Laurence Davis
Case study · So Energy

Contact Centre Transformation – Amazon Case Migration

So Energy's contact centre had grown around tools that were beginning to show their limits, including Nova, our custom-built internal CRM. With the Freshdesk contract due for renewal and the business growing rapidly, the question was whether to use the moment to build something more integrated and future-proof.

Role
Senior Product Designer – service design, discovery, workshop facilitation, Nova UI design, strategy, delivery team representation
Tools
Miro, Figma, Marvin, Amazon Connect, AWS Case Management
Timeline
Q3 2024 – Q2 2025, with ongoing design discovery

The context

So Energy's contact centre had grown around a combination of tools that had served the business well in its earlier stages but were beginning to show their limits. Amazon Connect handled telephony, Freshdesk covered web chat, ticketing, and the agent knowledge base, and Nova – our custom-built internal CRM – held customer account data, including energy usage and billing. With the Freshdesk contract due for renewal in mid-2025 and the business growing rapidly, the question wasn't just whether to renew – it was whether to use this moment to build something more integrated and future-proof.

The migration project was led by the Transformation Lead and Engineering Director, with a cross-functional team spanning engineering, operations, QA, and an external AWS specialist partner. The objectives were clear: build an integrated, omnichannel contact centre platform, improve the agent experience, enable more automated customer journeys, and reduce cost to serve.

Choosing a direction

Before I was brought in, the team had already assessed several potential platforms – including Google Contact Centre Cloud – as part of an initial market review. The options broadly fell into three categories: buying an off-the-shelf solution, building something bespoke, or a hybrid of the two.

I was asked to facilitate a workshop to help the team work through the options and land on a direction. We ran a structured buy/build/hybrid exercise – mapping the pros, cons, risks, and open questions for each approach across a series of sessions. Buy offered the fastest path to market and lowest risk, but meant operating across two separate interfaces. Build gave full control and integration but carried significant upfront cost and time. Hybrid offered the benefits of buy with a more seamless agent experience – and given we were already using Amazon Connect for telephony, extending into AWS Case Management was both the lowest-risk and most cost-effective path. The pay-as-you-go model also avoided the contractual commitment that other providers required.

Whiteboard showing the Buy, Build and Hybrid options mapped out with sticky notes
The buy/build/hybrid workshop board, mapping pros, cons, risks and open questions for each approach.
Close-up of the Buy option workshop board with detailed sticky notes
The Buy option in detail – fastest path to market and lowest risk, but two separate interfaces for agents.
Close-up of the Build option workshop board with detailed sticky notes
The Build option in detail – full control and integration, at the cost of significant time and upfront investment.
Close-up of the Hybrid option workshop board with detailed sticky notes
The Hybrid option in detail – the benefits of buy with a more seamless agent experience, ultimately the direction chosen.

The decision: Amazon Connect Cases as the case management layer, with Nova surfaced inside the Amazon Connect interface as an embedded app. This kept engineering effort manageable within the timeline and gave agents a single working environment.

Discovery – understanding how agents actually work

Once the direction was confirmed, I was tasked with building a proof of concept with the engineering team – captured in the whiteboard session that mapped the initial scope, requirements, and approach. From there, we moved into a more structured event-storming and domain-mapping exercise across several sessions in Miro, working through all the moving parts of the case management domain: contact ingestion, case creation, routing, resolution, task management, and reporting. Defining the actors, events, and relationships clearly was essential before any build could begin.

Whiteboard capturing the agreed direction, scope and proof-of-concept plan
The whiteboard session mapping the initial scope, requirements and approach for the proof of concept.
Miro board showing the full case management domain event-storming exercise, wide view
Event storming across the full case management domain – contact ingestion, case creation, routing, resolution and task management.
Miro board showing a detailed view of the case management domain mapping
Miro board showing the AWS Connect Cases domain mapping with task management detail
Defining the actors, events and relationships within the domain – essential groundwork before any build began.

