A custom AI agent for a SaaS startup

A young tech startup (we'll call it NovaTech to preserve anonymity) builds a SaaS platform for managing digital marketing campaigns.

The company is growing fast, but the team stays very small. This leads to a situation common among startups scaling up:

• lots of repetitive operational work
• customer support handled manually
• slow lead qualification
• a backlog of emails and requests
• meetings with no structured summary

The result is that the team's time gets absorbed by day-to-day operations instead of strategic work and product development.

Meanwhile, the market is changing fast. The fastest-growing startups are adopting AI as an operational and competitive lever.

According to Salesforce's Small and Medium Business Trends report, growing startups invest in artificial intelligence 1.8 times more than declining companies.

NovaTech needed to:

• reduce the team's operational workload
• speed up response times to customers
• improve lead management
• free up time for growth activities

In particular, the team identified five critical areas:

Customer support

About 70% of incoming requests were about frequently asked questions or recurring issues.

Lead qualification

The sales team manually reviewed every contact coming from marketing campaigns.

Meetings and operational tasks

Meetings generated long follow-up work: manual summarising and task creation.

Email management

The email backlog was growing fast.

HR and onboarding tasks

Many internal requests were repetitive and ate into the team's time.

The goal, then, was to design a system able to automate these tasks without adding organisational complexity.

QMates designed a modular AI agent built on a Large Language Model (LLM) with a RAG (retrieval-augmented generation) architecture.

The system is made up of several specialised sub-agents, orchestrated by a main agent that coordinates operations and manages integration with the company's systems through secure APIs.

1. Customer assistant

A multilingual chatbot and voicebot answers the most frequent questions, tracks order status and handles support requests. The system runs 24/7 and brings in a human operator only when needed.

2. Lead qualifier

An agent automatically analyses new contacts coming from marketing campaigns.

• enriches the data with public information
• assigns a priority score
• creates tasks in the CRM
• notifies the sales team

3. Meeting summariser

A module integrated with Google Meet automatically records meetings. The agent:

• generates the transcript
• summarises the key points
• extracts action items
• sends tasks to Notion or Asana

4. Content & email generator

A sub-agent drafts emails, marketing content and social posts using the company's knowledge base. The system learns from the company's historical tone of voice to keep the style consistent.

5. HR assistant

The agent supports the onboarding and recruiting process:

• answers candidates' questions
• collects documents
• schedules interviews
• helps screen CVs

6. Email triage bot

The system analyses incoming emails and:

• classifies requests
• suggests automatic replies
• routes messages to the right department

This drastically cuts the operational backlog.

The project was developed in three main phases.

1. Analysis and data collection

QMates analysed the company's workflows to identify the most repetitive tasks. The following were collected and pre-processed:

• FAQs
• internal documentation
• historical emails
• sales scripts

This data was indexed for use through RAG techniques.

2. Agent development

Each sub-agent was trained on specific tasks using fine-tuned GPT-4 models. Integrations were built with:

• CRM
• customer support systems
• collaboration tools

The main agent coordinates activities and manages security and privacy.

3. Testing and iteration

The system was tested in pilot mode. The team gave continuous feedback to improve:

• response accuracy
• system integration
• user experience

Beyond the quantitative results, the project had a significant impact on the organisation.

Higher customer satisfaction

Faster response times and round-the-clock availability improved the customer experience.

Lower operating costs

Automation avoided new hires for operational support. The system frees up around 100 hours a month across all teams.

Sharper strategic focus

The team can focus on:

• product development
• marketing experimentation
• data analysis

NovaTech's experience shows that a well-designed AI agent can radically transform how a startup operates.

It's not just about automating tasks, but about redesigning how the team works.

For startups that want to adopt similar systems, it's essential to:

• identify the most costly processes
• integrate AI with existing company tools
• work on data quality
• take an iterative approach to improvement

When these elements line up, AI becomes a real lever for scaling without a proportional rise in costs.