Your next ERP program, minus the paperwork.
A Dynamics 365 program is mostly documents. Requirements, design specifications, configuration workbooks, test cases, runbooks, training material. AIM put AI against that drafting work and left every decision, approval and sign-off exactly where it was. You get the months back. You do not give up control of anything.
- AIRequirements documents
- AIFunctional design specifications
- AIConfiguration workbooks
- AITest cases and test data
- AITraceability matrix
- AICutover runbooks
- AITraining material
- ◆What goes in scope
- ◆Architecture and security
- ◆Accepting the system
- ◆The decision to go live
AI drafts. People decide, approve, and stay accountable.
What the framework changes.
Months back on the calendar
Roughly three months off a program modelled at over two years, on the conservative case AIM is willing to put in a contract. Not because anyone works faster, but because the drafting queue between decisions stops being the long pole.
A smaller team in your building
Peak consultant headcount falls with the hours. That shows up as fewer people to coordinate, shorter decision cycles, and less of your own team’s week spent bringing consultants up to speed.
Documentation that outlives the program
Traceability, configuration rationale, edge-case test coverage. The artifacts that get cut first when a program slips, and that you pay for at the first upgrade or the first audit. Cheap enough now to simply keep.
Seven stages. Where AI drafts, and where it is not permitted to.
The bar shows how much of each stage is drafting work AI absorbs, and how much stays with people. The split is not uniform, and the places it collapses are the point.
AI is strongest where the job is turning a defined input into a defined document, and it fades wherever the work carries a consequence, an ambiguity, or a person who has to be in the room. AIM models it that way because the evidence says so, not because it makes a better slide.
Discovery
42.6%42.6% drafting AI absorbs · 57.4% stays with people
AI drafts
- Writes the question set for every workshop before it happens
- Fills in the process boards in advance, so the room starts from something
- Turns what the room said into decisions, risks and open questions the same day
Stays with people
- Running the workshop and reading the room
- Telling a real constraint from a long-standing habit
- Deciding what you are not going to implement
Requirements
47.2%47.2% drafting AI absorbs · 52.8% stays with people
AI drafts
- Turns workshop notes into written requirements
- Writes the acceptance criteria for each one
- Keeps the traceability matrix current, which is the document every program abandons by month four
Stays with people
- Approving the requirements and owning them contractually
- Settling the ones two departments disagree about
- Deciding whether a gap is worth building for or absorbing
The biggest single impact in the framework, and the least glamorous. Traceability is the most automatable artifact in an ERP program and the first one dropped when a program falls behind.
Solution design
36.9%36.9% drafting AI absorbs · 63.1% stays with people
AI drafts
- Drafts the design documents from approved requirements
- Proposes who should be able to see and do what
- Maps your existing data onto the new system, field by field
Stays with people
- The architecture call, and approving the blueprint
- Trade-offs between what you want now and what you can upgrade later
- Anything an auditor or regulator will ask about
Configuration
36.2%36.2% drafting AI absorbs · 63.8% stays with people
AI drafts
- Generates the configuration workbooks straight from the design
- Builds the data load packages instead of typing into forms
- Records what was set and why, as it goes
Stays with people
- Making the changes in the real environment
- Confirming the numbers come out right
- Approving what gets promoted to production
Documentation effort falls by 70%. Execution falls by 30%. The gap between those two numbers is the whole thesis in miniature.
Customization
27.4%27.4% drafting AI absorbs · 72.6% stays with people
AI drafts
- Scaffolds the code and writes the unit tests
- Drafts the integration logic between systems
- Does the first pass of code review against your standards
Stays with people
- Deciding the pattern, and whether it survives the next update
- Final review and release approval
- Security and performance assessment
Deliberately the most pessimistic stage in the framework. This is where published evidence is worst, so we cap the assumption well below the figures usually quoted for AI coding. Complex work in an established codebase is exactly where measured gains have failed to show up.
