AI without context is an expensive parrot — why AI is not plug-and-play

The problem

In 2026, companies arrive saying “we want to add AI”. We ask: where is your customer data? “In WhatsApp.” What about order history? “In a spreadsheet Antonia maintains.” Do you have a documented process for handling complaints? “No, everyone does it as best they can.”

There is the problem. AI is not a plug. You do not arrive, connect it to the power supply and immediately create value. AI without clean data, defined processes and context is an expensive parrot: it repeats what you tell it, hallucinates what it does not know and costs €200–800 a month in tokens without returning anything useful.

The expensive parrot analogy

A parrot repeats phrases. Teach it insults and it insults. Teach it nothing and it stays silent or makes random noise. Generative AI does the same, with more words and more hallucination. Give it clean context — your catalogue, FAQs, ticket history and brand rules — and it responds like a senior team. Give it a dirty spreadsheet with three badly named columns and it returns a plausible, incorrect answer.

AI knows nothing about your business that you have not given it. If your business lives in WhatsApp and the heads of three people, AI has nothing to work with.

That is why, at SOM-OS, the first step is never “deploy an agent”. The first step is to audit, organise the data and design the architecture. After that, yes: AI starts producing real value.

The demo’s 5 scenarios

We have a demo with five real scenarios that we show every client before talking about a budget. Each shows what happens when AI has context and when it does not. With concrete numbers:

1. Customer service

Without context: the agent responds in 4 seconds but says “please contact our team for that”. With context — product catalogue, returns policy, opening hours — the agent resolves 68% of enquiries without human intervention. Average response time: 4 seconds, 24/7. A person takes 3–8 minutes and only during working hours.

2. Classify emails

Without context: the agent labels randomly and misses 40% of urgent messages. With context — your category taxonomy, labelled examples, priority rules — it classifies 500 emails in 90 seconds with 94% accuracy. A person takes 3–4 hours to do the same, with fatigue-related errors.

3. Respond to complaints

Without context: the agent gives generic apologies and enrages the customer. With context — customer history, compensation policy, brand tone — it drafts a response acknowledging the specific case, proposing a solution and closing the ticket without escalation in 72% of cases.

4. Detect stock issues

Without context: the agent does not know what stock is or where to look. With context — a connection to your inventory database, product thresholds, suppliers — it detects stockouts 48 hours before stock runs out and alerts the purchasing manager. A person checks stock once a week, if they remember.

5. Summarise calls

Without context: the agent invents parts of the conversation. With context — a clean transcript, summary template, fields to extract (agreements, next steps, risks) — it delivers an actionable summary in 30 seconds, ready to paste into the CRM. A person takes 15–20 minutes to summarise a 30-minute call.

Why the order matters

The right order for digitising a company is not “buy AI”. It is:

• Analyse. Understand what you do, with which data and where the bottlenecks are. This is the Chaos Map.
• Optimise. Organise data, clean up ineffective processes and unify separate tools. This is custom development.
• Scale with AI. Now deploy AI agents on clean data and defined processes. Each agent has context, rules and a measurable goal.

If you skip steps 1 and 2, step 3 is a parrot. An expensive one. It costs tokens every month and returns hallucinations. Do all three in order and AI works like another team member — one that does not sleep or get sick and scales to a thousand simultaneous conversations without hiring anyone.

AI does not replace processes. It relies on them. If there is no process, there is nothing to rely on.

What we tell someone who only wants “the AI”

If you arrive saying “I need a chatbot with ChatGPT” and your data is not connected, we tell you clearly: first we need to assemble the foundation. We do not sell you a parrot. We build the system that makes AI worthwhile — and explain why €200 a month in tokens without context is €200 thrown away.

To see an agent working with real context, look at Sando Capital: an agent qualifying financial leads with the product catalogue loaded, the right questions and fit rules. That is the difference between a parrot and a system.