Home / How we work
How we work
Assess. Connect. Build.
Check. Run.
Every engagement follows the same steps, so you always know what happens next, what it costs and who approves what.
Assess
We start with one process, not "AI for everything."
We map how the work really runs and pick one number to improve.
Pick one process
With a clear start and end, like "a customer asks for a refund" to "refund done."
Talk to the people who do it
Not just the owner. The person doing the work often walks through ten times more steps.
Time it
Minutes of actual work versus days until it's done. The days are usually what costs you customers.
Sort every step
Delete it, make it a simple rule, AI with a person checking, or person only.
Pick one number
Such as days to get paid, or hours of data entry a week, and measure it before and after.
Recommend the smallest fix
In writing. If an off-the-shelf tool will do the job, we'll say so.
How we work
Assess. Connect. Build. Check. Run.
The same five steps behind every solution, so you always know what happens next.
Assess
We map how one piece of your work really runs, time it, and pick one number to improve.
Connect
We connect the tools you already use, with only the access each job needs.
Build
We build the automation, assistant or report, with clear rules on what needs your yes.
Check
Every answer is sourced and tested against your own documents and data before launch.
Run
We run and maintain it for you, re-test after every change, and report what it did.
You stay in control
Three lines for every task.
Before anything goes live, we agree with you which line each task sits on.
Routine, low-risk work, like sorting email or updating a spreadsheet.
Replies, invoices and reports wait for your yes. Sending and spending always do.
Things like changing bank details, making payments or hiring decisions stay with people.
Safeguards in every build
Built so mistakes cost little.
- Every workflow has a named owner on your side and an off switch.
- Each AI sees only what its job needs, and only what the person using it could already open.
- Incoming email and web content are treated as untrusted: nothing sends, pays or exports because an email asked it to.
- Tested on a copy first, with backups and a log of every action.
- Trust is given in stages: watch it, then schedule it, then let it run.
- Before launch, assistants must answer known questions correctly and refuse questions they can't answer from your files.
Where are you today?
Four levels of AI in a business.
Tasks
People ask AI for help one task at a time, then copy the answer into their other tools.
Processes
One job runs on its own when something triggers it, like answering every new lead in minutes.
Departments
Several processes share one customer record and one set of rules.
Operation
Each area hands work to the next, and you approve and read the numbers.
Most businesses are at level 1. We usually start by getting one job to level 2, and we only build level 4 with people approving the key steps.
The Nebacore Check
No AI is perfect. That's why we check it.
Even the best AI still gets things wrong. Every solution we build runs the same five checks, so mistakes get caught before they cost you.
Sourced
Every answer points to the document, contract line or record it came from. No source, no answer.
Checked against your documents and data
Numbers are compared with your real contracts, prices and records, not taken on the AI's word.
Knows when to stop
When it's unsure, or money or records are involved, it hands off to a person instead of guessing.
Tested before and after
Nothing goes live without passing our tests, and every change is re-tested.
Scored and reported
You see what was found, what was caught and how accurate it was.