Summary
Many SMEs that start AI projects see no results in the first few months. The reason is almost never a lack of technology — it's data fragmentation.
When we talk to technology teams during audits, we find an uncomfortable truth: ChatGPT is not the problem. Data silos are.
The typical scenario
An SME decides to "implement AI". It buys access to ChatGPT Enterprise, assigns a budget, and expects teams to become more productive.
What actually happens:
- Data lives in 4–8 different systems (ERP, CRM, Sheets, Mail, Drive, legacy database)
- There is no unified data dictionary
- Every question to the model requires manual context-switching
- The team doesn't trust the answers because they don't know whether the model has access to the truth
Result: AI improves productivity by maybe 10%. It gets used as a desktop ChatGPT, not as a strategic tool.
Why agents fail without architecture
An AI agent is a system that:
- Reads data from your systems
- Makes decisions based on that data
- Acts (writes to your CRM, generates reports, runs workflows)
But without a clear data architecture, agents:
- Don't know which data is reliable → they make decisions on outdated or contradictory data
- Generate more work, not less → someone has to verify and correct every decision
- Create technical debt → every decision they make gets encoded into behaviors that later become impossible to change
What "architecture before agents" means
We're not talking about 6 months of expensive consulting. We're talking about:
- Data consolidation — connecting your main systems into a unified data layer
- Business dictionary — documenting what each field means in your context
- Synchronization flows — keeping data fresh in real time
- Security and compliance — knowing who accesses what
With this in place, agents work. Without it, they are expensive and unreliable.
The alternative that works
Start small and architecturally correct:
- Identify 1–2 processes where an agent adds value (not: "automate everything")
- Make sure the required data is consolidated
- Implement the agent in 2–4 weeks
- Measure real impact (time saved, quality, faster decisions)
- Expand to other processes
Illustrative cases:
- Logistics SME: an agent that reads orders, ERP and schedules → generates optimized routes
- Services company: an agent that consolidates billable hours from 3 systems → automatically updates profitability reports
- B2B startup: an agent that reads inbound leads and the CRM → prioritizes the leads most likely to close
Next steps
If your SME has:
- Multiple disconnected systems
- Teams losing time consolidating data manually
- Curiosity about AI but caution about the "hype"
It's time to talk: a technical audit (90 minutes, free of charge) lets us map where the silos are and which agents could genuinely work in your business.
We don't sell expensive implementations. We identify the highest point of leverage and execute from there.
About the author
Teo García Pascual is CEO and CTO of Nodalix Consulting. As the company's sole founder, he designs the data and product architecture of every AI portal he deploys. LinkedIn · contacto@nodalix.es