AI & automation
AI put to work on the jobs it is actually good at
Reading enquiries, sorting them, pulling detail out of documents, summarising long threads and drafting replies for someone to approve. These are the language-heavy tasks that quietly consume a growing team, and they are where AI earns its place. What you get is a working business process, not a chatbot and not a folder of prompts.
Beneath this: business automation for the rule-based work, and missed call & AI receptionist for phone-driven enquiries.
In short
Porthline builds AI into real business workflows: enquiry triage and first-draft replies, document and PDF extraction, email and conversation summaries, internal knowledge assistants, lead sorting and routing, reporting support, and AI features inside portals and internal tools. Rules handle the predictable steps, AI handles the language, and a person approves anything that carries risk. Every run is logged so the work can be checked.
Rules first. AI where it earns its place.
Most of what removes repetitive work in a business is not AI. It is clear, reliable rules moving information between the tools you already have, triggering the next step and chasing what has not happened. That is steadier, cheaper to run and easier to reason about, and it is covered on the business automation page.
AI goes on top of that, for the parts where something has to be read, understood, summarised or written. It is genuinely good at those, and genuinely unsuited to being the thing that decides. People stay in control wherever judgement matters. If a rule would do the job better, I will tell you to use the rule.
What I build with AI
Each of these is a job a person is doing by hand today, not a feature looking for a use.
Enquiry triage and first drafts
Incoming enquiries are read, categorised and routed, with a first-draft reply prepared for someone to check and send.
Document and PDF extraction
Pull the fields you actually need out of forms, invoices, specifications and scanned paperwork, and put them where they belong.
Email and conversation summaries
Turn a long thread or a call transcript into a short summary and a clear next action, so nobody re-reads twenty messages.
Internal knowledge assistants
Let staff ask questions across your own documents, procedures and records, and get an answer with its source attached.
Lead sorting and routing
Separate the serious enquiries from the noise and get each one to the right person, with the detail already gathered.
Reporting support
Pull the week's numbers together and draft the commentary around them, ready for a person to check and adjust.
Drafting for approval
Quotes, updates, follow-ups and standard replies drafted from your own templates and tone, then held for a human to send.
AI inside a portal or internal tool
The same capabilities built into the system your team already works in, rather than another separate app to remember.
The first of these, enquiry triage with a draft held for approval, is the one I have built and recorded as a demonstration on a fictional business, evaluation and limitations included. Where these sit inside a system your team logs into rather than a background process, they are built as part of a portal or internal tool.
Assistant, workflow or agent?
These words get used loosely. Here is what each means in a Porthline build, what it is allowed to read and do, and where a person stays in the loop.
AI assistant
Answers a question or drafts something when a person asks it to, using your own documents and records where that is set up.
What it may read and do
Reads what it is pointed at. Produces a draft or an answer. Takes no action on its own.
Where a person stays in
The person asking decides what to do with the result, every time.
AI workflow
A fixed sequence that runs on a trigger, such as a new enquiry or a document arriving, with an AI step inside it for the reading or drafting.
What it may read and do
Sorts, extracts, summarises or drafts at its step; rules do everything before and after it. Runs the same route each time.
Where a person stays in
Anything that reaches a customer or changes a record you care about is held for a person to approve. Cases that do not fit are flagged, not guessed.
AI agent
A workflow that is allowed to choose between a small number of agreed actions, for example which template to use or which record to update, within limits set in advance.
What it may read and do
Reads only the sources it has been given access to, and does only the actions it has been given permission for. Each permission is agreed in writing and can be narrowed.
Where a person stays in
Bounded actions can run; anything consequential still waits for a person. Every action is logged so it can be checked, and the boundaries are the design, not an afterthought.
Two practical dependencies apply to all three. Each connects to your tools through whatever those tools expose, so what it can read or change depends on the CRM, inbox or booking system you already use, and a change on their side can break the connection. That is why every build is monitored after launch, and why looking after the connections is part of ongoing support rather than something left to chance.
What is delivered, and what is not yet
Delivered
The Adventra customer portal, live and in daily use, with rule-based automation inside it: status changes trigger customer emails without anyone typing them. It is a portal, not an AI system. Two client websites are also live.
Demonstrations
One published AI demonstration: an enquiry assistant shown on a fictional business, with a recording, screenshots and the evaluation written up, limits included. It is a prototype, not client work. The Work page also has a live demo portal you can open, and automation concepts labelled as concepts.
What you can commission today
A £750 Workflow & Systems Review, or an automation sprint from £2,000: one process mapped, built, tested and monitored, with anything the AI drafts held for a person to approve. The applications on this page describe how that work is done, not client case studies. No AI client project is live yet, so the first piece is kept small enough to judge.
How it is made dependable
AI that works once in a demo is easy. AI you can leave running on a Tuesday afternoon is a different job, and most of that job is the plumbing around it.
Rules do the deterministic parts
Moving a record, sending a confirmation, setting a status: those are rules, and rules are more reliable than a model. AI is used only where language is involved.
A person approves what matters
Anything that goes to a customer, commits you to something or carries risk is drafted, not sent. You decide which steps need a human and which do not.
Everything is logged
Each run records what came in, what the AI did with it and what went out, so you can check the work rather than trust it blindly.
It is tested on real cases
Including the awkward ones. If it cannot handle a case reliably, that case is routed to a person instead of guessed at.
Exceptions surface, they do not vanish
When something does not fit the pattern, it flags for review. Silent failure is the main way automation loses people's trust.
Your data stays yours
I am explicit about what is sent where, what is retained and what is not. Where information should not leave your systems, the workflow is built so it does not.
Where I would talk you out of it
Start with one process
The first piece of work should be small enough to judge. One workflow, built and tested against real cases, running for a few weeks before anything else is added. That way you find out whether it holds up before committing to more.
Automation sprints usually start from £2,000. If you are not sure which process is worth doing first, that is what the Workflow & Systems Review is for.
AI and automation FAQs
Which task would you hand over first?
Tell me the language-heavy job your team repeats most: the enquiries, the documents, the summaries, the drafting. I will tell you honestly whether AI is the right tool for it, or whether a plain rule would serve you better.