AI Implementation
Build Your AI BI Analyst: The Report That Builds Itself
Most organisations under a hundred people sit at one of three reporting levels, and it is worth knowing which one you are on before buying anything.
Level one: the spreadsheet ritual. A capable person spends part of every week exporting, pasting and tidying so a meeting can happen. The knowledge of how it works lives in their head.
Level two: the dashboard nobody opens. A BI tool was purchased, a consultant built some charts, and eighteen months later the team still asks the capable person for a spreadsheet.
Level three: reporting that arrives, and answers back. The regular report turns up on time without anyone building it, and a follow-up question gets a reply the same hour.
An AI analyst is the quickest route from the first level to the third, and it usually skips the second entirely.
The work it absorbs
We estimate 50 to 80 percent of analyst hours at this size go on three activities:
- Producing the recurring report. Same sources, same layout, every week or month.
- Re-cutting it on request. By program, by location, by funder, year on year, with or without one-offs.
- Scanning for what changed. Reading rows to find the two that deserve a conversation.
All three are rule-following over structured data. A service business tracking utilisation, a not-for-profit reporting outcomes to funders and a wholesaler watching margin have the same problem in different clothes.
Build sequence
1. Capture the current process. Whoever builds the report today records a screen walkthrough explaining each figure. We treat that recording as the spec.
2. Stand up a small data layer you own. Each source system gets a connector: an API where available, a timed export where not. The connectors write tidy tables into a modest database in your own cloud account.
3. Reproduce the report. Generate it from that layer and run it next to the hand-built one. Differences are investigated until there are none.
4. Retire the ritual. The scheduled version becomes the real one.
5. Open it to questions. Leaders ask in ordinary language through Claude, and new cuts that prove popular are added as named, tested views.
Every run is logged, and any figure can be traced to the rows beneath it.
Guard against the two-answers problem
The cheaper analysis gets, the more of it appears, and sooner or later two confident people bring conflicting figures to one meeting. Settle this up front:
- Each headline measure has a single authoritative definition and a named owner.
- Anyone may explore, and exploratory figures carry that label.
- Analysis of someone else’s area goes to them first.
This is a culture rule as much as a technical one, and it needs to come from the top.
What it does to your software bill
Because the report and the questions run on your own data layer, the BI subscription becomes optional rather than structural. If your core system later ships an AI reporting feature, you can adopt it, ignore it or replace it without rebuilding anything.
Is someone on your team still at level one? Tell us what the report is made of and we will tell you how much of their week we can return.
If this sparked something, let's talk.
No pitch, no pressure — just a conversation about what you're working on.
Let's talkRelated posts
AI Implementation
Build Your AI Executive Assistant: What It Does, What It Never Does
AI can carry 50 to 80 percent of the repetitive work of an executive assistant: triage, drafting, scheduling, chasing, the morning brief. Here's how we build one, the rule that keeps it safe, and whether a hand-written email still matters.
AI Strategy
Did AI Just Kill BI? What to Stop Paying For, and Where to Spend Instead
Traditional business intelligence spent most of its budget on integration: ETL, data wrangling and merging systems. AI has collapsed that cost. Here's what leaders should stop funding, what deserves the money instead, and the questions to put to your BI vendor.
AI Implementation
Build Your AI Middle Manager: Tracking, Chasing and Reporting on Autopilot
Good middle management is mostly diligence on a schedule: tracking commitments, chasing at the right moment, unblocking, and reporting upward. AI can carry 50 to 80 percent of it. Here's how we build an AI manager, and the line it never crosses.