AI-Powered Contact Center Transformation for Energy Retail Customer Experience

energy · Energy · 2025 · source Agentic: no · promotional bias: med · source-backed · extracted with prompt 0.5.4 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Energy retail customer service

So Energy needed to rebuild customer-service operations after fragmented channels, manual context tracking, and back-office escalation gaps worsened already-frustrated energy customers' experiences.

Constraints:

Customers contact energy providers when something is already wrong
Voice, chat, email, and other channels previously lacked shared context
Transformation had to bring people and processes along, not just swap platforms

Architecture

So Energy replaced fragmented voice/chat/email systems with Amazon Connect as a unified cloud contact-center platform integrated with its internal CRM. Amazon Connect provides omnichannel context, identity verification, intent recognition, routing, case management, task tracking, contact summaries, agent assist, and a foundation for RAG chatbots and voicebots while agents work from one consolidated workspace.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  amazon-connect-platform["amazon-connect-platform"]
  identity-and-intent-recognition["identity-and-intent-recognition"]
  routing-engine["routing-engine"]
  agent-workspace["agent-workspace"]
  crm-integration["crm-integration"]
  case-and-task-management["case-and-task-management"]
  rag-chatbot-and-voicebot-roadmap["rag-chatbot-and-voicebot-roadmap"]
  amazon-connect-platform --> identity-and-intent-recognition
  amazon-connect-platform --> agent-workspace
  amazon-connect-platform --> crm-integration
  amazon-connect-platform --> case-and-task-management
  amazon-connect-platform --> rag-chatbot-and-voicebot-roadmap
  identity-and-intent-recognition --> routing-engine
  routing-engine --> agent-workspace
  agent-workspace --> crm-integration
  agent-workspace --> case-and-task-management

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`amazon-connect-platform`	Unified cloud contact-center platform for voice, chat, email, messaging, routing, analytics, case management, and AI features.	`identity-and-intent-recognition`, `agent-workspace`, `crm-integration`, `case-and-task-management`, `rag-chatbot-and-voicebot-roadmap`
`identity-and-intent-recognition`	Automatically verifies the customer and identifies intent across channels before agent handling.	`amazon-connect-platform`, `routing-engine`
`routing-engine`	Matches demand, urgency, and topic to agent capacity so customers reach appropriate support quickly.	`identity-and-intent-recognition`, `agent-workspace`
`agent-workspace`	Single UI replacing multiple screens, with full contact-center context and CRM data for agents.	`amazon-connect-platform`, `crm-integration`, `case-and-task-management`
`crm-integration`	Connects Amazon Connect to So Energy's internal CRM so energy-specific customer context appears in the workspace.	`amazon-connect-platform`, `agent-workspace`
`case-and-task-management`	Tracks customer cases, back-office tasks, and waiting states so escalations do not disappear between teams.	`amazon-connect-platform`, `agent-workspace`
`rag-chatbot-and-voicebot-roadmap`	Future self-service layer based on Amazon Connect that starts with a RAG chatbot and moves toward context-aware voicebots.	`amazon-connect-platform`

Data flow

A customer reaches So Energy through voice, chat, email, or messaging. Amazon Connect recognizes the customer and intent, keeps the case context across channels, routes the request to the right agent, and surfaces CRM/contact-center context in one workspace. Back-office tasks and customer waits are tracked in the same platform so escalations remain visible, and future chatbot/voicebot flows can reuse the Amazon Connect foundation.

“After Amazon Connect, we enabled instant recognition across all of the channels. So, we had initial automatic identity and verification along with intent recognition”transcript, stated

“So, literally operating from three screens down to one screen, and that was really powerful for the agents.”transcript, stated

Why this architecture

So Energy chose Amazon Connect because it needed a future-ready, lower-risk contact-center foundation that unified channels, preserved context, integrated deeply with its CRM, and could grow into chatbots, voicebots, and agentic capabilities without another migration.

“we got the unified cloud platform, we got all the channels that we wanted, voice, emails, chat initially, in one cloud-native solution, with complete context in one place”transcript, stated

“we weren't really comfortable going with a completely new entrant, given the importance of this to our business, but also the risk of it, because this is pretty much a one-way door decision”transcript, stated

Models used: Amazon Connect AI capabilities · RAG chatbot

Scale

So Energy serves around 300,000 customers and about half a million gas/electric meter points.
“We serve around 300,000 customers, about half a million meter points, gas and electric, and we are really formed with a mission”transcript, stated
The team trained more than 250 people in roughly four weeks during rollout.
“We trained about over 250 people in around four weeks on a new platform and then we finished rollout in June where we exited the old platform.”transcript, stated
The overall migration took about six to seven months.
“So all in all, it took about six to seven months to get that done. So, you can accelerate pace, it depends on the complexity of your own ecosystem”transcript, stated
Call-wait times dropped about 33% after the transformation.
“we've seen remarkable performance improvements, especially in reduced call-wait times. It's about 33%, but we expect that to even get better over the next few months.”transcript, stated
Chat grew from less than 1% to 15% of contact volume.
“before our chat volumes were pretty much negligible but increased that channel to 15% from like pretty much less than 1%, all at the same time.”transcript, stated

Alternatives considered (roads not taken)

Traditional CCaaS platforms with limited or repackaged AI features

Rejected because: They were strong but too traditional, with AI often limited to NLP recognition, summaries, or writing assistance.

