AI & Workflow Systems
Lean automation and internal systems that remove manual operating drag without adding another layer of complexity.
Build the operating system before automating it.
Automation works when the underlying process is clear. I start by understanding how work enters the business, where decisions happen, what information is required, which steps repeat, and where professional judgement needs to stay human.
From there the work is practical. Connect the systems that should already talk to each other. Remove duplicate data entry. Generate repeatable tasks and outputs automatically. Introduce AI where it can classify, extract, summarise, or draft. Build a dedicated internal interface only when the existing tools create too much operational friction.
The aim is not more software. It is a business that is easier to run.
Six patterns behind most requests
These are the operating patterns that usually sit behind a request for automation. Most businesses recognise three or four of them.
Repetitive administration
The same client, job, or project information is being entered in several places, and standard work is recreated every time.
Fragmented tools
The systems work individually, but people spend too much time moving context between CRM, email, documents, spreadsheets, and task tools.
Knowledge bottlenecks
Important processes, precedents, and judgement sit with one or two senior people and are difficult for the wider team to use consistently.
Weak workflow visibility
Leaders cannot quickly see what is active, what is blocked, what is overdue, or what needs a decision.
Manual documents and reporting
Standard correspondence, summaries, reports, and internal briefs are assembled by hand despite following repeatable patterns.
AI without an operating model
Employees are already using AI, but it is disconnected from trusted company information, process controls, and the systems where work actually happens.
What I can build
Five areas of work. Most engagements use two or three of them, in the order the operating problem requires.
Workflow automation
Map the process, remove unnecessary steps, and automate the repeatable work around it.
- Structured intake
- CRM or job creation
- Task generation
- Approvals and reminders
- Status transitions
- Onboarding, hand-offs, and closure
AI-assisted operations
Use AI for the parts of knowledge work where it creates leverage, while keeping important decisions and external outputs under human control.
- Extraction and classification
- Summarisation
- Document comparison
- Missing-information checks
- Internal drafting
- Status briefs and structured capture
Knowledge and internal AI
Turn approved procedures, templates, guidance, and accumulated internal knowledge into something the team can actually find and use.
- Controlled knowledge base
- Context-aware internal assistant
- Retrieval with source references
- Role and job-specific access
- Approved template retrieval
Connected operating systems
Keep specialist systems as systems of record and connect them into a cleaner operating workflow.
- CRM
- Microsoft 365 and Google Workspace
- Document storage
- Email and forms
- E-signature
- Finance and operations platforms
Internal workspaces
When the team is spending too much time moving between systems, create one internal interface over the tools already doing the underlying work.
- Operational dashboard
- Customer, job, or matter view
- Task queue
- Documents and timeline
- Approvals
- Embedded AI assistance
One workflow, existing systems underneath
I do not rebuild tools that already work. CRM, document storage, email, and specialist platforms can stay where they are. The work is to create a clearer way to coordinate work across them and, where it is warranted, a single internal interface for the team.
- Internal workspaceOnly where switching between systems is a real constraint
- Dashboard
- Work
- Tasks
- Knowledge
- AI assistance
- Workflow coordinationWhere the business rules live
- Workflow
- Automation
- Permissions
- Audit
- Systems of recordKept in place, connected through APIs
- CRM
- Documents
- Finance and operations
- AI routerTask type decides the provider, not the other way around
- General model
- Fast extraction model
- Specialist service
- Human review
Use the right technology for the work
I am not tied to a single AI model. Provider choice follows the workload, information sensitivity, quality required, and controls the business needs.
The business workflow should own the rules. The underlying model should remain replaceable.
| Work type | Typical approach |
|---|---|
| Classification and extraction | Fast, cost-efficient approved model |
| Summaries and drafting | General reasoning model |
| Company knowledge | General model plus approved internal sources |
| Specialist research | Appropriate specialist provider |
| High-risk output | Source-backed workflow plus human review |
| High-volume simple work | Lowest-cost model that meets the quality threshold |
Four stages, and what each one earns
Four stages. Each one has to earn the next, and most engagements stop before the fourth.
- 01 · Diagnose
Map how work currently moves through the business, where information is duplicated, where decisions occur, and what is consuming senior attention.
Typical output- Current-state workflow
- Priority problems
- System inventory
- Risk and control points
- Automation shortlist
- 02 · Simplify and automate
Standardise the process first, then connect the existing systems and automate the repeatable work around them.
