CDBAMImplementation Projects
Best practices and intelligent tools

What we do differently is not the steps. It is what remains after them.

Any consultancy can list the phases of an implementation. The difference shows in the detail: how evidence is produced, what happens when a figure does not reconcile, and how much of the knowledge stays with the client at the end.

Practices

Five rules we do not negotiate

These are not slide-deck slogans: they are built into the tools we use, and they are what we negotiate scope and schedule against.

01

Evidence is born with the deliverable

Nothing is produced without validation material. Every file, every load, every task delivered comes with what makes it checkable — and the evidence is filed separately from what goes into production. Validation pushed to the end of the project is validation that never happens.

02

Data used is data discarded

Payroll is sensitive personal data. Data handled in the project never travels through the cloud or the internet and is discarded as soon as it has served the purpose it was used for. Access credentials are never written to a file. The safest way to protect information is not to keep it alive longer than necessary.

03

We never alter the source data

The tool reads, transforms and writes — it never fixes the legacy system on its own initiative. What is recorded at source is historical record, to be located and flagged. Correcting it is the client’s decision, taken with the problem in hand.

04

Silence is a defect

A routine that fails quietly is worse than one that fails loudly. An unmapped code, a date loaded as text, one column too many: none of these complain on their own — they complain during the load, or after it. Our tools complain early, with the reason written down.

05

The knowledge stays with the client

At the end of the project the client receives more than a working system: the discovery document, the documentation of what was configured and why, the test scripts with the reconciliation results, and the dictionary explaining every migrated field. That is what lets the next person understand the decisions without having to call whoever made them.

Intelligent tools

The practices became software

A rule that depends on human discipline fails on a long project. That is why we built our own tools — so the right way is the easiest path, not the most laborious one.

Tool 1

Technical Discovery

The system where the client’s landscape is recorded, organised and consulted. It exists because a discovery that lives in a spreadsheet on someone’s laptop does not survive the project — and discovery is precisely what the schedule, the scope and every later decision rest on.

  • A structured script. The same set of questions on every project, so nothing depends on a consultant remembering to ask.
  • A living, versioned record. What was found is stored with its history — not in a spreadsheet version someone renamed.
  • Consolidated into a document. The discovery becomes a printed document, handed to the client as the basis of the implementation plan.
  • A single deliverable. The discovery becomes one document the client receives as the basis of the plan — with no versions traded by email.

Why it starts here

Discovery is the first phase of the project and the only one that delivers value even on its own: by the end of it the client knows the real size of what lies ahead, with the risk points named.


It is also what separates a defensible estimate from a guess — and what avoids the awkward mid-implementation conversation about what was or was not agreed.

Artificial intelligence built in

Used where it is genuinely good — and blocked where it is not

Both tools have artificial intelligence support built in. Not to generate the project or to decide business rules: to read large volumes of technical text quickly and suggest a path when the structure is ambiguous. The decision remains the consultant’s.

Reading logs and errors

Error messages and execution logs translated into plain language, with the likely point flagged — instead of scrolling a thousand lines looking for the one that matters.

Assistance writing the rules

Suggestions and review of extraction rules while they are being built, with the result always testable before it is applied.

The residue of tracing

When a field’s origin is ambiguous for deterministic analysis, the AI digs only into that remainder. The document comes out complete even with it switched off.

Limits

What our AI does not do

Artificial intelligence on a payroll project demands clear rules. These are ours — and they live in the software, not in a policy nobody reads.

It never sees payroll data

What goes to the AI provider is structure: a column’s calculation expression, the column map, the error message. Never anyone’s salary, never a line of payroll history.

It never stores credentials

The access key exists only while the tool is open. It is never written to a configuration file and no response ever returns it.

It never spends unseen

The cost of each call is deducted from a credit you enter, by actual consumption. Until that credit is entered, the percentage stays blank: saying “I don’t know” is better than showing 0%.

It never uses half an answer

If the response comes back truncated, the batch is split and redone, instead of saving half an analysis and assuming it worked.

Want to see this running on your data?

A thirty-minute technical conversation, with your scenario on screen, shows more than any presentation.