The problem, as you feel it
You ask an AI assistant to analyse something and it hands you back text. A lot of text, with headings and bullet points and the occasional table that does not quite line up. You copy it into a Word document, fix the formatting, try to make a chart from the numbers, give up, and paste a screenshot.
The output of the most capable writing and analysis tool ever built ends up as an unrendered document that nobody wants to read, that cannot be sorted or filtered, and that looks worse than what your team produced by hand ten years ago.
That is not a limitation of the AI. It is a default. The default output is markdown, a plain-text format designed for programmers, and everyone accepts it because nobody told them to ask for something else.
Ask for a web page instead
A modern AI can write a complete, self-contained web page as easily as it writes a paragraph. When you ask for the report as HTML, you get a single file that opens in any browser, on any device, and can do things a document cannot:
- Interactive charts. Real charts drawn from the data, with hover values, built with proper charting libraries (D3 and its relatives), not screenshots.
- Tables you can sort and filter. Click a column header. Type in a search box. Show only the accounts over ninety days.
- Tabs, collapsible sections, and a table of contents so a forty-page analysis reads like a well-organised site rather than a scroll.
- Print to PDF that actually looks right, with page breaks where they belong, because the page carries its own print layout.
- One link. Host it privately and send a URL. The client opens it on their phone. No attachments, no “which version is this.”
- It can be a small application. A calculator with your pricing rules. A what-if model where the reader drags a slider. A checklist that remembers what was ticked.
The same request that used to produce a wall of text now produces something you would be comfortable putting in front of a board, a client, or a lender, and it took the AI no longer to make.
Why this matters more than it sounds
Most of the value of analysis is lost between the analysis and the decision. A report that is hard to read gets skimmed. A chart that is a screenshot cannot be interrogated. A document that is out of date the moment it is sent gets forwarded anyway. When the report is a page, the reader can check the number that surprised them, see the trend behind the summary, and print the one section they need. The report becomes something people use, rather than something they receive.
There is a second effect that is easy to miss. Once the AI is producing pages, the pages can be regenerated. The Monday report is the same page with new data. The client proposal is the same page with a new client. The quarterly review is a template that fills itself. Reporting stops being a task and becomes a thing that happens.
Take a business like this one
The business in this section is a composite of the kind of company I talk to, not a named client. The numbers are the shape of the problem, not a case study.
Picture a fractional CFO practice: one principal, two associates, eleven small-business clients. Every month each client gets a financial review, which means an associate pulls the numbers, writes a Word document from a template, builds four charts in Excel, pastes them in, exports a PDF, and emails it. Two days per associate per month, and the clients admit they mostly read the first page.
What I set up for a practice shaped like this:
- A report template as a web page. The practice’s branding, their standard sections, a print stylesheet, and chart components ready to receive data.
- A data step that pulls each client’s numbers from their accounting system and prepares them for the page.
- An AI writing step that fills the narrative sections from the numbers, in the principal’s voice, flagging anything unusual and keeping the tone consistent across all eleven clients.
- Interactive charts and tables in every report: cash runway you can hover, an expense table you can sort by change, a scenario slider for the one decision each client is facing that month.
- Private hosting with a link per client per month, plus a one-click PDF for the clients who still want the file.
- A review step where an associate reads the generated page, edits any sentence she disagrees with, and publishes.
In a practice shaped like this you should expect the two days per associate to become a couple of hours of review, the reports to be materially better than the Word versions (because they can now show the trend, not just the number), and, the part the principal cares about, clients who start replying with questions about page three because they read page three.
Doing this in your own business
You do not need a consulting practice to benefit. The pattern works for any recurring report: sales reviews, project status, board packs, marketing summaries, inspection reports, client deliverables. What it takes is a template built once, a reliable way to feed it data, and an AI step set up to write into the template rather than into a blank chat window. That is what I build.
What it costs to run
Generating a page costs cents in AI usage. Private hosting for reports is a few dollars a month. The cost is the setup: designing the template around your actual report, connecting your data, and tuning the writing until it reads like you.
What I need from you
Two or three of your current reports, the ones you are proudest of and the ones you are embarrassed by. Where the numbers come from. An hour to talk through what the reader actually does with the report. Then a couple of rounds of feedback on the generated pages until the voice and the layout are right.