Part 6 of our series: AI-Powered Websites
There's a pattern we see repeatedly across B2B businesses. A client sends a document: a brief, a specification, a floor plan, a contract. Someone on the team opens it, reads through it, manually extracts the relevant information, enters it into their systems, and then follows up with the client to clarify the things the document didn't cover clearly enough.
This process happens dozens of times a week in businesses that deal with project-based work. It's slow, error-prone, and expensive in terms of the skilled human time it consumes. It's also one of the areas where AI can make the most immediate, tangible difference. Not as a nice-to-have, but as genuine operational infrastructure.
What we're building
For one of our clients, a company that designs and builds custom exhibition stands, we've designed an AI-powered client portal that fundamentally changes how they receive and process client information.
The portal works like this. A client receives a magic link via email. No password, no account creation, just a single secure link that grants access to their project space. They upload their documents: event floor plans, brand guidelines, design briefs, stand specifications, whatever they have.
The AI reads every document. Not skimming, not keyword-matching: reading. It extracts the structured information: stand dimensions, event dates, venue name, power requirements, branding colours, specific instructions. It identifies what's missing. And then it presents the client with a clear summary of what was understood and a set of intelligent follow-up questions to fill the gaps.
The client responds to those questions in a streaming chat interface, a conversation that's contextually aware of everything in their uploaded documents. If the floor plan specifies a 6m x 4m space but the brief mentions wanting a meeting room and a product display area, the AI can flag that the space might be tight for both and suggest discussing priorities.
On the business side, the project team receives a structured project summary extracted from all the documents and the conversation, ready to be dropped into their workflow. Information that previously took three emails and a phone call to gather is available in minutes.
Why magic links matter
A small but important design decision in this portal is the authentication model. We chose magic links over traditional username/password accounts for a specific reason: these are clients who interact with the system occasionally, not daily. Asking them to create an account, remember a password, and log in through a standard authentication flow creates friction at exactly the moment you want to reduce it.
A magic link arrives in their email. They click it. They're in. The link is time-limited and single-use for security, but the experience is frictionless. This isn't a new concept, as Slack, Notion, and others have popularised the pattern, but it's surprisingly rare in B2B client portals, which tend to default to traditional login flows because that's what the framework provides out of the box.
The result is dramatically higher engagement. Clients actually use the portal because using it requires less effort than not using it.
The document intelligence layer
The most technically interesting component is the document extraction pipeline. Different document types require different handling. A PDF floor plan contains spatial information that needs to be interpreted visually. A Word document brief contains structured and unstructured text. An email might contain critical details buried in a thread of forwards and replies.
The AI processes each document type appropriately, extracting not just the obvious structured data (dates, dimensions, quantities) but also the implicit information that a human project manager would pick up on: tone of urgency, design preferences expressed through adjectives rather than specifications, constraints mentioned casually rather than formally.
This is where current AI capabilities genuinely shine. Large language models are extraordinarily good at reading a document the way a human would, understanding context, inferring intent, and identifying what matters, but doing it consistently, instantly, and without the fatigue that causes humans to miss details on the twentieth document of the day.
Beyond exhibition stands
The pattern we've designed for this client is applicable to almost any business that receives client documents as part of their workflow. And there are plenty of them across our region.
An architecture practice in Cambridgeshire could use the same approach to process planning briefs. A client uploads their wish list, site photos, planning constraints, and budget indication. The AI extracts the structured requirements, identifies conflicts (the wish list exceeds what the budget likely supports, for instance), and prepares intelligent questions for the architect before the first meeting.
A legal firm in London could use it to process client onboarding documents. A new client uploads their existing contracts, company formation documents, and a description of their legal issue. The AI extracts relevant dates, parties, obligations, and potential risks, presenting the solicitor with a structured case summary rather than a pile of PDFs.
An accountancy firm in Essex could use it to process annual accounts preparation materials. Clients upload bank statements, receipt records, and previous year accounts. The AI categorises transactions, identifies anomalies, and flags items that need clarification, turning what's traditionally a weeks-long back-and-forth into a single focused review session.
A construction company in Bedfordshire could use it for tender document processing. Clients upload specifications, drawings, and bills of quantities. The AI extracts scope, deadlines, and key requirements, giving the estimating team a structured starting point rather than hundreds of pages to wade through.
In every case, the portal doesn't replace the professional. It replaces the administrative extraction work that precedes the professional's involvement. The architect still designs the building. The solicitor still provides legal advice. The accountant still prepares the accounts. But they start from a structured, verified summary rather than a raw stack of documents.
The streaming conversation
One feature worth highlighting is the streaming AI chat within the portal. After the initial document extraction, the client can have a conversation with an AI that has full context of their project. This isn't a generic chatbot. It's a contextually aware assistant that understands what the client has uploaded, what the project requires, and what information is still missing.
The streaming aspect matters for user experience. Rather than waiting ten seconds for a complete response, the client sees the AI's response appearing word by word, creating a sense of real-time conversation. It's a small detail that significantly impacts how the interaction feels.
The conversation history is preserved and available to the project team, creating a searchable record of every client interaction that's richer than email threads and more structured than phone call notes.
The operational impact
The business case for this kind of portal is straightforward. Every hour that a project manager spends extracting information from client documents is an hour not spent on project delivery. Every email thread chasing missing information introduces delay and creates opportunities for miscommunication.
A well-built client portal with AI document intelligence compresses the information-gathering phase from days to hours. It improves the quality of the extracted information because the AI doesn't skip fields or misread handwriting. It creates a better client experience because the client feels their documents were understood rather than lost in an email inbox.
And it scales. The portal handles its first client and its hundredth client with exactly the same efficiency. That's not something you can say about email-based processes.
Building trust through transparency
One design principle we've applied throughout this project is transparency about what the AI understands and what it doesn't. When the AI extracts information from a document, it shows the client what was found and asks them to confirm. When it isn't confident about an interpretation, it says so explicitly and asks for clarification rather than assuming.
This matters enormously for trust. The client isn't handing their documents to a black box. They're interacting with a system that demonstrates its understanding and invites correction. That collaborative dynamic is what makes AI-powered portals feel like a helpful tool rather than an intrusive automation.
Next in the series: [How AI Generates Structured Data and Technical SEO Faster Than Any Developer](/blog/ai-structured-data-technical-seo)