Most companies start their AI implementation with the question "which model should we choose?" - and that's the first mistake. AI isn't the goal; it's a tool for a specific process. This guide is for a board that wants to quantify the potential and start with something that pays back - instead of funding yet another "AI strategy".
Start with the process, not the technology
Asking "which AI model?" is like asking "which engine?" without knowing whether you're building a motorbike or a lorry. Process first, then the tool. A good first AI project is one that is repeatable, measurable and currently eats up a lot of people's time.
For us, the most common "first process" is customer service and store search - because the effect shows up in the numbers within weeks, not after a year.
How to choose the first process
Hold your processes up against four criteria - the more "yes" answers, the better the candidate:
- Repeatability - it happens every day, many times over (e.g. the same customer questions).
- Measurability - you can count "before and after" (time, conversion, number of tickets).
- Time cost - it currently eats up team hours that could be moved to more valuable work.
- Data availability - you have material for the AI to draw on (catalogue, FAQs, ticket history).
Quantify the potential before you spend a single pound
Before any project gets going, do a simple back-of-the-envelope calculation: number of events per month × time per event × hourly cost = the pool AI can recover. Then add the sales effect (e.g. service outside working hours).
At Ciepło24.pl the AI assistant closes 88% of conversations without a human agent and handles 43% of enquiries outside working hours. Those are real team hours and real orders that used to slip away - quantifiable before the implementation even starts.
A pilot instead of a grand strategy
You don't need a year-long "AI transformation". You need one process, implemented in 2-4 weeks, with a clear measure of success. A pilot gives you data, and data - not slide decks - is the best argument for scaling.
If the pilot pays back, you expand. If it doesn't - you've lost weeks, not quarters. That's risk management, not a leap into the deep end.
FAQ
Does AI require us to replace our systems?
Usually not. Solutions like ecChat or ecSearch plug into your existing store via API - a pilot doesn't require rebuilding your infrastructure.
How long until the first meaningful result?
With a well-chosen process, the first measurable results appear 2-6 weeks after the pilot goes live.
Is our data safe?
Yes - we keep data in the EU, under a data processing agreement and with retention controls. Models don't learn from your customers without your consent.