for the organizational context; (2) initial ROI estimates for priority opportunities; (3) recommendations for next steps, including whether a Pilot Build is worth pursuing.

This includes sessions with the client's team, a review of candidate workflows, and a concise written report shareable with internal stakeholders. Scope and investment are discussed in the initial conversation; there are no hidden fees.

The Pilot Build and Managed Retainer are engagements whose scope is significantly influenced by the complexity of the target workflow, the number of systems that must be integrated, and the level of governance rigor required. Publishing a single public figure would be misleading, as two organizations with different requirements would receive materially different estimates.

For transparency, the pricing framework is fully explained after the Diagnostic Workshop or after an initial discussion session with our technical team, ensuring the estimate provided is genuinely based on actual scope.

Pilot Build scope is determined by: (1) the boundaries of the target workflow — one specific end-to-end process, not a general platform; (2) the number of internal systems that must be integrated, such as ERP, CRM, knowledge base, or ticketing; (3) the level of governance requirements, including audit trail and human-in-the-loop controls; (4) the transaction volume to be served per period.

We encourage focused, measurable scope so the Pilot can be objectively evaluated against clear business metrics before any expansion discussion toward a Managed Retainer.

Technically yes, particularly if the internal team already has a mature workflow analysis and ready requirements documentation. However, for most organizations the Diagnostic Workshop actually reduces total engagement cost, because it validates whether the chosen workflow is truly the best AI agent candidate and whether the expected ROI is realistic.

In practice, approximately two out of three clients who enter via the Workshop proceed to a Pilot, and the resulting Pilot scope is typically more precisely targeted.

The Diagnostic Workshop uses full upfront payment via bank transfer, with a formal invoice (VAT as applicable per PKP status). The Pilot Build uses a milestone payment structure — payments split across delivery stages agreed in the Statement of Work, typically three to four milestones with objective validation per stage. The Managed Retainer uses a monthly structure with invoices issued at the start of each service period.

All payments are made via bank transfer to an account in the name of PT Widigital Tri Buana. Terms and details are set out in the contract signed before the engagement begins.

Not as a prerequisite. Our engagement already includes the engineers, solution architects, and AI specialists who handle end-to-end implementation. Many clients choose not to form an internal AI team at the outset, precisely because one of the key benefits of a vendor engagement is the elimination of hiring risk in a rapidly evolving area.

For longer-term Managed Retainers, we recommend that the client have one internal Product Owner as a coordination point — but this is a business stakeholder role, not an engineering position.

Integration is a standard part of the Pilot Build and is designed around the APIs or access mechanisms available in the client's systems. For mainstream systems such as SAP, Oracle, Salesforce, HubSpot, Zoho, or internal platforms with REST APIs, integration follows standard patterns using dedicated service accounts.

For legacy systems without a formal API, we use an adapter approach — lightweight middleware bridging the agent with the target system. Integration strategy is mapped during the Diagnostic Workshop and documented in the Statement of Work before implementation begins.

Stack selection is matched to workflow requirements, the client's security boundaries, and the cloud provider preferences of the organization. For the LLM component, we work with models from mainstream providers (Anthropic, OpenAI, Google) as well as open-source models that can be run on client infrastructure where privacy requirements demand it.

For orchestration, retrieval, observability, and integration components, we use a combination of proven production-grade open-source and commercial tools. The final stack is discussed transparently during the Pilot Build architecture phase, including the long-term operational cost implications.

All three options are available, depending on the client's security requirements, regulatory profile, and operational cost profile. Cloud-hosted is the default for organizations without specific regulatory requirements, as it offers the most efficient cost profile and the most flexible scaling.

On-premise or hybrid is used for organizations in regulated sectors (banking, healthcare, government, strategic state-owned enterprises) or those with strict data residency policies. For these configurations, we design an architecture that keeps sensitive components within the client's perimeter while allowing non-sensitive components to leverage cloud services.

Data handling follows the principle of minimization — agents only access the data necessary to complete the target workflow, through credentials with limited scope. Client data is not used for model training and is not shared with third parties beyond the infrastructure providers agreed in the contract.

For inter-component communications, we apply standard in-transit encryption. For data at rest that forms part of the agent workflow, storage mechanisms are designed to match the retention and access requirements agreed in the engagement. Engagement-specific details are documented in a Data Processing Agreement accompanying the master contract.

Yes, and this is a standard part of every implementation we deliver. Every agent is designed with decision logging, capturing the inputs received, tools called, tool outputs, and final output generated. These logs form an audit trail that can be queried by the client's team through the observability dashboard included in the delivery.

For workflows with material impact (financial, legal, customer-facing), we design human-in-the-loop checkpoints requiring human approval before the agent executes certain actions. The granularity of audit logging and checkpoints is configurable to match the client's internal governance policies.

We design implementations to comply with the principles of the Personal Data Protection Law No. 27 of 2022, including data minimization, processing transparency, data subject rights, and data breach notification obligations. For every engagement involving the processing of personal data, we prepare a data flow mapping and document the legal basis for processing in accordance with the statute.

Full compliance requires collaboration between the technology provider and the data controller (the client). WTB provides the technical framework and documentation supporting the client's obligations as data controller. For compliance verification against external audits or interpretation of sector-specific regulations, we recommend engaging a legal counsel or internal Data Protection Officer.

Technically yes. At the end of the Pilot Build, the client receives runbook documentation, access to code and configuration, and a knowledge transfer to the designated internal team. For organizations with a strong engineering team and the capacity to monitor, update prompts, and handle emerging edge cases, continuing without a retainer is a valid option.

However, most clients choose to continue with the Managed Retainer, primarily because agents evolve continuously alongside model changes, user behavior, and business requirements. The retainer provides ongoing access to the WTB team for iteration, proactive monitoring, and capability expansion without forming an internal AI team.

Model and prompt updates are an ongoing activity for production-grade agents. Major model providers (Anthropic, OpenAI, Google) periodically release new versions with different characteristics. User behavior also evolves, and new edge cases emerge after the system has been running for several months.

For Managed Retainer clients, these updates are part of the standard service, with a regression test process run before any changes are promoted to production. For clients not continuing with a retainer, we document the update procedure in the runbook so the internal team can perform updates independently.

Yes, knowledge transfer is a standard part of the Pilot Build delivery. During the Pilot closing phase, we conduct a transfer session with the client's designated technical team and business stakeholders, covering system architecture, monitoring procedures, troubleshooting guides, and runbook documentation.

For organizations seeking to build deeper internal capacity, we offer a training program that can be appended to a Pilot or Retainer engagement, covering prompt engineering practice, evaluation fundamentals, and agent operations. Training program details and fees are structured separately to match the client team's requirements.