Phase 3: Implement

How We Deploy AI Without Disruption

The Implement phase is where the strategy becomes reality. We deploy AI capabilities in structured waves — each wave focused on 2–3 workflows, validated before the next wave begins.

This approach prevents the chaos of trying to transform everything at once. It builds organizational confidence with early wins before moving to more complex workflows.

The Wave Methodology

Each implementation wave follows the same structure:

Wave Scoping (Week 1). Define the specific workflows in scope, success metrics, data requirements, and acceptance criteria. Nothing moves forward without agreed criteria.

Build and Integration (Weeks 2–4). Our team configures, integrates, and tests the AI solution in your environment. We work with your IT team — not around them.

Pilot Deployment (Weeks 4–6). Deploy to a controlled group of users. Measure against baselines. Gather structured feedback. Identify edge cases.

Validation and Adjustment. Review pilot results against acceptance criteria. Adjust configuration, training data, or process design based on findings.

Full Rollout. Once validated, roll out to the full user group. Document everything. Create SOPs.

Wave Retrospective. Formal review of what worked, what didn't, and what to carry forward to the next wave.

What We Build

Depending on the Priority Matrix from the Strategy phase, a typical implementation program deploys combinations of:

• AI chatbots for citizen-facing or customer-facing interactions
• Document processing systems for OCR, classification, and data extraction
• Workflow automation agents that handle multi-step processes end-to-end
• Analytics dashboards with AI-powered anomaly detection and forecasting
• Internal AI assistants for staff productivity and knowledge access

Embedded Delivery Model

Our consultants work on-site or in close remote collaboration with your teams throughout implementation. This is intentional.

Transformation that happens in isolation from the client team creates fragile systems that fail when our consultants leave. Transformation that happens alongside client teams creates internal capability.

Deliverables

Deployed AI Workflows. Working AI systems in production, integrated with your existing technology stack, handling real workloads.

Integration Documentation and SOPs. Complete technical documentation, operational runbooks, and standard operating procedures for every deployed workflow.

Performance Baselines. Pre- and post-deployment measurements for each workflow: time saved, error rates, throughput, cost per transaction.

Wave Retrospectives. Documented learnings from each wave, including what was adjusted and why. This becomes institutional knowledge.

The Wave Structure

Each implementation wave follows a consistent pattern:

Week 1-2: Configuration and Integration. We configure AI systems for your specific workflows, data sources, and business rules. We build the integration layer connecting AI to your existing systems.

Week 3-4: Testing and Validation. We test with real data and real workflows. Domain experts validate the outputs. We refine configuration based on validation results.

Week 5-6: Training and Deployment. We train the staff who will work with the new system. We deploy to production with monitoring and support. We measure against the success metrics defined in the Strategy phase.

Week 7-8: Stabilization and Review. We monitor production performance, address issues, optimize configuration, and conduct a wave retrospective. Lessons learned feed into the next wave.

Why Wave-Based Deployment Works

Waterfall AI implementations — design everything, build everything, test everything, deploy everything — fail at a rate that should give any organization pause. The requirements change. The technology evolves. The organization learns things during implementation that invalidate original assumptions.

Wave-based deployment reduces this risk by delivering working AI in 6-10 week cycles. Each wave delivers measurable value. Each wave teaches the organization something about how AI works in its specific context. And each wave builds the organizational confidence needed for the next wave.

Integration With Your Existing Systems

We do not require you to replace your existing systems. Our implementation approach builds AI capabilities that connect to your existing infrastructure — document management systems, ERPs, CRMs, case management platforms, and legacy applications. The AI layer augments what you have rather than replacing it.

What Comes Next

As each wave delivers deployed AI capability, Phase 4: Empower runs in parallel — building the organizational competency to use, maintain, and evolve the AI systems we deploy.

What Happens During This Phase

Week 1: Wave Scoping and Acceptance Criteria Definition. We define the specific workflows in scope for the wave, the success metrics and acceptance criteria that will determine whether the wave has succeeded, data requirements, integration dependencies, and the pilot user group. Nothing proceeds to build without agreed acceptance criteria — this prevents the common failure of building a system and then arguing about whether it works.