Typical output- Structured intake
- Workflow automation
- CRM and job structure
- Document workflow
- Task generation and reporting
- 03 · Add AI where it earns its place
Introduce AI for defined workloads such as extraction, summarisation, retrieval, and drafting, with controls appropriate to the work.
Typical output- AI-assisted workflows
- Knowledge retrieval
- Provider routing
- Source handling
- Human approval rules
- Usage and cost controls
- 04 · Consolidate where necessary
If the team is still moving across too many systems, build a focused internal workspace over the existing stack rather than replacing everything underneath it.
Typical output- Internal dashboard
- Unified work view
- Task and document access
- Embedded AI
- Operational reporting
The smallest useful solution comes first. Custom software is used when it removes a real operating constraint, not because a custom build is more interesting.
What this looks like in practice
Illustrative patterns rather than client work. They show the shape of a typical result.
- Service intake to live job
- An instruction arrives through a structured form. The system identifies the client, creates the job, creates the document structure, generates the correct checklist, drafts an acknowledgement, and gives the owner a clear next action.
- Internal knowledge assistant
- Approved procedures, templates, and guidance are available inside the work context. Someone can ask a question, get a concise answer, and open the source material behind it.
- Document-heavy workflow
- New documents are classified, key information is extracted, the work record is updated, and the responsible person sees the action generated from it.
- Management visibility
- Instead of compiling a weekly status report by hand, leaders see active work, overdue actions, blocked items, and workload from the same operating data.
- Single internal workspace
- The team works from one interface while CRM, document storage, email, and specialist tools remain behind it as the underlying systems of record.
Automate the repeatable work, keep the judgement
AI should not be inserted into every decision. I separate repeatable information work from decisions that require professional judgement, accountability, or authoritative sources.
Important outputs can be designed to require review, approved knowledge can be held separately from live client information, and specialist research tools can be used when a general model is not the right source.
Scope, sequence, and cost
Available as a defined project, hourly support, or an ongoing optimisation retainer. Defined projects set the blueprint, deliverables, dependencies, acceptance criteria, and end point before a larger build is committed to.
Technology is selected to fit the workflow. I work with existing SaaS, low-code automation, APIs, and targeted custom software rather than moving every client onto the same stack.
Engagement Model
Engagements are shaped around the work, not a fixed format. Defined projects set deliverables, dependencies, acceptance criteria, and an end point. Hourly support suits bounded advice or execution. Ongoing retainers set available capacity, priorities, decision rights, cadence, and a review point.
Delivery is senior and hands-on. Larger builds are scheduled rather than run in parallel, so timing is agreed upfront and the work gets the attention it needs.
What comes before this
Operating Model & Scale Readiness
Where accountabilities, hand-offs, and decision rhythm are the constraint, that work comes before any automation.
Fractional COO
Ongoing senior ownership of company-wide operating practices, rather than a defined systems project, points to this instead.
Further reading.
What Automation Costs You When You Use AI Instead
Using a model for work a rule could do has four costs, and the invoice is the smallest of them.
Automation Is Not AI: Where the Boundary Sits in a Small Business
Automation follows rules you write. AI produces judgement you did not write. The difference determines what you can rely on.
Zapier, Make, or AI? How to Choose the Right Tool for the Work
A decision rule for founders choosing between a workflow automation tool and an AI model, and why most useful systems need both.
Frequently asked questions
Do you start with the AI tool or the business process?
The process. I map the work, information, decisions, and controls first, then select the smallest useful technology to support it.
Do we need to replace our existing systems?
Usually not. The preferred approach is to keep useful systems of record in place and improve the workflow between them.
Can you build a dedicated internal application?
Yes, where the workflow justifies it. I generally validate the process and automation first, then build a focused internal workspace if moving between systems remains a material operating constraint.
Are you tied to one AI model?
No. The architecture can route different workloads to different approved providers or specialist tools, and the business workflow calls the task type rather than a fixed vendor name.
Can AI send work directly to customers?
It can technically support external workflows, but important communications and professional outputs should use approval controls appropriate to the business and the work.
Start where the time is going.
If a critical process is held together by inboxes, spreadsheets, repeated admin, or one person's memory, start there. I will scope the workflow, identify what should change, and determine whether automation, AI, or a more focused internal system is the right response.
I aim to respond within two business days.