colorMEMBA Professional Thesis · Topic proposal

Case study · Vietnam · Education SME

The impact of an internally developed AI system on operational efficiency and cost reduction in student recruitment

A case study of colorME

Setting
Creative & design school

Function
Student recruitment

Design
Single embedded case · mixed methods

colorME01 · Context

Context

Recruitment is the engine of a training business.

colorME02 · Problem

Problem statement

High cost, slow cycles, weak signals.

Marketing

Spend follows the wrong signal

Ad platforms optimised for page views and "conversions" that were mostly not real leads.

reported

real leads

Operations

Lead source is lost

Leads arrive from ads, Messenger and the website, with no link from click to enrolment. CAC per channel is unknown.

Content

Production bottleneck

Landing pages, posts, visuals and campaigns queue behind a small team, slowing every launch.

SMEs cannot afford enterprise MarTech. Can an AI system built in-house close the gap?

colorME03 · The intervention

The intervention · from Oct 2026

An internal AI workforce.

26 specialised agents, coordinated by an Orchestrator. Staff request work in plain chat through Megamind. One human approves anything that spends, sends or ships.

26 agents 1 approver rollback on every change

colorME04 · In practice

Megamind · the front door

Any staff member can ask the AI for work.

Report a bug, request a feature, ask for an email, an image or a video. Megamind turns the chat into a structured request, routes it, and reports back when it is done.

Chat in the staff app or web

Screened, approved, built by an agent

Requester notified: "done"

manage · Megamind

500+ requests in 3 weeks

colorME04 · What the AI team shipped

Output · live on colorme.vn

Built and shipped by the agents.

Every screen below is in production and serves students today. None of it existed in this form before October.

colorME05 · Where AI acts

AI agents mapped onto the recruitment funnel

Every stage has an AI teammate.

colorME06 · Research questions

Objectives & research questions

Four questions.

1

Efficiency

How does the AI system change the speed and throughput of recruitment operations?

Lead response time · launch time · output per staff

2

Cost

What is its effect on recruitment costs?

Cost per lead · cost per enrolment (CAC) · cost per content asset

3

Mechanisms

Through which mechanisms do these effects occur?

Automation · decision support · attribution · staff self-service

4

Conditions

Which organisational conditions enable or limit the impact?

Human-in-the-loop governance · adoption · skills · risk controls

colorME07 · Conceptual framework

Conceptual framework · RBV · Dynamic Capabilities · TAM

From capability to outcome.

colorME08 · Methodology

Methodology

Single embedded case, mixed methods.

Quantitative

Time-series & cost model

GA4 · Search Console · Ads platforms · CRM leads & registrations · AI running costs

Qualitative

12–15 interviews

Marketing, admissions/sales, management. Thematic analysis of mechanisms & conditions.

Survey & logs

Adoption (TAM)

Staff questionnaire + 500+ request logs and agent work logs as behavioural data.

colorME09 · Measures

Key measures

What we will measure.

Dimension	Indicators	Source
Efficiency	lead response timerequest → delivery timepage / campaign launch time	CRM · request logs · deploy logs
Throughput	assets shipped / weekrequests done / week	Agent logs · CMS
Acquisition	organic sessionssearch clicksleads per channel	GA4 · Search Console · CRM
Cost	cost per leadCACcost per assetAI running cost	Ads platforms · finance
Adoption	active requesterssatisfactionperceived usefulness	Request logs · survey

colorME10 · Early evidence

Early evidence · first 3 weeks

The signal is already visible.

Google Search Console · clicks per day · colorme.vn

GA4 · organic sessions / day

3,084+44% in 7 days

Search clicks / day (19/10)

2,0314-month high

GA4 · organic sessions · 15–21/10

colorME10 · Evidence · demand

Acquisition · Google Search Console & GA4

More people find colorME on their own.

Average search clicks per day, by month

Oct = 1–19 Oct · AI system live from 1 Oct · dashed line = August level

Where sessions come from

GA4 · 21/10 · 6,856 sessions

colorME10 · Evidence · operations

Internal requests via Megamind · 5–18 Oct

Staff turned into a product team.

Requests by area

405 requests from 36 staff in two weeks

Outcome

status on 18 Oct

colorME10 · Evidence · measurement

New measurement layer

The funnel is now measured daily.

Before October these numbers were not collected in one place. The AI system now reports sign-ups, course registrations and consultation requests every morning. This becomes the post-intervention series for the thesis.

Recruitment funnel, per day

16–21 Oct 2026 · automated daily report

colorME11 · Contribution & limits

Expected contribution & boundaries

Why it matters, honestly.

Academic

Evidence on agentic AI in an SME

Education sector, emerging market (Vietnam). Extends RBV / dynamic capabilities to internally built AI.

Managerial

A replicable SME playbook

KPI set to evaluate AI ROI in recruitment, and a governance model: approval, rollback, risk screening.

Limitations → mitigation

Single case, short window, insider

Analytical generalisation · time-series controls & triangulation · reflexive journal & audit trail.

colorMEThank you

"Can a small company build its own AI workforce, and does it make recruitment faster and cheaper?"

Thank you.
Questions welcome.