Weeks 2–4: Configuration and Integration Build. Our team configures the AI system for your specific workflows, business rules, and data sources. We build the integration layer connecting AI to your existing systems — document management, CRM, case management, or ERP — using your approved integration patterns and working directly with your IT team rather than around them. Integration development is typically the most technically complex component and the most common source of delays when poorly managed.

Weeks 4–6: Pilot Deployment and Structured Review. We deploy to a controlled pilot group — typically 5–15 users representing the range of use cases the system will handle in production. We monitor daily, conduct structured feedback sessions with pilot users, and measure outputs against the acceptance criteria defined in Week 1. We document every edge case, every failure mode, and every unexpected behaviour encountered in pilot.

Weeks 6–8: Adjustment, Full Rollout, and Stabilization. We refine configuration, training data, and process design based on pilot findings. We deploy to the full user group with monitoring, post-deployment support, and immediate response to production issues. We measure against performance baselines and document outcomes in the wave retrospective.

Client Involvement

Successful implementation requires sustained client engagement — it cannot be outsourced entirely to Remolda.

IT team: Active participation in the integration build is essential. Your IT team knows your infrastructure, your security requirements, and your approved integration patterns. We work with them, not around them. Expect 4–8 hours per week from 1–2 IT staff during the build phase.

Domain experts: Staff who understand the workflows being automated must be available to validate AI outputs during testing and pilot. Their judgement is the ground truth against which we measure AI accuracy. Expect 2–4 hours per week from 2–3 domain experts during the testing and pilot phases.

Pilot users: The 5–15 staff selected for pilot deployment are active participants, not passive testers. We ask them to document their experience, flag problems, and attend a structured feedback session at the end of each pilot week. This investment produces significantly better calibration than passive usage monitoring alone.

Executive sponsor: Available for wave retrospective presentation and for rapid decision-making if issues arise during pilot that require scope or priority decisions.

Common Challenges

Challenge 1: Integration complexity underestimation. Legacy systems that appear to have API connectivity in design reviews frequently have undocumented constraints, authentication issues, or performance limitations that surface only during development. Remolda addresses this by building extra integration assessment time into the scoping week and by maintaining a prioritised list of integration fallback approaches (file-based, RPA connectors, or manual data transfer bridges) for each integration dependency, so that an integration challenge does not become a full wave delay.

Challenge 2: Scope creep during build. Stakeholders frequently identify new requirements during the build phase — additional features, additional data sources, or expanded scope — that were not in the original acceptance criteria. Remolda addresses this through a formal change control process: new requirements are assessed for effort and impact, a decision is made whether to include in the current wave or queue for the next, and the decision is documented. This prevents scope creep without requiring stakeholders to feel that their input is being dismissed.

Challenge 3: Pilot resistance. Pilot users sometimes approach the pilot with scepticism or anxiety that affects their engagement and the quality of feedback they provide. Remolda addresses this through the Change Management Plan from the Strategy phase: pilot users are selected with care, briefed on what the pilot is and is not, given clear guidance on what "good feedback" looks like, and supported by an AI Champion from their team who has already used the system and can provide peer-level reassurance.

Output: What You'll Have

At the conclusion of each implementation wave, your organisation holds:

• Deployed AI workflows: Working AI systems in production, integrated with your existing technology stack, handling real workloads — not prototypes or demos.
• Integration documentation and SOPs: Complete technical documentation of the integration architecture, operational runbooks, and standard operating procedures for every deployed workflow.
• Performance baselines: Pre- and post-deployment measurements for each workflow — time per transaction, error rates, throughput, cost per transaction — establishing the evidence base for ROI measurement.
• Wave retrospective: A documented review of what was built, what was learned, what was adjusted, and what should be carried forward to the next wave. This becomes institutional knowledge.

---

Further reading: AI Automation: A Business Guide | Automating the Wrong Things