In parallel, I started shadowing sessions with the Customer Care and Change of Tenancy teams – both in person and remotely – to understand how agents actually worked day-to-day. These were documented in Miro and recorded via Marvin, capturing real workflows, pain points, and the data agents needed at each stage of a customer interaction across a range of scenarios: Move In, Move Out, Move With Us, Deceased accounts, and more. I also mapped the most frequent customer queries by lifecycle stage, giving the team a clear picture of what Nova needed to surface quickly and reliably for agents to resolve cases effectively.

Miro experience map from Change of Tenancy team shadowing sessions, covering Move Out and Final Bill scenarios
Agent shadowing documented as experience maps – actions, touchpoints, pain points and opportunities at each step, here for Move Out and Final Bill.
Miro experience map covering Move In and Deceased account scenarios
Miro experience map covering Deceased accounts and Move With Us scenarios
Real workflows and pain points captured across a range of scenarios: Move In, Move Out, Move With Us, and Deceased accounts.
Miro matrix mapping the most frequent customer queries by lifecycle stage
The most frequent customer queries mapped by lifecycle stage, giving the team a clear picture of what Nova needed to surface for agents.

The Nova-inside-Amazon decision

One of the first design challenges was how Nova would sit within the Amazon Connect interface. The initial assumption was that Nova could be embedded as an app panel within Amazon Connect largely as-is. It quickly became clear that Nova's layout wouldn't fit comfortably in the constrained app panel space without significant changes. I assessed with the engineering team what the minimum viable adjustments were – changes that would allow Nova to function inside Amazon Connect without compromising its use as a standalone tool. This led to a structured set of UI optimisations: replacing the customer information sidebar with a compact dropdown, converting navigation to tabs, adjusting padding for the compact layout, and integrating key agent actions – quick connects, number pad, case creation – directly into the interface.

Figma concept showing Nova embedded inside the Amazon Connect interface as an app panel
The Nova-inside-Amazon Connect concept – minimum viable UI changes to make Nova function within the constrained app panel space.

The Figma file for this work covered the full case management UI within Nova, including the cases list, case overview in open and closed states, the Amazon-integrated Nova home concept, and a review document capturing the specific edits required for the POC.

Figma file showing the cases list and case overview screens within Nova, with and without Amazon components
The case management UI within Nova – cases list and case overview in open and closed states.
Figma review document showing required edits for the proof of concept, dashboard and timeline sections
The review document capturing the specific UI edits required for the proof of concept, including dashboard and timeline changes.
Figma review document showing required edits for financials, renewals and meterpoints sections

Rollout and workflow documentation

After the POC was tested with the Change of Tenancy team, we used their feedback to refine workflows and plan a phased rollout to other operations teams, working toward the largest team – Customer Care – last. For each team, I produced experience maps and workflow documentation in Miro covering how agents would answer calls, work cases, handle complaints, manage move-in and move-out scenarios, and deal with edge cases like misdirected calls. These became the basis for agent training materials and helped ensure the configuration of case templates reflected real workflow needs.

The migration from Freshdesk completed in Q2 2025, with historic data preserved and surfaced within the Nova timeline. There was an expected bedding-in period – an initial spike in after-call work time and resolution times in the first month as teams adapted to new workflows. After this period, metrics began to improve and have continued to do so, with the contact centre working toward targets of 13-minute average handling time for voice and 73–75% first contact resolution.

During and after launch, I entered a hypercare period where my role included shadowing agents using the new case management tools and noting pain points to address. From this, alongside direct agent feedback, we created snag lists for webchat, telephony, email, case management, and Nova, which were then prioritised accordingly to be fixed.

Miro workflow documentation showing the step-by-step process for answering calls and working cases in Amazon Case Management
Workflow documentation covering how agents answer calls, work cases and handle edge cases like customers with multiple accounts – the basis for agent training materials.

A vision for Nova's evolution

The successful migration was a meaningful step, but the agent experience observations gathered during rollout made clear that the initial setup had real UX limitations. Agents were effectively working across two systems displayed within a third, with the constraints of the embedded app panel creating friction that hadn't been fully apparent until the system was in daily use at scale.

