FORGE · Business AI for UK organisations
AI implementation and workflow automation for UK businesses
Put AI to work across your data, systems, and business processes - locally, in the cloud, or both - with human review and operational control.
MPED implements AI agents and automates document processing and CRM workflows for UK businesses. FORGE connects your existing systems to check documents, update records and complete tasks. Your team controls where data is processed, what AI may do and when human approval is required.
Built for: sensitive information, regulated work, and multi-step processes that a general-purpose chatbot cannot complete safely on its own.
Operating principle
The model proposes. FORGE governs execution.Business AI platform
Move beyond isolated AI tools and fragile automations
Business AI has to work across real data, systems, responsibilities, and exceptions. FORGE provides the controlled execution layer around the model, so an AI agent or automation can become part of an accountable operating process.
Produces an answer or proposes an action
Useful intelligence, but usually without the complete process around it.
Moves controlled work from evidence to outcome
Integrations, state, policy, recovery, human review, and a retained run history.
The UK Business Data Survey 2026 found that only 21% of businesses using AI had integrated it into existing business systems, while 73% were uncomfortable with business data being used to train external AI models. Read the UK government research.
Protect confidential business data
Keep sensitive evidence within an agreed local, private, or hybrid boundary instead of sending every document and prompt to a general-purpose public AI tool.
Integrate AI with existing systems
Connect documents, CRM, finance platforms, internal applications, and approved APIs without giving the model unrestricted access to business operations.
Automate multi-step processes reliably
Use checkpoints, a durable execution journal, bounded retries, and validation so long-running AI workflows can recover instead of silently losing their state.
Keep human oversight and evidence
Give reviewers the source evidence, missing information, proposed outcome, and an explicit approval point before a controlled business action is completed.
AI workflow automation
One controlled path from evidence to action
The workflow remains understandable to the people responsible for it, even when several models, systems, and review steps are involved.
- 01
Connect the evidence
Documents, approved business systems, and structured inputs enter a defined process.
- 02
Let the model propose
A suitable local or hosted model prepares an answer, plan, or next action.
- 03
Govern execution
FORGE applies policy, state, budgets, validation, retries, and permitted-effect boundaries.
- 04
Review and approve
A person sees the evidence, missing information, proposed output, and required decision.
- 05
Act and record
Approved changes are applied to connected systems and retained in an auditable run history.
Private, local, hybrid, and cloud AI
Choose local AI, hybrid AI processes, or cloud AI
FORGE does not lock the process to one model or one hosting pattern. Each workflow can use the deployment option that fits its data sensitivity, capability, integration, cost, and operating requirements.
Local AI
Run approved open-weight models on infrastructure controlled by your organisation when sensitive data or operating policy should keep work inside your environment. This includes on-premises AI deployments.
Hybrid AI processes
Keep sensitive steps and evidence local while approved cloud models handle selected tasks. FORGE controls what can leave the local boundary and how each result re-enters the workflow.
Cloud AI
Use selected hosted models where their capability and commercial model fit the task, while FORGE continues to govern evidence, state, permissions, budgets, and approval.
AI agents and business process automation
FORGE use cases - AI for sales, logistics and insurance
Three anonymised examples show how business AI turns calls, email and documents into completed work in existing systems. FORGE configures CRM demos, assembles transport documents and registers assistance cases, then verifies the result.
01 / Sales and CRM
CRM automation - HubSpot demo provisioning
- Specialist preparation after the call
- ~4 hours
- Automated demo build
- 22-32 min
FORGE reads the discovery call transcript, interprets the requirements and configures a tailored HubSpot demo with synthetic data. It then verifies the setup and prepares a presentation scenario for the salesperson.
The salesperson can typically return to the buyer within an hour of the transcript becoming available.
This compares post-call preparation. The roughly one-hour customer discovery call still takes place.
02 / Logistics
AI document automation for logistics
- Manual document collection and forwarding
- ~60 min
- FORGE aggregation and onward transmission
- Up to 4 min
In food-sector logistics, FORGE identifies the documents needed for a shipment, collects them from incoming email and checks completeness against workflow rules. It forwards the package, typically 2-8 documents, to the carrier and records what was sent.
Less manual attachment handling and a lower risk of missing a required document or forwarding the wrong file.
The comparison covers document handling for one shipment, not the transport time.
03 / Insurance assistance
AI for insurance assistance case registration
- Manual case registration
- ~5 min
- Average registration time with FORGE
- 2 min 3 sec
FORGE interprets a call, transcript or incoming message, extracts the required facts and assigns the relevant assistance business line, such as motor, home, travel or electronics protection. It registers the case in the existing system, including attachments where the source channel supports them.
Operators can focus on the customer while consistent classification and registration rules handle the system entry.
This compares case registration. It excludes the roughly four-minute conversation with the insured customer.
