EZRUN TECH

How we work

We take on what the client usually pays for

Every request follows one path: the task, how we understood it, the solution on a copy of the system, the check, confirmation in the live system.

  1. 1 · youA task in your own words
  2. 2 · weA solution on a copy of the system
  3. 3 · weCheck
  4. 4 · youAccepted in the live system

Five promises.

1

We start without a finished specification

A request is written in plain words. We reply with how we understood it and propose a solution. What can be seen in the system itself, the AI analyst looks up first, and asks the author only what the author alone knows: the decision, the intent, how to check the result.

Example In an approval system the AI at first sent all ten requests of a batch back with questions. Most of the answers were in the system itself. Since then it looks there first and asks afterwards.

2

Clarifications are part of the work, not an extra charge

Before approval, any participant can edit the requirements. Approval is two decisions: “we take it on” and “the specification is fixed”. When a detail changes, only the specification is fixed again. Inside the agreed function there are no separate invoices.

Example In a treasury report a user noticed a wrong amount. The calculation rule was clarified and corrected on the test database the same day. The next day the author of the request confirmed it.

3

Mistakes are allowed: the live system is not touched

Development runs on a copy of the system, with a backup before changes are loaded. Before a person sees it, an independent AI checks the result against the acceptance scenarios. We take comments until you accept the result.

Example The first transfer of changes between servers failed twice. The cause was found by comparing files. That transfer was accepted only after a full comparison: all 504 files matched.

4

Documented as we go, closed on evidence

Every request has its own page: what the task was, how we understood it, how we solved it and what is still missing. A load into the test database is not acceptance: a request is closed when it is confirmed that the change is in the live system.

Example For one report the object was already in the live database, but the request stayed open: one field and the match of settings had not been confirmed.

5

We hand over the tools and the way of working

Along with the result you get everything it was made with: the way requests are handled, the AI agent settings, the checks and the documentation. Your team can carry on by itself, without us.

Example The finance function of a ten-company group runs its year's automation plan in-house.

We name the function in the agreement in one sentence. A new function is a new conversation.

Your ERP stays. A workspace goes on top.

We don't replace your ERP and we don't move data out of it. It remains the place where data, access rights and postings are kept. On top of it we build a workspace: whatever lives today in email, spreadsheets and messengers moves there.

Stays in the ERP
Data, reference lists, balances
Moves to the workspace
Collecting data from payments, contracts, documents and email
Stays in the ERP
Access rights: who sees and can do what
Moves to the workspace
Requests and their path from ask to decision
Stays in the ERP
Postings, period close, limit control
Moves to the workspace
Checks, questions and approval
Stays in the ERP
Reporting for the auditor and the tax authority
Moves to the workspace
One screen per role: what needs deciding today

Changes to the ERP itself are made on a copy.

Why it takes days.

1

Nothing to migrate

The data and access rights already exist in your ERP. We work with them where they are, so there is no data migration to a new system, the riskiest part of any rollout.

Example The accruals module takes its data from payments, contracts, electronic documents and email: from what the company already has.

2

AI works with the employee's own rights

It sees exactly what that person is allowed to see in the ERP, and by default it only reads. No new permissions and no workarounds appear.

Example Our assistant Nulio, which answers questions from ERP data, takes its permissions from the ERP, and only an administrator can switch changes on.

3

Rework takes days

AI agents do the development and testing, so a fix or a redo is a matter of days. That is why you can start without a finished brief and refine as you go.

Example After the accruals module was demonstrated, the team gave 10 to 15 comments. They went into work as ordinary requests.

4

One screen, one decision

People don't have to learn a new program. The system fills in what it can and shows what to check and what to decide.

Example In the accruals module, each role only checks what the system has already filled in.

What eats your team’s time?

One message. We’ll tell you honestly whether AI will help.