How we work
Most firms we talk to have the same question before anything else: what actually happens if we say yes? This page answers it plainly. It describes how we run a project, what we won't automate, and the questions about data we expect you to ask.
The short version
- We start with one repetitive workflow, not a programme of change.
- We agree a fixed scope in writing before building anything.
- We connect the tools you already use, where they allow it.
- People at your firm sign off anything sensitive.
- You get logging, a plain-language handover and a way to switch it off.
The four steps
1. Squeeze: find the work worth removing
We start with a conversation about real tasks, not a wish list. Which job comes back every week? Who does it, how long does it take, which systems and inboxes does it touch, and what goes wrong when it's late? If it turns out automation isn't a good fit, we'll say so.
2. Strain: agree the smallest system that solves it
We write down one workflow and agree it with you before any build starts: what the automation will do, what it's allowed to access, which steps need a person's approval, who owns it at your end, what the risks are, and what it will cost. If something can't be connected, for example because a system doesn't offer an API or your licence doesn't include it, we say so at this stage rather than discovering it later.
3. Pour: build it and test it on real cases
We build against real examples from your work, with the people who do the job today. Early on, every outgoing message or system update can be held for a person to approve, so you can see exactly what the automation would do before it does it. Sensitive steps keep a human sign-off permanently.
4. Refill: measure, tune, then extend
Once it's running, we look at what changed and adjust what didn't work. You get logging, alerts for failures and a plain-language handover document that explains what the automation does, what it touches and how to pause it. We only suggest extending it to the next workflow once the first one has earned your trust.
What we won't automate
Some steps should stay with people however good the technology gets. As a rule, we won't build automations that:
- make professional judgements, such as tax treatment, legal advice or whether to act for a client
- approve pay, refunds or payments without a person signing off
- make safety decisions, such as what to do about a dangerous gas or electrical result
- carry out identity, right to work or anti-money laundering decisions in place of your staff or your chosen verification provider
- send anything in your name that you haven't agreed it can send
An automation can prepare, sort, draft, chase, file and flag. Your team decides.
Questions about data we expect you to ask
We'd rather you asked these up front. For any workflow we propose, we'll answer:
- What does it access? Which mailboxes, folders, systems and records, and with what level of permission.
- Where is data processed and stored? Including which third-party services are involved.
- Can an AI provider keep or train on our data? And what settings or agreements apply.
- How long is anything kept? Including logs.
- Who can see what it did? Logs should show each action the automation took.
- How do we switch it off? There should always be a simple way to pause it.
We design around least-privilege access, keep humans in the loop for sensitive steps and document where data flows. We don't claim certifications we don't hold. If your firm has its own data protection or IT requirements, we work to them.
Who this suits
A five-person firm with one painful process is as good a starting point as a larger operations team. The sectors we focus on are accountancy, recruitment, freight and logistics, law firms and letting agents, and the same approach applies elsewhere.
Email us or write to luke.sewell@melonjuice.ai, or use the contact form
Fair questions
How much does a project cost?
There is no public price list, because it depends on the workflow and the systems involved. We agree the scope and the cost with you in writing before any build work starts.
How long does it take?
It depends on the workflow and on access to your systems, so we don't promise a number of days on this page. Scope, risks and owners are agreed before a build, and the first system stays small until it proves itself.
Do we need to replace our existing tools?
Usually not. Most of the work connects and automates the systems you already pay for. Where a tool can't be connected, we'll tell you before we start.
What happens after handover?
You get documentation your team can follow, logging and alerts. Ongoing support or changes can be agreed separately if you want them.
How do we start?
Email us with a short description of the task you'd like to get rid of. We'll reply to say whether it looks like a fit and suggest a call if it does.
Email us or write to luke.sewell@melonjuice.ai, or use the contact form
Background reading: Insights · AI automation by sector · FAQ