Most "AI for small business" articles are written by people who have never shipped an automation in production. They list the tools, not the flows. Here are the five we install most often for Israeli SMBs — each one ships in under a week, each one survives the founder going on vacation, each one shows up in the P&L within 60 days.
1. Inbox triage to Slack — 2-3 days to ship.
An n8n flow listens to your hello@ inbox. Every incoming email passes through a GPT classifier that tags it as urgent / needs reply / FYI / spam. Urgent items go straight to a Slack channel with a one-click "draft reply" button that uses the company tone-of-voice prompt to write a response. Spam disappears. The founder reads Slack twice a day instead of email every five minutes. Typical save: 2-3 hours/day for the founder of a 10-30 person SMB.
2. Lead-to-CRM with auto-enrichment — 2 days to ship.
Every website form submission flows through a webhook into n8n, which calls a Clearbit-style API to enrich the lead (company size, industry, LinkedIn). The enriched record is written to Greeninvoice or Pipedrive with a tag (cold / warm / hot) based on a heuristic. Sales gets a Slack ping for hot leads with the enrichment summary attached. No more "I will Google them" before the discovery call. Typical save: 15 minutes per qualified lead, plus a real qualification rate jump.
3. Document drafting from briefs — 3-4 days to ship.
A simple form ("project name, scope items, tier, addons") triggers a flow that drafts the proposal, the SOW, and the first invoice — using a templated GPT prompt that injects your standard terms, payment schedule, and discount rules. The drafts land in your Drive as Google Docs ready for review. Hebrew + English, both directions. Typical save: 90 minutes per proposal. If you write 4 proposals a month, that is a billable day back, every month.
4. RAG chatbot on your knowledge base — 4-5 days to ship.
Point a Firecrawl crawler at your Notion, Confluence, or Google Drive. The output goes into Supabase + pgvector. A small Next.js chat UI on /support answers customer + employee questions from the indexed content, with citations linking back to the source doc. Hebrew works as well as English if you embed with the right model (multilingual-e5-large). Typical save: 30-40% of incoming support tickets answered without a human, with the rest going to a human pre-tagged with the relevant doc.
5. Weekly KPI digest to WhatsApp — 1-2 days to ship.
A scheduled flow pulls revenue, expenses, new leads, and project status from your accounting + CRM + project tools every Sunday night. GPT writes a 200-word summary in your voice ("this week, revenue was X, the standout was Y, the risk is Z"). It lands on the founder WhatsApp at 7am Monday. No dashboards to log into. No "let me check the numbers". Decisions get made on Monday morning instead of Wednesday afternoon. Typical save: less measurable than the others, but the time-to-decision drop is the real ROI.
The order matters.
Do not start with the RAG chatbot. It is the most impressive demo but the longest to build and the hardest to maintain. Start with #1 (inbox triage) — it has the shortest path to "felt different at the founder level" and validates the team's appetite for AI tooling. Once #1 is loved, #2 + #3 follow naturally because they free up sales + ops time. #4 and #5 are quarter-2 work.
What to skip.
Skip "ChatGPT seats for everyone". Skip the AI consultancy that wants three months of discovery before recommending anything. Skip the Custom GPTs that need 40 prompt-engineering hours and break the moment OpenAI ships a new model. Skip the AI agent that "autonomously runs your business" — those demos are real but the production version is six months away.
AI for SMBs in 2026 is not strategy work. It is plumbing work. Ship five pipes that move real water, and the business feels different inside 90 days.