Q1
We are not sure what to automate. Where do we start?
With the process audit. We sit with the team doing the work, write down every step and time it, then rank what we found by payback. It is normal to discover that the process everyone complains about is not the one worth automating first. You keep the map either way.
Q2
Does our data get sent to OpenAI or Google?
Only the fields a step genuinely needs, and only to the model you approve. n8n, your database and your documents stay on infrastructure you control. Nothing is used for training. If a process cannot leave your network at all, we can run it against a model hosted inside it.
Q3
What happens when the AI gets something wrong?
Two safeguards. Anything irreversible waits for a person to approve, so a wrong answer costs a click rather than a customer. And every run is logged with its input and output, so a bad batch can be found and reversed rather than guessed at. We tune assistants to say they do not know instead of inventing an answer.
Q4
Do we need n8n, or can you use something else?
n8n is our default because it self-hosts, the flows are readable by your team, and there is no per-task licence growing with your volume. If you already run Make, Zapier or Power Automate we will work with it. If a flow is better as plain code in your own codebase, we will say so.
Q5
How much does it cost to run every month?
Two parts: the server n8n runs on, which is usually small, and the model usage, which depends on volume. Every AI call is logged with its cost, so you see the real monthly figure per flow rather than an estimate. Most pilots land in the tens of dollars a month, not the hundreds.
Q6
Will this replace people on our team?
In our experience it moves them, rather than replaces them. The work that disappears is retyping, copying between systems and assembling the same report every week. What is left is the judgement part, which is also the part people would rather be doing. We will tell you plainly if a flow only makes sense as a headcount cut, because that changes how you should plan it.
Q7
Can you work with the systems we already have?
Usually yes. Anything with an API, a webhook or a database we can reach is straightforward. Older systems without an API are still reachable through scheduled file exchange or the screens themselves, which is slower but works. We check this during the audit, before quoting, so there are no surprises.
Q8
When is automation not worth it?
When the process runs a handful of times a month, when it changes shape every time, or when the rules live only in one person's head and nobody agrees on them. In those cases the honest advice is to fix or document the process first. We would rather say that than sell you a flow that breaks in a month.
Q9
Who owns what you build?
You do. Flows, configuration, prompts and any code we write are handed over, and your team is trained to make small changes without us. Your credentials stay yours. There is no component that only we can operate.
Q10
Which AI model do you use?
Whichever one fits the step. Classification and extraction run on a small fast model, because the volume is high and the quality difference is nil; drafting and reasoning go to a stronger one; anything that cannot leave your network runs on an open model hosted inside it. We work with Gemini, GPT and Claude behind one interface, alongside self-hosted open models, so changing a model later is a configuration change rather than a rebuild. Every call is logged with the model that answered and what it cost, so you can see the trade-off instead of taking our word for it.
Q11
How long before we see the benefit?
The pilot is designed so you can see it in about three weeks, measured against the numbers we captured in week one. If the first flow has not paid for itself in a quarter, we picked the wrong process and we will say so.
