Services / AI Integration
We have shipped AI products of our own, which is mostly an education in where it does not help. The capabilities below are the ones that survived contact with real implementations: reading documents, writing procedures, tracking tests, and answering questions in context.
Where it pays back
Each one is high-frequency, low-judgement and expensive to get wrong by hand — which is the whole test. A task that happens twice a month has nowhere to pay back from.
Supplier invoices, scanned paperwork and email attachments turned into Business Central documents, reconciled against the order and the posted receipt — so a mis-read figure raises a legible exception instead of posting quietly.
Procedures captured from what people actually click, rather than from what someone remembers to write down afterwards. It is the difference between training material that exists and training material that is planned.
Test scenarios generated from recorded sessions and tracked to the requirement, so “are we done testing?” stops being a judgement call at the end of a go-live.
In-product guidance written against your processes and your vocabulary, so a new user reads your procedure rather than generic help for a product they have never used.
Where we draw the line
Anything that writes to a ledger goes through Business Central’s own posting engine with a human decision in front of it — the same rule we hold our commercial apps to. An AI feature that silently books inventory or finance entries is not a feature we will build.
“Classical statistics, back-tested against a baseline they have to beat — deliberately no AI in the arithmetic.”
We built these first
That is the difference between an integration practice and a slide: these are running, versioned and supported, and the engineers who would build yours are the ones who maintain them.
Process documentation
An AI process-capture platform. A Chrome extension records what you actually click; the guide writes itself — narrated, in four reading modes, shareable by link. Includes a Business Central assistant that explains the screen you are on, cited to Microsoft Learn.
On Microsoft AppSource (opens the marketplace listing in a new tab)
User acceptance testing
AI-powered end-to-end test scenario management for Business Central implementations. Generates test scenarios from recorded sessions, tracks execution with RTM traceability, and exports professional test workbooks.
On Microsoft AppSource (opens the marketplace listing in a new tab)
Supplier invoice processing
Vendor invoice to posted purchase invoice, inside Business Central. The invoice is treated as a claim and reconciled against BC’s own order and posted receipt, so a mis-read number raises a legible exception instead of posting a wrong figure.
On Microsoft AppSource
Financial foundation health
An always-on review of a Business Central financial foundation. It reads the Chart of Accounts, dimensions and posting groups, flags inconsistencies against a rule catalog, scores foundation health, and either writes fixes back to BC or exports a RapidStart configuration package.
Available direct
In your own tenant
We build the integration; you keep the data, and you keep the permissions model you already have.
Questions asked in plain language against Business Central data, scoped by the same permissions the user already has — a question in English still cannot read a table the asker is not entitled to.
Structured data pulled out of documents and landed on the right Business Central record, with the extraction shown to a human before anything is committed.
Guidance for your own processes and terminology, available on the screen where the question actually gets asked.
Straight answers
No. That is a rule, not a default we could be talked out of. Anything that writes to a ledger goes through Business Central’s own posting engine with a human decision in front of it — the same standard we hold our commercial apps to. An AI feature that silently books inventory or finance entries is not a feature we will build.
Into your own tenant’s processing path, scoped by your existing permissions, and that scope is part of what gets specified before anything is built. If a proposed capability would require moving data somewhere you would not be comfortable naming to your auditor, that is a reason to design it differently.
Ask what happens when it is wrong. A capability with a defined failure mode — an exception raised, a human shown the extraction, a suggestion left unposted — has been engineered. One that has no answer to that question has been demoed.
No, and one of our own products is the argument. ForecastQ forecasts demand with classical statistics, back-tested against a baseline they have to beat, and we say so on the product itself. Where arithmetic should be auditable, auditable arithmetic beats a model.
Yes, and that is the more common engagement. The work starts with which task is actually slow and how often it happens, not with which capability is available — a feature applied to a task that runs twice a month has nowhere to pay back from.
Next step
Tell us which part of the work is slow. If one of our products already solves it, we will say so. If AI is the wrong tool for it, we will say that too.
sales@dynamicspro.ca · Serving North America · 416-843-6575