When AI makes sense vs when workflow is enough
Technology is chosen last — after process map and ROI.
Workflow without AI
Rules, routes, integrations — when process is predictable: lead → CRM → task → proposal → send. Document approval chains. CRM–1C sync.
Tools: amoCRM robots, Bitrix24 BP, n8n, custom API. Cheaper, predictable, easier to maintain.
When AI is justified
Unstructured text: knowledge base answers, ticket classification, field extraction from arbitrary PDFs. Dialogs with variable client wording.
Prerequisite: structured data and process. AI without CRM and documents — a pilot that won't scale.
How to decide
Audit: calculate manual labor cost in hours × rate. Compare «workflow only» vs «workflow + AI». AI adds 20–40% to budget and needs MLOps.
Start with workflow. Add AI on a narrow slice with measurable ROI — L1 support, document search, proposal assembly from dialog.
Corporate perimeter
For company data — private LLM, YandexGPT or GigaChat in your contour. RAG with source citations. Guardrails and human escalation.
Don't send client data to public ChatGPT without DPA and policy.
Checklist
- Process map before technology choice
- ROI table for workflow vs workflow+AI
- Pilot on one slice, not «AI for whole company»
- Data in perimeter for corporate scenarios
Related pages
AI & Automation Audit
We'll audit, calculate ROI by scenario and honestly say whether AI is needed or workflow suffices.