“We looked at the regular traditional CCaaS platforms, very strong offering, but they're very traditional in that sense. Everyone started at that point in time, having like AI functionality”transcript, stated

New market entrants for the contact-center platform

Rejected because: The contact center was too critical and risky for a less-proven vendor choice.

“We also saw some new entrants coming to the market which offered some interesting things, but we weren't really comfortable going with a completely new entrant, given the importance of this to our business”transcript, stated

Using an implementation partner as the primary migration team

Rejected because: So Energy tried a partner for one part and found it slowed the project because internal engineers had the necessary context.

“we actually, to be honest, we did try implementation partner for one thing and we didn't really work out, because it was actually slowing us down, than actually speeding us up”transcript, stated

Patterns

`context-compaction` (agent-assist)

Contact summaries let agents quickly understand repeated or ongoing customer issues without rereading all prior interactions.

“You had contact summarization and so on, so customers could really glance really quickly what the customers were contacting us about before they even start speaking to the customer.”transcript, stated

`tool-use-orchestration` (workflow)

The unified platform keeps channel, case, CRM, and task context together so agents can solve the issue rather than reconstruct it.

“After Amazon Connect, we enabled instant recognition across all of the channels. So, we had initial automatic identity and verification along with intent recognition”transcript, stated

`rag-retrieval` (self-service)

So Energy launched a RAG chatbot and planned to add more context for customer self-service.

“We did the chatbot at the same time with a RAG 11 chatbot based on Amazon Connect. We wanna start giving it more context and making it more context aware”transcript, stated

`structured-output-contract` (automation)

The team is experimenting with deterministic tooling that reads email intent and requests missing details before processing.

“where we can give some deterministic tooling and see if we can give the agents the ability to solve customer queries in more complex ways themselves.”transcript, stated

Tradeoffs

How much to emphasize AI during the transformation.

Chose customer and agent benefits first over AI hype as the headline.

Rationale: The team explicitly tempered AI expectations and treated AI as an enabler, not the solution itself.

“AI isn't going to solve world's problem like, on its own. It's really an enabler and we really want to focus on the benefits that it would give to the customers and our agents”transcript, stated

How to build for uncertain contact-center futures.

Chose a flexible foundation that can extend over repeating expensive transformations whenever AI capabilities change.

Rationale: So Energy did not want to stay on a transformation flywheel as technology evolved.

“you invest enough that you allow it to extend for the future, and that's really important because you don't really want to do another transformation or keep on doing a transformation every single time.”transcript, stated

Gotchas (what bit them)

Fragmented channels forced customers and agents to rebuild context manually.

Root cause: Voice, chat, and email were separate platforms with no shared customer story.

Resolution: So Energy moved to a single context across channels with case management and tasks.

“Customers were contacting us across channels and when they went from one channel to another, they really had to retell the story because there was no context between those channels.”transcript, stated

Migrating a broken process preserves the same problems.

Root cause: A new contact-center platform does not automatically fix poor workflows or unclear objectives.

Resolution: The team advises understanding current problems and co-creating the solution before migration.

“There's no point in doing the same processes that might be broken and you take 'em to a new system because you're gonna still have the same problems.”transcript, stated

Early rollout performance can dip before benefits appear.

Root cause: People and process teething issues emerged after the June rollout.

Resolution: So Energy diagnosed concerns, tweaked workflows, and turned performance around within months.

“our performance went down a bit, but we were confident in not just the platform but also our ability to turn that around, because we knew there was teething issues”transcript, stated

Evaluation

So Energy measures operational and customer outcomes such as wait times, channel mix, CSAT, Trustpilot rating, and first-time query resolution rather than vanity AI metrics.

Metrics: call wait time · email response performance · chat volume share · CSAT · Trustpilot rating · first-time resolution

“Don't get trapped in the vanity metrics. Focus on a few and we focus on CSAT and getting customer queries right first time.”transcript, stated

Outcomes

Call wait times fell by about 33%.
Chat grew from less than 1% to 15% of contact volume.
CSAT and Trustpilot improved after rollout.

The implementation served about 300,000 customers and became the foundation for future self-service and agentic tooling.

“we've seen remarkable performance improvements, especially in reduced call-wait times. It's about 33%, but we expect that to even get better over the next few months.”transcript, stated

“before our chat volumes were pretty much negligible but increased that channel to 15% from like pretty much less than 1%, all at the same time.”transcript, stated