# AI systems

> Messages sorted, calls summarised in the CRM and leads ranked, with tests and a measured cost.

Source: https://fcgsystems.es/en/services/applied-ai-systems/ · FCG Systems

**Applied AI for your training business, measured by what it changes.**

We put AI into your training business where it genuinely saves hours: student and lead messages sorted, calls summarised in the CRM, leads ranked before they’re assigned and an assistant that answers from your documentation. Every use case starts from a metric, is tested on real situations and measured in production.

AI isn’t a project on its own. It’s one more component of the system, worth using when it does something specific better, faster or cheaper than the alternative. If a simple rule solves the problem, we’ll say so instead of selling you a model.

If you already tried AI and it never got past the demo, the usual reason is that nobody defined what it had to improve. Here that comes first, with clear limits, human approval where a mistake is expensive and a cost per run you can see.

## Signs you need it

- Someone reads and sorts hundreds of messages a day
- Call notes never make it into the CRM
- The team keeps asking what’s already documented
- You tried AI and it never got past the demo
- You can’t tell which assistant answers are right
- Leads are prioritised by gut feeling

## Who it is for

- **Founders with a stretched team:** Student messages, admissions and calls that someone currently reads, summarises and sorts by hand. You get hours back without hiring.
- **CMOs and sales teams with lead volume:** Leads ranked on criteria you can review, every call summarised in the CRM and context for the closer before they pick up the phone.
- **CTOs and technical owners:** AI in production with logs, cost limits, test cases and the option to swap models without rebuilding the integration.
- **Agencies with training clients:** AI assistants and workflows for your clients, built and maintained without growing your technical headcount.

## What we build

- **Classification and routing:** Messages, forms and tickets tagged and sent to the right person.
- **Summaries and extraction:** Calls, meetings and documents turned into tidy data inside your CRM.
- **Lead scoring:** Lead and opportunity prioritisation based on explicit criteria you can review.
- **Agents with tools:** Agents that query your CRM, calendar or database and carry out scoped steps, with human approval wherever a mistake is expensive.
- **Assistants and chatbots:** Answers grounded in your documentation for the team or for students, with cited sources, defined limits and a handoff to a person.
- **Conversation analysis:** Patterns across chats and sales calls: objections, questions and loss reasons.
- **Human-in-the-loop workflows:** AI suggests, a person approves, wherever a mistake is expensive.

## Chatbot, AI agent or AI automation.

Three ways of using AI that often get mixed up. Choosing well saves you from paying for an agent where a workflow was enough, or a chatbot that can’t actually do anything.

| Criterion | Chatbot | AI agent | AI automation |
|---|---|---|---|
| What it does | Answers questions inside a conversation | Decides the steps and uses tools (CRM, calendar, database) to reach a goal | Runs a fixed workflow with an AI step: classify, summarise or extract data |
| When it fits | Frequent student or lead questions answered in your documentation | Multi-step tasks with real variation from case to case | Repetitive, high-volume tasks with an output you can check |
| Main risk | Made-up answers if it isn’t tied to sources | Wrong actions if it has no limits or approval | Silent errors without test cases or logs |
| Control we add | Answers with sources and a handoff to a person | Scoped tools, stopping conditions and human approval on sensitive steps | Output validation, retries and alerts |

## How we work

1. **Discovery:** We pick one process, measure how it works today and set the metric that has to improve.
2. **Design:** We decide what the model does, what a rule handles and where a person signs off.
3. **Build:** It plugs into the tools your team already uses: CRM, WhatsApp, email or support, all versioned.
4. **Validation:** Real test cases and edge cases before anything touches production.
5. **Deployment:** Production with logs, cost limits and a fallback if the model fails.
6. **Improvement:** We review accuracy, usage and cost on real data and adjust what underperforms.

## Deliverables

- Defined use case and success metric
- Integration into your team’s tools
- Versioned prompts and models, in your name
- Test cases to control quality
- Usage, cost and accuracy metrics
- Documentation

## What changes

- **Time:** Repetitive reading, sorting and note-taking come off your team’s plate.
- **Quality:** Every change is evaluated against test cases before it reaches production.
- **Known cost:** You know what each run costs and what value it delivers, with no surprises on the bill.

## Format

- Format: AI Workflow · fixed-scope project
- Duration: Depends on the use case, set after discovery
- Typical tools: OpenAI, Anthropic, Vercel AI SDK, TypeScript, Python, n8n, Postgres, pgvector

## FAQ

### What exactly do you do with AI in a training business?

We connect it to your tools to solve specific tasks: sorting messages, summarising calls into the CRM, scoring leads or answering from your documentation. We test it on real cases and measure what changes in time, quality or cost. We don’t sell it as an end in itself.

### Do you do AI consulting or implementation too?

Both, in that order. First we define the use case, the success metric and whether AI is the right option; then we build it, deploy it and maintain it. If a rule or a simple automation solves the problem, we’ll tell you even if it costs us the project.

### Which processes in my business can be automated with AI?

The ones that involve reading, interpreting or summarising information at volume: student and lead messages, documents, sales calls, lead prioritisation or questions already answered in your documentation. Decisions with legal, financial or student impact always keep human review.

### How much does it cost and when does it pay back?

It depends on the process, the integrations and the volume, so we don’t publish fixed rates. After discovery you get a fixed scope with deliverables and timelines. Model usage is measured per run: you see what each task costs and compare it with the manual hours it replaces.

### Which AI is best for my business?

No single model is best at everything. We work with OpenAI and Anthropic, among others, and choose per task by running them against the same test cases for accuracy, cost and speed. The system is designed so the model can be swapped without rebuilding the integration, so you’re not locked into one provider.

### We already tried AI and it never got past the demo. What’s different?

Here AI handles a specific task inside a process that already exists, and you can measure the before and after. We don’t build demos or decorative chatbots. If a simple rule solves the problem, we use the rule.

### Is our data used to train models?

We use providers and settings that don’t train on your data, and we only send the model what it needs. When the use case requires it, data is anonymised before processing.

### How do you stop the AI from getting things wrong?

With real test cases before deployment, human review on sensitive steps and accuracy metrics in production. A model can fail: the system has to catch it and have a fallback.

## Contact

Email hola@fcgsystems.es or use the form at https://fcgsystems.es/en/contact/. Reply within one working day.