AI where it helps. Rules where they're enough.
Some steps in a business workflow need judgement: reading an incoming document, working out what a request is about, drafting the right follow-up. AI can handle steps like these. Everything around them stays as clear business rules, and a person approves the decisions that matter.
Where It Fits
The repetitive work that still needs a bit of thinking
Plain automation is good at fixed rules: send a reminder on day seven, copy a form into a spreadsheet. It struggles when someone has to read something first. These are the steps where AI tends to earn its place.
Processing incoming documents
Checking what a file is, which period it covers and whether pages are missing, before anyone has to open it.
Spotting missing information
Comparing what arrived with what was needed, and listing exactly what's still outstanding.
Sorting emails and requests
Working out what each message is about and how urgent it is, so it lands in the right queue.
Preparing reports
Pulling the figures together and drafting the summary, with the numbers calculated by plain code, not estimated by AI.
Following up on open items
Drafting a specific reminder for each outstanding item, instead of one generic chase email for everything.
Routing work to the right person
Sending each item to whoever handles that type of work, with the context they need attached.
If a fixed rule can do the job, we use the rule. It's cheaper, faster and easier to trust. AI is added only for the steps that actually need it.
How It Runs
AI handles one step. The workflow keeps it in check.
Each workflow is a fixed sequence that we design with you in advance. In the workflows we build, the AI doesn't decide what happens next on its own. It does one defined job, such as classifying a document or drafting a reply, and hands its result back to the workflow.
Rules check the AI's result, and the actions you mark as important wait for a person on your team to approve them. Where appropriate, the workflow can record processing status, errors and key actions, such as what was approved and by whom, to support troubleshooting and review. What gets recorded is agreed for each project.
When Things Aren't Clear
What happens when information is missing, unclear or wrong
It gets flagged, not skipped
The workflow is designed to list what's missing and draft a specific request for it. The item waits until the gap is filled or someone on your team decides to proceed.
It goes to a person, not a guess
When a result falls below the agreed confidence level or fails a check, the item is passed to a person with a short note on what was unclear. AI confidence isn't a perfect signal, which is why important actions still need a person's approval.
It's caught before the next step
Validation rules are set up to catch results that don't add up, such as the wrong period, a total that doesn't match or a duplicate, before they go any further. Rules reduce errors; they don't replace review.
We design workflows so that text inside an uploaded document or an email is treated as information to check, not as instructions. The AI step can't trigger actions by itself, so a document that says “pay this” or “ignore the rules” still goes through the same checks and approvals.
Built With a QA Mindset
Tested like any other system, because it is one
Noera is led by Radita Pratiwi, who spent eight years in QA leadership and business process reengineering before founding Noera. That shapes how AI steps get built. The question isn't whether the AI is clever. It's whether the workflow behaves correctly when real, messy input arrives.
Input validation
Checking that what arrives is complete and in the expected format before any processing starts.
Exception handling
Deciding in advance what happens when a step fails, times out or receives something unexpected, and testing those paths, not only the expected ones.
Missing or inconsistent data
Testing the messy cases, not only the clean ones: blurry scans, wrong files, half-filled forms, details that don't match.
AI uncertainty
Setting the point at which the AI must hand an item to a person instead of making the call itself.
Human approval
Agreeing with you which actions always need a person's sign-off, and building that in from the start.
Testing and acceptance criteria
Agreeing what "working" means before building, then testing each AI step against agreed sample cases before go-live. The testing scope is set out in each project's written scope.
Example
Chasing missing documents for an accounting firm
A small accounting practice sends each client a list of the documents it needs: bank statements, receipts, payroll reports. Clients send some of them, the wrong ones, or nothing at all. Staff spend a large part of each week checking what arrived and writing chase emails.
In this example, the firm's checklists and client records stay where they are. The workflow sits alongside them and takes over the checking and the first draft of every chase.
- Each client has a checklist of the documents needed for the current period.
- When a client uploads a file, the AI suggests what it is, which period it covers and whether it looks complete.
- The checklist updates. Anything unclear goes to a staff member instead of being guessed.
- For each gap, the workflow drafts a specific reminder, for example: "We have your March and April statements. May is still missing."
- A staff member approves or edits each reminder before it's sent.
- Follow-ups repeat on a schedule, and the team gets a weekly list of who is still outstanding.
The approval step follows the same pattern as Signofly, our live approval-tracking product: every item has a status, an owner and a timeline.
How a Project Works
Start small, prove it, then extend it.
A first project usually covers one part of one workflow, not the whole business.
-
01
Discovery Call
A short call to understand the workflow and the problem. If AI isn't the right tool for it, we'll say so.
-
02
Map the Workflow
We look at how the work happens today and mark which steps need judgement and which a simple rule can handle.
-
03
Agree the Scope
A written scope covering what the first version does, what success looks like, how it will be tested and how data will be handled, including any outside services. Nothing is built before this is agreed.
-
04
Build & Test
We build it and test each AI step against sample cases based on your real work, anonymised where practical, then review the results with you.
-
05
Run With Full Review
The workflow goes live with a person approving every output at first, so you can see how it behaves before relying on it.
-
06
Hand Over & Adjust
Handover can include documentation covering the workflow, key settings, operating instructions and known limitations, as agreed in the scope. Then adjustments as new cases appear, with ongoing support available if you want it.
Your Data
How project data is handled
- Minimum access. We aim to use the minimum access necessary to deliver the agreed work. Access requirements are agreed with you before work starts, and at the end of a project we list the access we used so your team can review or remove it.
- Working within your systems. Where practical, we aim to work within your existing systems and minimize unnecessary data transfers. Some workflows need an outside service, such as an AI API, hosting or an automation platform. When that's the case, we discuss it with you before implementation.
- Outside services agreed first. Any third-party service that will handle your data, such as an AI API, cloud hosting, an automation platform, an email provider or an external database, is disclosed and agreed before it's used. The exact services and data flows depend on the design, so they're confirmed once the scope is clear.
- AI services reviewed before use. Before using a third-party AI service, we review its data-handling terms and training policies, along with the relevant service and account settings. These differ by provider, plan and configuration, so the AI services and data-processing arrangements for your project are discussed with you before implementation.
- Proportionate records. Where appropriate, workflows can record processing status, errors and relevant actions to support troubleshooting and review. What gets recorded depends on the project, your privacy requirements and the technical design, and we avoid copying documents or sensitive content into logs when it isn't needed.
Noera is a small studio and doesn't currently hold security certifications such as SOC 2 or ISO 27001. If your industry has specific data or compliance requirements, tell us at the start and we'll say plainly whether we can meet them. How this website handles enquiries is covered in the privacy policy.
Discuss a Workflow
Tell us which part of the work keeps repeating.
You don't need to know whether AI is the answer. Describe the workflow and where it gets stuck, and we'll tell you honestly what would help, AI or not.
The first step is a short call to understand the problem and check whether we're the right fit. Detailed solution design and consulting are handled separately when needed. We typically reply within one business day.
Send the details: