"Automate your business" sounds like a mantra, but in practice 70% of automation projects fail: either the tool doesn't fit the process, no one maintains it after rollout, or the savings are smaller than the cost.
This playbook is our experience rolling out automation at 20+ companies in Kazakhstan. Here's what actually works.
Step 1. Which processes to automate first#
Simple formula:
Automation ROI = (hours × rate × 12 months) − implementation and support cost
Good first-wave candidates:
- Repeating - happens ≥10 times per week
- Predictable - clear rules (if-then)
- Painful - currently done by hand, with errors
- Measurable - you can count time/money before vs after
Bad candidates (don't start with these):
- Rare events (once a quarter)
- Lots of exceptions
- Require subjective judgement
- Touch many departments on day one
Step 2. Three levels of automation#
Level 1. No-code integrations (Zapier, n8n, Make)
Good for:
Action in system A → notification/record in system BExamples:
- New customer in Kaspi → CRM card + Telegram to a chat
- Payment in 1C → email to the client with the receipt
- Missed call → task assigned to the account manager
Cost: from $20/month, rollout in days.
When it stops being enough
No-code hits a wall past ~20 scenarios, or when you need complex branching logic, or custom API calls. Then - level 2.
Level 2. Custom Node.js / Python pipelines
When no-code becomes a spaghetti of 30 Zaps nobody understands anymore.
Move to managed code:
// Example: auto-invoice after payment
export async function handlePayment(payment: Payment) {
const invoice = await generateInvoice(payment);
await storeInDb(invoice);
await sendEmailWithPdf(payment.clientEmail, invoice);
await notifyTelegram(invoice);
await syncTo1C(invoice);
}Upsides:
- A single git repo - the whole business process is visible
- Tests, types, code review
- No limits on the number of runs
Cost: 2-6 weeks of dev + hosting.
Level 3. Your own platform
When you have 10+ processes, 5+ departments, and need an admin panel and roles. That's an internal SaaS.
Example - we built one for a logistics client: document flow, warehouse, bank integration, fuel accounting. Vendors like 1C + SAP didn't fit on price and flexibility.
Step 3. Success metrics#
Lock in a baseline before rollout:
| Metric | Before | Goal | How we measure |
|---|---|---|---|
| Order-processing time | 2 h | 15 min | CRM log |
| Document error rate | 8% | <1% | Monthly audit |
| Cost per order | ₸2,000 | ₸400 | Salary / order count |
| Client response time | 4 h | 30 min | First-response metric |
30 days after rollout - check. If the numbers aren't there, figure out why.
Step 4. Common mistakes#
Automating a bad process
Before rolling anything out, re-read the as-is process. If it's crooked - fix it first, then automate. Otherwise you'll just make mistakes faster.
Other gotchas:
- No rollout owner - the whole thing rots a month later
- No documentation - when someone leaves, no one can maintain it
- Skipping monitoring - you hear about outages from clients a week later
- Missing edge cases - "what if the customer clicks pay twice?"
Takeaway#
Automation isn't an IT project, it's a product project. You need to:
- Understand the process first
- Pick the right level of complexity (don't write code if Zapier is enough)
- Measure before and after
- Assign an owner and write documentation
At BAI Core we help with all three levels - from the first integrations to a fully-fledged internal platform.
Newsletter
New articles and case studies - every two weeks
No spam, no marketing blasts. Just engineering and real-world problems. Unsubscribe with one click.