Testing
38.0%38.0% drafting AI absorbs · 62.0% stays with people
AI drafts
- Writes the test cases from your scenarios, including the edge cases teams skip when they are behind
- Builds the test data sets
- Assembles the regression pack from what has already been run
Stays with people
- Your people running acceptance testing and accepting the system
- Deciding whether a defect blocks go-live
- Signing each test cycle in and out
The largest block of saved hours, because test cases multiply with scenario count. Coverage should improve as well as cost less: generated suites include the negative cases a stretched team quietly drops.
Deployment
32.0%32.0% drafting AI absorbs · 68.0% stays with people
AI drafts
- Writes training material and job aids from how the system was actually configured
- Drafts the cutover runbook and the rollback plan
- Writes the support articles from resolved issues
Stays with people
- The decision to go live
- Running the cutover, including the decision to call it off
- Standing in front of people whose jobs are about to change
Go-live execution itself is modelled at 15%, the lowest assumption anywhere in the framework.
Five decisions that stay yours.
Every document the framework produces is drafted by AI, reviewed by a named AIM consultant, and approved by a named accountable owner. These five carry no AI assistance at all, which means AIM does not price a saving against them and does not claim one.
What goes in scope
AI profiles the process catalogue and proposes a shape. A person decides what the business is actually going to implement, and owns that decision when it gets expensive later.
Requirements sign-off
AI writes the requirement, the acceptance criteria and the traceability entry. Approving it is an accountability act, not a drafting act, and the model assumes almost no saving on it for exactly that reason.
Architecture and security
Extensibility, upgrade safety, cost of ownership, and who can see what. Anything with a compliance or audit consequence is approved by a named architect, on the record.
Accepting the system
Your people run acceptance testing and accept the system, or they do not. AI writes the test cases and builds the data. It does not decide whether a defect is something you can live with.
The decision to go live
Go, no-go, and the decision to abort mid-cutover. Command and control stays with a named person who carries accountability for the business continuing to run.
The number we will sign to, and why it is the low one.
11.2%
Modelled reduction in delivery effort
Conservative case
Modelled estimate, not an achieved result
The framework produces three figures. AIM commits externally to the lowest one, which assumes AI performs at 60% of modelled rates and that checking its output costs a third more than planned.
The expected case is materially higher, and we will walk you through it with every assumption attached. It is not printed here, because an internal planning number quoted as a public promise is how partners end up defending a figure they never had grounds to give.
AIM has not yet completed a client engagement on this framework. The model is built from published Dynamics 365 delivery benchmarks and measured AI productivity studies, discounted before use, with the cost of verification and governance charged back against the benefit. It is a forecast, and we say so.
Two things most AI claims skip.
Every hour traces to something countable.
- Each delivery activity is tied to an object in the Microsoft Dynamics 365 Business Process Catalog, then given an effort norm and an activity-specific AI factor. That makes the estimate auditable rather than directional.
- That is the full reference catalogue, not a typical program. Nobody implements all of it. Your engagement is priced against your own in-scope counts after a discovery workshop, and AIM re-bases the catalogue at the start of every engagement because Microsoft keeps changing it.
Nobody is asked how much time AI saved them.
- A randomized controlled trial of experienced developers on complex work found participants believed AI had made them faster while measurement showed them running slower.
- AIM treats that as governing rather than inconvenient. Self-reported savings are inadmissible. Leverage factors come from measured studies, are discounted before use, and no activity is credited above a 70% reduction.
- The validation plan measures effort at activity level across a two-process pilot, with instrumentation in place before any AI-assisted work begins. It is deliberately built to be capable of disproving the business case.
What would three months and a smaller team be worth on your next program?
The framework is applied inside a Dynamics 365 engagement, not sold separately. It starts with a working session, then a Discovery Workshop that scopes your actual process counts and produces the business case. The estimate you get back shows the AI assumption behind every activity and the cost of checking it as a separate line, rather than folding it into the benefit. You can challenge any of it. That is the point of estimating this way.