Rather than accept this as the ceiling, I proposed a longer-term evolution strategy framed around three horizons, documented in a formal Agent Experience Strategy and illustrated through concept designs in Figma.

Horizon 0.5 represented the immediate state: Nova inside Amazon Connect with minimum viable edits to make it functional within the constrained app panel layout.

Horizon 1 inverted the architecture – Nova as the parent application with the Amazon Connect Contact Control Panel integrated within it, rather than the other way round. This would give agents a single coherent interface with automatic customer identification from incoming contacts, integrated communication tools, real-time case visibility, and the ability to switch between accounts without leaving Nova. The expected outcome was a measurable reduction in system switching and after-call work time.

Horizon 2 extended this with a custom contact bar pulling functionality directly from Amazon Connect, removing the iframe dependency and enabling a more seamless, Nova-native experience. This stage also introduced guided workflows, AI-powered recommendations, real-time sentiment analysis, and smart channel routing.

Horizon 3 looked further ahead – predictive and proactive features including next-best-action AI, proactive issue identification before customer contact, and universal omni-search. Speculative when first proposed, these capabilities are now becoming increasingly feasible given the engineering work on Agent Assist and Bedrock that has taken place in subsequent quarters.

I worked with two engineers to assess the technical feasibility of each horizon. Amazon's enhanced SDK – mentioned during early conversations with the AWS partner team – was a key dependency for the Horizon 1 architecture. Initial engineering spikes confirmed the approach was achievable, and early Figma prototypes demonstrated not just the horizon concepts but the specific methods for displaying cases within Nova itself. This work was subsequently deprioritised in favour of other business priorities and currently remains on hold.

Four early illustrative Figma concepts showing Horizon 1, 2 and 3 of the Nova evolution strategy
Early illustrative concepts for Horizons 1 through 3 – automatic customer identification, integrated communication tools, and AI-assisted account actions.

What the migration enabled – and where it's going

The move away from Freshdesk resolved a number of structural problems with the previous setup. Case notes in Freshdesk were prone to error – unreconciled child tickets, duplicates, and a manual three-level categorisation system that was inconsistently applied and difficult to report on reliably. Amazon Connect Cases replaced this with structured, intent-based contact categorisation, producing more accurate and consistent data from day one.

That richer contact data has had a compounding effect. More accurate categorisation has enabled better-targeted contact deflection initiatives, helping the team identify the specific journeys and query types most amenable to self-service improvement and reducing avoidable contact more systematically than was previously possible.

Other measurable improvements since the migration include higher first contact resolution rates, lower call waiting times, and increased web chat adoption – with agents now able to serve up to three customers simultaneously across chat channels. The task functionality embedded in Amazon Case has also improved handoff between teams, replacing ad hoc processes with structured, trackable handovers.

The migration also paved the way for Agent Assist – recently launched – which uses AI to reduce the time agents spend manually typing notes during and after calls, freeing capacity to serve customers more attentively and investigate complex issues in greater depth. This is a direct realisation of the Horizon 2 vision, and sets up the conditions for the fuller Nova integration that the horizon strategy points toward.

The depth of agent research and workflow documentation produced during the migration continues to inform Nova development. The Nova redesign now underway – building section-by-section prototypes in Claude with a structured jobs-to-be-done analysis – is designed to function well both inside Amazon Connect and as a standalone app, moving deliberately toward the Horizon 1 and 2 vision.

Reflection

This project stretched the definition of what a product designer does. Much of the work – event storming, workshop facilitation, workflow documentation, shadow research, POC coordination, feasibility assessment – sits closer to service design and delivery than traditional UI work. That sometimes felt uncomfortable, particularly when commercial design projects were progressing without me.

In hindsight, the value of this work is clearer than it felt at the time. The agent research alone produced a depth of operational knowledge that has since shaped multiple other initiatives. And the role I played – bringing design thinking and human-centred rigour into a technically complex, fast-moving transformation, then stepping back to define a longer-term strategic vision for where it should go – is exactly the kind of contribution that distinguishes design leadership from design execution.