AI-Powered Contact Center Transformation for Student Support Services

anthology · Education · 2024 · 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: Higher-education student support

Anthology needed to handle seasonal spikes in student-support demand, repetitive questions, and reliability issues while improving accuracy and reducing dependence on large temporary hiring waves.

Constraints:

Peak back-to-school periods drive extreme seasonality
Student data requires compliant technology
Password resets and financial-aid questions can affect student persistence

Architecture

Anthology migrated student-support operations to Amazon Connect in phased releases, combining voice/chat virtual agents, student verification, human handoff with AI summaries, agent assist, Contact Lens interaction analytics, and automated/hybrid evaluations. The same AI-agent infrastructure supports both student self-service and human-agent guidance during seasonal enrollment spikes.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  amazon-connect-core["amazon-connect-core"]
  virtual-student-agents["virtual-student-agents"]
  student-verification["student-verification"]
  human-agent-handoff["human-agent-handoff"]
  agent-assist-panel["agent-assist-panel"]
  contact-lens-analytics["contact-lens-analytics"]
  automated-evaluations["automated-evaluations"]
  release-train["release-train"]
  amazon-connect-core --> virtual-student-agents
  amazon-connect-core --> human-agent-handoff
  amazon-connect-core --> contact-lens-analytics
  virtual-student-agents --> student-verification
  student-verification --> human-agent-handoff
  human-agent-handoff --> agent-assist-panel
  contact-lens-analytics --> automated-evaluations
  release-train --> amazon-connect-core

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

Components

Component	Role	Interacts with
`amazon-connect-core`	Cloud contact-center foundation for Anthology's voice, chat, routing, interaction analytics, and AI capabilities.	`virtual-student-agents`, `human-agent-handoff`, `contact-lens-analytics`
`virtual-student-agents`	Self-service AI agents for repeatable student questions such as password reset, login, refund status, and registration support.	`amazon-connect-core`, `student-verification`
`student-verification`	Identity verification layer using student information before the virtual or human agent handles account-specific requests.	`virtual-student-agents`, `human-agent-handoff`
`human-agent-handoff`	Transfers unresolved interactions to a person with the prior AI context so students do not restart from zero.	`virtual-student-agents`, `agent-assist-panel`
`agent-assist-panel`	Provides real-time guides and knowledge to human agents during student-support interactions.	`human-agent-handoff`
`contact-lens-analytics`	Summarizes interactions, analyzes sentiment, records transcripts, and supports automated and hybrid quality evaluations.	`amazon-connect-core`, `automated-evaluations`
`automated-evaluations`	Reviews interactions at full coverage with AI-generated scorecards while allowing supervisor adjustment.	`contact-lens-analytics`
`release-train`	Phased delivery plan that launched core capabilities before peak season and then progressively added AI features.	`amazon-connect-core`

Data flow

A student reaches Anthology through voice or chat, where a virtual agent handles repeatable needs such as password or financial-aid status questions and verifies identity. If the request needs a person, the conversation and summary transfer to a live agent, who receives relevant guides and side-panel help. Contact Lens captures transcripts, sentiment, summaries, and scorecards so supervisors can review far more interactions than the old sampling process.

“Then we progressively introduced AI, such as Contact Lens. And we started, you know, with that we were able to do student self-serve, agent assist, as well as interaction insights.”transcript, stated

“the same tech being used to develop AI agents for self-service as it is AI agents for human agent assistance on the right hand panel.”transcript, stated

Why this architecture

Anthology needed reliability and repeatable answers during extreme education seasonality, so it avoided a one-time big-bang AI launch and phased Amazon Connect capabilities into production before peak. Virtual agents absorb repetitive questions, while handoff and agent assist keep humans available for complex student-support work.

“we're gonna do release one on July 1st. We're gonna do release two on July 22nd, we actually have our fourth release in December.”transcript, stated

“The seasonal hiring was another one. We thought a success measurement would be after the launch. If we can reduce our reliance on seasonal hiring, not eliminate it, but just reduce it”transcript, stated

Models used: Amazon Connect AI features · Contact Lens

Scale

Anthology supports 8 million inbound and outbound student interactions per year.
“And we support student experiences, 8 million inbound and outbound per year.”transcript, stated
Peak season occurs in July, August, September, and January.
“there's a huge spike over the summer, July, August, and September when students are going back to school. And then also in January.”transcript, stated
Wait times dropped by 50% during the busiest month of the year.
“We've reduced our wait times by 50% during our busiest month of the year.”transcript, stated
Accuracy improved by 14 points after introducing virtual agents and human-agent guidance.
“with virtual agents, we're seeing 14 point increase on accuracy.”transcript, stated
Attrition dropped by 10% after going live.
“we were able to reduce attrition by 10% just by going live.”transcript, stated
Back-to-school unplanned outages fell from 12 to two.
“every back to school we would have 12 unplanned outages in our busiest months of July, August and September because of the load and this back to school we had two.”transcript, stated

Alternatives considered (roads not taken)

Launching every Amazon Connect and AI capability at once

Rejected because: The team moved to a release train so core service could go live before peak while AI capabilities continued to roll out safely.

