Intelligent ERP.
Not ERP + hype.
Your ERP already knows your business. AI makes it act on that knowledge — reading documents, answering customers, preparing decisions, automating approvals. I build these systems inside Odoo, where the ROI is measurable.
What "AI inside your ERP" actually means
Six patterns, each wired into Odoo's records and permissions — not chatbots bolted onto broken processes.
AI agents & assistants
Agents that read ERP context and prepare work for approval — triage, drafts, summaries, follow-ups. Built on modern standards (MCP), with audit trails.
Document intelligence & OCR
Vendor bills, POs, delivery notes, bank statements → validated Odoo records. Auto-posted above a confidence threshold; one-click review below it.
Conversational service
Chatbots and WhatsApp flows where your customers already are — subscriptions, requests, payments, status. 4,000+ customers in production.
Approval & workflow automation
Multi-level approvals, exception routing, and hand-offs that run themselves — humans decide, the system does the chasing.
Predictive operations
Reorder signals, demand patterns, maintenance windows — prediction added only after data discipline makes it trustworthy.
AI reporting & insight
Ask questions in plain language, get answers from your live ERP data — plus scheduled narrative summaries leadership actually reads.
AI by business function
Sales
- Quote drafts from emails & RFQs
- Lead triage and enrichment
- Customer history briefs before every call
Inventory
- Delivery-note OCR → instant receipts
- Reorder signals from real demand
- Discrepancy detection before stocktakes
Manufacturing
- Work-order summaries & shift handovers
- Quality-check capture from the floor
- Maintenance-window prediction
Where AI pays — and where it's just expensive
AI multiplies whatever your ERP already is. Clean processes become leverage; chaos becomes faster chaos. I hold clients to the sequence.
Worth your money
- Documents your team retypes into the ERP
- Questions customers ask on repeat
- Approvals that wait on someone noticing
- Reports that take a day to assemble
Not yet — fix the base first
- Forecasting on top of unreliable inventory data
- Chatbots bolted onto broken processes
- "AI strategy" that can't name a workflow
- Autonomy without audit trails
Fair questions
Do we need AI, or “just” automation?
Often the honest answer is deterministic automation first — rules, integrations, and workflows with no model involved. AI earns its place where judgment is repetitive: documents, conversations, triage. The opportunity map tells you which is which.
Which AI models and tools do you work with?
Model-pragmatic: OpenAI and Anthropic (Claude) APIs, plus workflow tooling and MCP-based integrations where they fit. The architecture keeps you portable — models change; your workflows shouldn’t break when they do.
What about data privacy?
Automation designs specify exactly what data leaves your system, to which processor, under what terms — with redaction where needed. For sensitive workflows, extraction can run on EU-hosted or self-hosted models.
Which workflow would pay for automation first?
Bring your process list. You leave with an honest opportunity map — ranked by ROI, including the "not worth it" column.