Omnichannel assistance intake
The same intake process can handle telephone calls and voice transcripts, online messengers, website chat, B2C and B2B portals, partner APIs and email. Attachments are included with the case where the source channel makes them available.
Operational reference measurements from the described use cases; customer identities are withheld. Timings depend on inputs, integrations, models, infrastructure, workload and required approvals. They are not contractual SLAs or guaranteed labour savings.
See FORGE in action
HubSpot automation and AI-assisted data review
Two walkthroughs show task execution and result verification in an existing CRM. Explore demo configuration and case completeness review. Both recordings have English audio and English subtitles.
01 / CRM automation
3:15From a call transcript to a configured HubSpot demo
Pawel Kiezynski shows how FORGE turns a brief into a HubSpot configuration: contact properties, a custom object and associations between records. The result is verified and then demonstrated in the CRM.
- Requirements and process launch
- Verified configuration result
- Properties, objects and associations in HubSpot
The recording lasts 3:15. The process described takes around 20-30 minutes; the walkthrough shows the result of an earlier run. This is a demo environment with fictional data. Free rein expands the example within the selected process boundaries.
02 / AI-assisted case review
4:16AI checks case completeness in HubSpot
Using a financial-services conversation as an example, Pawel demonstrates transcript analysis in HubSpot. FORGE identifies missing evidence, cites the analysis sources and sets out the next steps for the adviser, client and administrator.
- A conversation transcript in the CRM
- Missing documentary evidence identified
- Sources and next steps for the team
BLOCKED means the case needs further documentary evidence. It is a review outcome, not a system failure. The 85% figure describes completeness of this example case, not AI accuracy. The recording uses a demonstration scenario.
Working reference, not a concept
A complete case workflow, run locally from evidence to review
MPED has run a synthetic financial fact-find case through a local FORGE reference environment. The test covered evidence extraction, structured drafting, missing-information handling, and review preparation rather than financial advice or automated suitability decisions.
This is engineering evidence from a synthetic benchmark. It is not a production SLA, a client result, or a guarantee of accuracy or throughput.
FAQ
Business AI, workflow automation, and private AI questions
The terminology overlaps. These answers explain how FORGE turns models, AI agents, deployment choices, and integrations into a controlled business workflow.
What is business AI?
Business AI applies models to real operational work: understanding evidence, preparing outputs, coordinating tasks, supporting decisions, and taking permitted actions across business systems. FORGE adds the workflow, controls, integrations, review points, and recovery needed to use those capabilities as a dependable process.
What is AI workflow automation?
AI workflow automation uses AI within a defined business process to interpret information, prepare an output, choose a permitted next step, or support a controlled action. Unlike a standalone chatbot, the workflow also needs integrations, validation, state, failure handling, permissions, and appropriate human review.
What is private AI for business?
Private AI is an AI setup designed so an organisation can control where its data is processed, which models are used, who can access the workflow, and what actions the system is allowed to take. It can run on premises, in a private environment, or through a controlled hybrid architecture.
What is on-premise AI?
On-premise AI, more formally called on-premises AI, runs models and supporting software on infrastructure controlled by the organisation rather than relying entirely on a public AI service. FORGE can support this pattern where it fits the security, performance, and operating requirements.
Does FORGE have to run fully locally?
No. FORGE supports three delivery patterns: local AI, cloud AI, and hybrid AI processes. The right option depends on data sensitivity, model capability, integrations, cost, and the governance requirements of the workflow.
What is a hybrid AI process?
A hybrid AI process combines local and cloud capabilities within one governed workflow. Sensitive evidence or selected tasks can remain local, while approved cloud models handle other steps. FORGE applies the routing policy, records the run, and keeps review and approval explicit.
Is FORGE an AI chatbot?
No. FORGE is a workflow and execution platform. A conversational interface can be part of a solution, but the core product manages evidence, state, permissions, budgets, recovery, approval, and controlled actions across a business process.
How does FORGE control AI actions?
The model proposes an output or action, while the FORGE runtime applies the rules around execution. Depending on the workflow, these can include permissions, validation, idempotency, retry limits, cost budgets, human approval, and a retained run history.
Can FORGE integrate AI with our existing business systems?
Yes. A FORGE workflow can connect to approved document sources, CRM, finance platforms, internal applications, databases, and APIs. Each integration is scoped around the minimum access and permitted actions required for the process.
How much does private or on-premises AI cost?
Cost depends on the workflow, model capability, integrations, expected volume, availability requirements, and whether dedicated hardware is needed. MPED scopes the software delivery, infrastructure, support, and operating costs separately so the organisation can compare local, hybrid, and cloud options on a like-for-like basis.
What is a sensible first FORGE project?
The strongest first project is usually a narrow, evidence-heavy workflow with a clear owner, repeatable inputs, measurable manual effort, and an explicit review point. MPED scopes the process, deployment option, integrations, and success measures before implementation.