“we're gonna do release one on July 1st. We're gonna do release two on July 22nd, we actually have our fourth release in December.”transcript, stated

Manual quality review of roughly 1% of interactions

Rejected because: Contact Lens allowed AI-assisted evaluation of 100% of interactions, changing QA from sampling to full coverage.

“our aspiration was to be able to review 1% of all the interactions before. I mean, now with the efficiency”transcript, stated

“Oh, we can do 100% now. And you can set up rules in Contact Lens to be able to say”transcript, stated

Patterns

`fallback-degradation` (customer-support)

Every interaction starts with voice/chat automation, but humans remain behind it for more complex student questions.

“every interaction starts with voice and chat powered through Connect, and then we have humans behind that to be able to support students for more complex questions that need to be resolved.”transcript, stated

`human-in-the-loop-gate` (agent-assist)

Automated self-service handles repeatable tasks, while trained teams coach students through financial aid and admissions.

“we can leverage our teams that know financial aid really well or admissions really well and coach that student through that process.”transcript, stated

`observability-tracing` (quality-management)

Contact Lens summaries, transcripts, sentiment analysis, and automated evaluation let supervisors review all interactions instead of tiny samples.

“With Amazon Connect with AI you can do 100% automated agent evaluations using what you have here with the sentiment analysis and these questions that you can help with prompts.”transcript, stated

`canary-rollout` (deployment)

Anthology switched from a big-bang launch plan to a series of releases during the peak season rollout.

“we should actually do this in a series of releases. And that's where we ended up to say, we're gonna do release one on July 1st. We're gonna do release two on July 22nd”transcript, stated

Tradeoffs

How to launch during a critical peak period.

Chose phased releases starting July 1 over shipping every capability at once.

Rationale: The phased model created a framework for learning from each release while still going live before back-to-school demand.

“I thought this was such a good plan, but as we went through it, you know, the thought was, okay, we should actually do this in a series of releases.”transcript, stated

What to optimize in student support.

Chose consistent self-service for repeatable questions and human expertise for complex guidance over using humans for every password reset and routine status question.

Rationale: Repeatable questions can be handled consistently, freeing specialized staff for financial aid and admissions coaching.

“they're 20 repeatable questions. How can we leverage technology to be able to answer those in alternative ways for users and make sure that they're handled consistently”transcript, stated

Gotchas (what bit them)

A lift-and-shift business case failed to win approval for years.

Root cause: The proposal focused on moving CRM/IVR infrastructure to the cloud rather than transforming the business model.

Resolution: Anthology reframed the work as a business transformation with change-management conversations.

“it was always presented as a lift and shift. And so every time we went through it, while the slides may have changed or our recommendations changed”transcript, stated

Seasonal hiring made quality and innovation hard to sustain.

Root cause: The organization doubled staff for summer peaks and spent much of the year just getting through the back-to-school period.

Resolution: The new model handles repeatable questions with technology and keeps expert teams focused on complex support.

“if we have a staff of a thousand today, we hire over a thousand every summer to be able to support student inquiries.”transcript, stated

Old systems created peak-period outages and manual failover work.

Root cause: Load during July, August, and September exceeded the reliability of the previous application stack.

Resolution: The migration reduced peak-period unplanned outages and removed much of the disruption.

“every back to school we would have 12 unplanned outages in our busiest months of July, August and September because of the load and this back to school we had two.”transcript, stated

Evaluation

Anthology tracks wait time, accuracy, attrition, outages, handle time, productivity, and automated interaction-review coverage.

Metrics: wait time · accuracy · attrition · unplanned outages · handle time · 100% automated evaluations

“we measure everything in 15 minute increments. So when we started this initiative, we wanted to make sure we were clear on our objectives.”transcript, implied

“our aspiration was to be able to review 1% of all the interactions before. I mean, now with the efficiency of being able to do this”transcript, stated

Outcomes

Wait times fell by 50% during the busiest month.
Accuracy increased by 14 points.
Attrition fell by 10%.
Peak-period unplanned outages dropped from 12 to 2.
Interaction review coverage rose from an aspiration of 1% to 100%.

Anthology supports about 8 million inbound and outbound student interactions per year.

3 supporting quotes

“We've reduced our wait times by 50% during our busiest month of the year.”transcript, stated

“what we've actually seen with human agents now with virtual agents, we're seeing 14 point increase on accuracy.”transcript, stated

“Oh, we can do 100% now. And you can set up rules in Contact Lens”transcript, stated