Answers, before you have to ask for them
The questions that come up in every first conversation, answered in full. Each one opens with the complete answer; the detail underneath is there if you want it.
- Engaging Genisys
How long does a custom ERP take?
A focused custom ERP release normally takes several months, while a multi-entity replacement with substantial migration and integration work can take a year or more. The controlling factors are process breadth, data condition, external dependencies and the organisation’s capacity to make and verify decisions.
Read in full - Engaging Genisys
What does a discovery phase actually produce?
A discovery phase produces an evidence-based delivery plan: agreed outcomes, process and system models, tested technical assumptions, release boundaries, risks and an estimate. Its value depends on resolving the uncertainties that could materially change the design or plan, rather than producing a large requirements document.
Read in full - Engaging Genisys
Who owns the code we pay you to write?
The client should own the bespoke source code and project artefacts paid for once the contract’s payment and assignment conditions are met. Pre-existing tools, third-party components and open-source software remain subject to their own ownership and licence terms, which must be identified rather than silently treated as bespoke work.
Read in full - AI & Machine Learning
Do we need a custom model or an API?
Most organisations should begin with a hosted model API or an established open model, then measure it against a representative evaluation set before considering custom training. A custom model is justified when specific quality, control, latency, deployment or unit-cost requirements cannot be met through prompting, retrieval, tools or fine-tuning.
Read in full - AI & Machine Learning
How do you stop an LLM inventing answers?
An LLM cannot be guaranteed never to invent an answer, but unsupported claims can be reduced and contained through grounded evidence, constrained tools, explicit abstention, verification and evaluation. High-consequence facts and actions must remain subject to deterministic checks or human approval rather than model confidence.
Read in full - AI & Machine Learning
What does an AI feature cost to run?
An AI feature’s running cost is the sum of model inference, retrieval, tool calls, storage, observability, human review and the capacity reserved for peak demand. It should be estimated from measured end-to-end requests and expected workload distribution, because a provider’s token price excludes retries, context growth and surrounding infrastructure.
Read in full - Enterprise Systems
Should we buy an ERP or build one?
Buy and configure an ERP where processes are standard and the product can remain authoritative without extensive workarounds; build where distinctive operations or integration constraints create durable requirements that products cannot support cleanly. Most organisations need a deliberate combination, with a packaged financial core and custom capability around genuine operational differences.
Read in full - Enterprise Systems
How do you migrate off a system with no API?
A system without an API can be migrated through supported exports, read-only database extraction, vendor reports or controlled interface automation, chosen in that order. The migration must preserve meaning through profiling, mapping, repeatable transformation and reconciliation, with screen automation used only where no safer extraction path exists.
Read in full - Enterprise Systems
What makes accounting software audit-ready?
Accounting software is audit-ready when every reported balance can be traced through immutable, balanced entries to an authorised source and reproduced for the relevant period. That requires controlled posting, approvals, access history, reconciliations, period management and retained evidence, not merely an audit-log screen.
Read in full - Engineering Practice
How do you hand a software project over?
A software project is handed over by transferring practical control of its code, environments, credentials, documentation, decisions and operating procedures, then verifying that the receiving team can use them. Handover should happen throughout delivery and conclude with observed build, deployment, recovery and support exercises rather than a document dump.
Read in full - Engineering Practice
What happens when a deployment fails?
A failed deployment should stop automatically, preserve diagnostic evidence and either restore the last known-good release or move forward with a tested repair. The safe choice depends on whether data or external contracts have changed, so rollback capability must be designed and rehearsed before release rather than improvised during failure.
Read in full - Engineering Practice
How do you price a fixed-scope build?
A fixed-scope build is priced from an examined delivery boundary, acceptance criteria, technical approach, dependencies and explicitly allocated risks. A defensible fixed price requires discovery evidence and controlled change; fixing cost before uncertain integrations, data and decisions are investigated either inflates contingency or conceals exclusions.
Read in full
Ask the one that is not here.
If your question is specific to the system you are running, it will not be on this page. Send it over and it will be answered directly.