Local AI vs cloud AI: a practical choice for small businesses
AI & automation · November 2026
"Local AI" means the model and the tools around it run on hardware you control: an office workstation, a private server or an edge device. "Cloud AI" means the model runs on a provider's infrastructure and you access it over the internet.
When local AI fits
- Documents contain client, financial or personal data that should not leave a controlled environment for routine work.
- The workflow is narrow and repeated: extracting fields, summarising documents, answering questions from an approved knowledge base.
- You can identify an owner for reviewing outputs and maintaining the system.
- The hardware budget and expected usage are understood before the model is chosen.
When cloud AI can be the better choice
- The task needs the largest current models and you do not have the hardware to run them locally.
- Usage is occasional and a local server would sit idle.
- The provider's data handling terms meet your obligations and the workflow is not sensitive.
Questions to ask before choosing
- Where will the data be stored and processed?
- Who can access the system, and what is logged?
- What happens when the model gives a wrong answer?
- What is the fallback if the local hardware fails?
- What will it cost to maintain over a year, not just to install?
The honest limit: installing a model locally does not, by itself, create data security or a useful workflow. Access control, backups, evaluation and human review still matter.
Explore private AI services
Data recovery vs phone repair: what is the difference?
Data recovery · November 2026
Repair aims to make the device work again. Data recovery aims to extract the data. They are related, but they are not the same service and they do not carry the same promise.
A phone can be repaired and still lose its data. A phone can remain non-functional after a successful recovery. A storage chip can sometimes be read directly even when the phone will not boot.
What affects the chance of recovery
- The type of failure: logical, firmware, hardware or physical damage.
- Whether the device was powered on repeatedly after the failure.
- Previous repair, reset or flashing attempts.
- The condition of the storage chip and circuit board.
- Whether the data is encrypted and the required key is available.
Before you send a device
- Stop using it and avoid repeated reboots or resets.
- Back up anything still accessible.
- Note the passcode or pattern if you have it; it may be needed.
- Ask about assessment terms, data handling and confidentiality.
- Do not assume a recovery attempt will also repair the phone.
No responsible provider can guarantee recovery. Be cautious of anyone who does before examining the device.
See data recovery services
When does a business actually need custom software?
Software · November 2026
Custom software is justified when the process is important, repeated and poorly served by the available tools. It is rarely justified simply because the business wants something bespoke.
Signs that a custom build may be justified
- Staff maintain the same data in several places and reconcile it manually.
- A core workflow is forced into a tool that does not match how the work actually happens.
- The business pays for several products that do not share data.
- Reporting requires manual exports and rework every week or month.
- The process is a source of customer errors, delays or lost revenue.
Try these first
- Simplify or standardise the process before automating it.
- Check whether an existing product can be configured rather than replaced.
- Fix the data ownership and reporting definitions.
- Estimate the annual cost of the current workaround, including staff time and errors.
Questions that keep a build honest
- Who will own the system after launch?
- What must the first version do, and what can wait?
- Which integrations are mandatory, and do their APIs exist?
- How will acceptance be tested against real work?
- What is the maintenance and hosting cost over three years?
Explore software development
What to prepare before a software project
Software · November 2026
A little preparation before a build saves far more time than it takes. These are the inputs that make scoping faster and the result more useful.
- A process map: the steps from enquiry or order to delivery, and who performs each step.
- Real examples: sample forms, invoices, reports and spreadsheets, with sensitive data removed.
- User roles: who needs access to what, and who approves changes.
- Data sources: where the data comes from, who owns it and how it is currently kept.
- Integrations: the systems that must connect, and whether they expose an API.
- Acceptance criteria: what “working” means in the real workflow, not just in a demo.
- An owner: one person on the business side who can make decisions and consolidate feedback.
If some of this is missing, a short discovery or technical assessment can produce it before a full build begins.
Discuss a software project
How to evaluate a private AI vendor
AI & automation · November 2026
Private AI is an infrastructure and workflow decision, not just a model choice. Ask these questions before committing budget.
- Data location: where will documents, prompts and logs be stored and processed?
- Hardware: what hardware is required, who owns it, and what happens when it fails?
- Model choice: which model, why it fits the task, and how it will be updated or replaced?
- Evaluation: what sample data will be used to measure accuracy and failure cases before rollout?
- Integration: which existing systems will connect, and what happens if an API is unavailable?
- Access and logging: who can see the data, what is logged, and how is access removed?
- Support: who maintains the deployment, and what is the response path when it breaks?
- Exit: how do you take your data and leave if the vendor relationship ends?
A useful pilot ends with evidence and a decision. It should not become an indefinite subscription by default.
Explore private AI services