James Towers
Technical Consultant at Bulien
At the start of every month the same file arrives and the same routine starts again. Clean the raw data, rebuild the report tabs, check the totals, and hope nothing has moved since last time. If your team has a workbook like this, it probably works. Until the person who built it is on holiday.
The good news is that the hard part is already done. A workbook that produces the right month-end figures is a complete description of the process: which data comes in, what gets calculated, how it is grouped and what the outputs look like. Automating month-end reporting is mostly a matter of lifting that logic out of the spreadsheet and putting it somewhere it runs the same way every time.
The tempting first step is to open Alteryx and start rebuilding tabs. We would hold off. Before anything is built, it is worth writing down what each tab does, which cells hold raw values and which hold formulas, and how every report tab uses the source data.
This is also where the awkward details turn up, the kind every long-lived workbook collects. A total that only works because someone pastes values over a formula. A lookup that silently drops a row. A category that exists in August and not in September. Finding those on paper is far cheaper than finding them in a board pack.
AI is genuinely useful at this stage. Given the workbook and a clear instruction, a model can inventory every sheet, list the functions in use and map each report back to its inputs. The instruction matters: ask it to describe and flag ambiguities, and tell it explicitly not to build anything yet.
The single most important design decision is that nothing about the current month is baked in. If August’s totals or August’s list of regions end up hard-coded, you have automated one month, not the process.
In practice that means the month becomes a single setting at the top of the workflow. Change it from August to September, run it again, and every report rebuilds from that month’s data. It sounds obvious. It is also the step most often skipped when a spreadsheet is copied into a new tool in a hurry.
Once the logic is written down, building the workflow is the quick part. In our own demonstration, James Towers gave an AI agent (Codex, using Alteryx’s official skills through the Alteryx MCP server) one month of synthetic contract data and a six-tab reporting workbook. It produced a complete, reusable Alteryx workflow in four minutes and 28 seconds.
That speed is real, and it changes the economics of automating small, repetitive processes. Two caveats come with it.
First, the result will not look like a workflow a person would build. James’s description was “an interesting-looking workflow, to say the least”, which is fair. It ran, it was correct, and someone would still want to tidy it before handing it over to a team to maintain.
Second, a fast build is worth nothing until the numbers have been checked. The test we would insist on is simple: run the new workflow for a month you already have, and compare every figure with the original spreadsheet. In the demonstration that meant contract count, gross annual contract value, discounts, net revenue, delivery and service costs, gross profit and margin. Every figure matched the August workbook, with the margin needing only rounding. Only then is it worth running September and October and treating the output as trustworthy.
This approach works best for processes that are repetitive and rule-based: the same inputs arrive, the same calculations run, the same reports go out. Contract reporting, regional and product breakdowns, sales performance and renewal risk lists all fit that description.
It works less well where the month-end depends on someone’s judgement each time. If a person decides which accruals to adjust or which exceptions to ignore, that decision needs to be agreed and written down first. Automating an undocumented judgement just makes the mistake faster.
There is also a choice about where the finished workflow lives. James built his on his own machine using the local Alteryx MCP server; pointing the same approach at Alteryx One would make the workflow shareable across a team instead. For anything that a finance team relies on monthly, the shared route is usually the right one.
James takes a month-end workbook, runs it through an AI agent and checks the result against the original figures, start to finish, in a few minutes of video.
We help finance teams move recurring reporting out of fragile spreadsheets and into processes that run the same way every month, with the checks left in. If your month-end still depends on one workbook and one person, have a look at our finance reporting automation work, or get in touch and tell us what takes the longest.