Consultancy / Forthreason AI

Ambitious AI.
A clear scope.

AI engineering and security consultancy for organizations with sensitive data, consequential workflows and accountable process owners. Start with the work that needs to improve; define the boundaries before connecting a model.

The first step can be small: a paid assessment of one workflow, its risks and acceptance criteria. We agree scope and fee before work begins.

Three ways to move the work forward.

Commission an AI security review, an agent-permission assessment or a scoped integration.

An engagement may begin with an assessment, a scoped implementation or a defined support requirement. The scope, responsibilities, deliverables and acceptance criteria are agreed before delivery.

Find the right problem.

Assess the workflow, the data, the risks and the business case. Define what useful and acceptable performance would mean before building.

  • Workflow and data-flow mapping, including the people who own each decision.
  • A review of permissions, failure modes and feasible deployment choices.
  • For one agreed workflow: permitted and prohibited actions, representative success and failure cases, and a prioritized test plan. Test execution is scoped separately.
  • A prioritized recommendation, with explicit assumptions and acceptance criteria.

A scoped assessment and a clear recommendation.

Related founder / Bento workQueenie's Mayflower
Discuss your workflow

Make the boundaries explicit.

Design and implement AI workflows with scoped access, appropriate evaluation, human approval and evidence of what happened.

  • A bounded implementation connected to the agreed data and systems.
  • Evaluation of legitimate tasks and relevant failure cases.
  • Documented access controls, approval points and a handover for the people operating it.

A defined implementation, with testable acceptance criteria.

Discuss your workflow

Keep ownership clear.

Agree how the system will be maintained, reviewed and supported as its use changes. Make responsibilities and escalation paths explicit.

  • An agreed maintenance and review schedule.
  • Defined responsibilities for incidents, changes and escalation.
  • Checks after relevant model, data, permission or software changes.

A support scope that matches the actual system.

Related founder / Bento workQueenie's Mayflower
Discuss your workflow
Selected work / Founder and Bento projects01 / 03

Experience that shapes the questions.

Forthreason builds on Matthew's work through Bento. These projects show specific engineering and verification work; they are not presented as commissions delivered by the new company.

01 / Financial-data workflow

Skyplex Properties

Developed workflow / Inspected reporting outputs

Bank exports turned into structured Excel and PDF reporting, with reconciliation checks and unresolved items flagged for accountant review.

Read the scope and verification summary

A custom Python workflow imports and orders bank transactions, applies client-specific classification rules and produces monthly, supplier, recurring-payment and property-work schedules. Balance and period checks help keep the reporting traceable.

What the records demonstrate

  • Custom generator and January-December 2025 Excel/PDF outputs inspected.
  • Rule-based classification, reconciliation checks and control totals.
  • Explicit unresolved items and accountant notes.

Accountant review remains part of the workflow. This is data automation, with no claim of autonomous AI accounting or measured time savings.

Have a similar workflow? Tell us about it
02 / AI integration / Application security / CRM

Queenie's Mayflower

Development and verification records

A model-backed support assistant, access-controlled application and restaurant tools, alongside a CRM with controlled sending and unsubscribe handling.

Application security reviews covering authentication, staff roles, conversation access, signed tokens and replay protection.

Read the scope and verification summary

Work spans an AI assistant, customer and staff workflows, POS software and a separate CRM. Development records cover authenticated conversations, user-owned history, usage limits and human-escalation proposals. CRM records cover cleaning, deduplication, operator-controlled batches and provider send IDs.

What the records demonstrate

  • A real-model opening-hours answer and refund-escalation proposal recorded in a development API test.
  • Application security reviews covering authentication, staff roles, conversation access, signed tokens and replay protection.
  • CRM cleaning and batch-send records, suppression checks and unsubscribe handling.

The recorded assistant test surfaced a domain-configuration failure in handover email. These records do not establish completed production handover or measured customer-retention gains.

Have a similar workflow? Tell us about it
03 / Gaming operations / Product engineering

Confidential regulated operations

Product development / Potential pilot discussions

Mobile, staff-dashboard and backend workflows for waiting lists, seating and operational records, with a synthetic-data tournament-management concept.

Read the scope and verification summary

A confidential product-development project connects player and staff workflows: waiting lists, seat calls, seating assignments and operational event records. An interactive tournament concept explores registration, clocks, table movements and reconciliation using synthetic data.

What the records demonstrate

  • Engineering work across a mobile application, staff dashboard and backend services.
  • Connected waiting-list, seat-call and seating-assignment workflows.
  • An interactive tournament-management concept using synthetic data.

A possible pilot is under discussion. This case describes product development, not an accepted production deployment or client endorsement.

Have a similar workflow? Tell us about it

A useful first conversation

What are you working on?

Five short questions, then project and contact details. A clearer brief to discuss together. Start wherever you are, including "not sure yet".

Your answers stay in this page until you choose to copy or email your brief. Share business contact details only; leave out passwords, confidential material and personal records about others.

Client data and materials do not automatically become research material. Reuse or publication requires a separate scope and permission.

Prefer to contact us directly? Guided interview

Your brief will bring together

  1. 01The workflow and the change you need
  2. 02Systems, constraints and a useful outcome
  3. 03Project context and a person to follow up with

Answer, review, then decide what to share.

01

Built around the actual workflow.

A document assistant, an internal integration and an agent that changes records need different controls. We consider the information involved, the actions the system can take and the person responsible for the outcome.

  • AI readiness and workflow assessments.
  • Security reviews of existing AI systems and integrations.
  • Private document workflows and AI-assisted internal tools.
  • Agent permissions, approval boundaries and reviewable action records.
02

A useful first conversation.

Tell us what the process does today, where it falls short and who owns the decision. A broad description is enough to begin; confidential documents, credentials and personal records are not needed in the first message.

We can then establish whether the right next step is an assessment, an implementation brief or further discovery. Timing, cost and any ongoing support are defined for the specific engagement.

03

Security is part of the design.

We treat access, data handling, evaluation and human intervention as implementation decisions. Any sector-specific legal or regulatory requirements need to be identified for the actual use, customer and jurisdiction.

Support coverage and response commitments exist only when expressly agreed. Research publication and reuse of client materials require their own permission.

Begin with a worthwhile question.Forthreason AI

What are you
thinking about?

A sensitive workflow. A difficult research question. A system that needs clearer boundaries.

Start a conversation