HOW WE WORK · AI ACCELERATION

Where AI actually compresses the timeline. And where it still comes down to a person.

Six stages, six different jobs for AI. Here's what genuinely speeds up at each one, and what doesn't move at all.

THE HONEST VERSION

"AI-powered delivery" means nothing on its own. Which specific step gets faster is what matters.

Every stage of the method uses AI differently, and in a couple of places it barely uses it at all. Below is the stage-by-stage breakdown, not a slogan.

Where AI genuinely helps

Pattern-matching and volume work, at each of the six stages.

  • ✓
    Diagnose: process miningScans system logs and transaction history to reconstruct how work actually flows, in days rather than the two or three weeks interview-only discovery usually takes.
  • ✓
    Design: gap mappingCompares your documented process against standard Dynamics 365 behaviour and surfaces every difference automatically, instead of relying on a consultant to spot each one by memory.
  • ✓
    Data: cleansing & matchingFlags duplicate customers, inconsistent units of measure, and orphaned records at a volume no person could review row by row.
  • ✓
    Build: assisted code reviewChecks custom extensions for obvious defects and inconsistent naming before a person looks at the pull request, catching the boring mistakes early.
  • ✓
    Prove: generated test casesExtends a base test script against edge cases pulled from your actual transaction history, not a generic library that ships with the methodology.
  • ✓
    Land: in-app help agentAnswers routine "how do I..." questions from day one of go-live, so they don't all land on one exhausted person.

Where it still comes down to a person

Judgment calls, at the same six stages.

  • 1
    Diagnose: what's actually load-bearingA process mining tool can show you a pattern. It can't tell you whether that pattern reflects a real business need or a habit nobody's questioned in years.
  • 2
    Design: change the process or the systemEvery gap gets one of three answers, and choosing the right one is a judgment call, not a lookup.
  • 3
    Data: what's worth migrating at allThe model flags the mess. Deciding whether a record, a customisation, or a dead vendor is worth carrying forward is still a person's call.
  • 4
    Build: the configuration decisions themselvesReview tools catch obvious errors. They don't decide how a workflow, an approval, or a costing method should actually be configured.
  • 5
    Prove: reading what a failure meansA generated test case can fail. Whether that failure is a real defect or an expected edge case still needs someone who understands the business.
  • 6
    Land: anything that isn't routineThe help agent handles the predictable questions. Anything genuinely unusual still gets escalated to a person who knows your setup.
TYPICAL SCALE, NOT AN AUDITED FIGURE

Rough scale of what this saves, on a project like a single-entity Business Central implementation.

These are illustrative figures based on how these stages typically run, not a published or independently audited benchmark. Treat them as a rough sense of scale, not a guarantee for your specific project.

40–60%
Typical cut to Diagnose's calendar time when system access is clean, versus interview-only discovery, on a project like this.
~30%
Typical reduction in the hours spent on data cleansing during Data, on a single-entity migration of comparable size.
1 in 3
Rough share of first-month post-go-live support questions the in-app help agent answers directly, on a project like this.
WHAT WE WON'T CLAIM

AI shortens the runway. It doesn't remove the pilot.

We won't tell you AI removes the need for an experienced consultant on your project, because it doesn't. What it removes is the slow, manual grind of scanning, cross-referencing, and reformatting that used to eat the first few weeks of a stage before the real judgment work could start.

FAQ

Questions about how AI is actually used.

Q.Does using AI mean fewer consultants on the project?

No. It changes what the team spends its time on. Fewer hours go into manual scanning and reformatting; the same or more judgment goes into the decisions that actually shape the design.

Q.Where do you draw the line on what AI touches?

Anything that's pattern-matching against volume: scanning logs, comparing configurations, generating test variants. Anything that's a call about your business specifically stays with a person.

Q.Are the percentages on this page audited?

No, and we've said so above. They're a rough, illustrative sense of scale based on how these stages typically run, not a published benchmark you should hold us to for your exact project.

Q.What happens if the AI tooling gets something wrong?

It gets caught the same way any draft gets caught: a person reviews it before it's acted on. Nothing generated by the tooling goes live unreviewed.

Q.Is the in-app help agent trained on our data?

Yes, it's configured against your real chart of accounts and configuration during Build, not a generic script handed over at go-live.

Curious where AI actually fits into your project

Get a straight read before you commit.

Tell us your systems and data situation. We'll tell you honestly which stages would speed up and which wouldn't.

Book a 30-minute call

With someone who has run this before.

→

See the full method

The six stages this page fits into, end to end.

→