The report you can talk to

4 Jul 2026 · 3 min read

Modern business intelligence lets you ask a question in everyday words and get a sourced answer back, ready to check. Below are ten questions a finance director asks and the kind of grounded answer a well-built model gives, each pinned to where the number came from and when it was last true. The answer is only as good as the data underneath, which is why we get that right first.

A happy client against a neon backdrop

Modern business intelligence lets you ask a question in everyday words and get a sourced answer back, ready to check. Below are ten questions a finance director asks, and the kind of answer a well-built, trustworthy model gives, each one pinned to where the number came from and when it was last true. (Illustrative examples on sample data.)

Ten questions, ten sourced answers

"Why did margin drop last week?" Margin fell to 21.8%, down 1.6 points. Nearly all of it is one thing: freight on the Benelux lane jumped 14% after a carrier surcharge on 28 June. Volume and pricing held. Sourced from Xero · the freight ledger · 3 invoices · as of Tue 09:14.

"Which customer is slipping?" Northgate Supplies is down 22% over the last three months against their trailing average: order frequency has halved since April, though average order value is steady. Worth a call before it becomes churn. Sourced from the sales ledger · CRM order history · rolling 90 days.

"Are we going to hit the Q3 number?" On current run-rate, Q3 revenue lands at £2.14m, about 3% under the £2.2m plan. The whole gap is in own-brand, which is tracking two weeks behind; factored lines are ahead. Sourced from the sales forecast model · budget v4 · last night's refresh.

"What's my real margin by channel, after ad spend and fees?" After ad spend, payment fees and fulfilment, Amazon nets 9.4% contribution, the webstore 21.8%, and trade 27.1%. Amazon's headline revenue flatters a thin true margin once the fees are in. Sourced from Shopify · Amazon · Stripe · the ad platforms · blended.

"Where will cash be at the end of next month?" Projected closing balance is £312k, dipping to around £180k on the 18th when the quarterly VAT and two large supplier payments land together. That is the tight point to watch. Sourced from Xero · the payment calendar · confirmed POs.

"Which stock is tying up cash and not selling?" £148k sits in lines with more than 120 days of cover. Of that, 60% is in three slow SKUs from the spring range. Clearing them would release roughly a month of working capital. Sourced from the inventory system · sales velocity · rolling 12 weeks.

"Why doesn't sales' revenue match finance's?" The £41k gap is timing, not error: sales counts an order when it's placed, finance when it's invoiced, and 118 orders that shipped in the last three days of the month invoice in the next. Same truth, two clocks. Sourced from the CRM · the sales ledger · reconciled.

"Is the warehouse hitting on-time dispatch?" On-time dispatch is 94.6% this month, down from 97%, and the dip is entirely the north pick face, where two staff were off last week. Everything else is on target. Sourced from the WMS · dispatch logs · rolling 30 days.

"What changed in overheads this month?" Overheads rose £9.2k. Two-thirds is the annual insurance renewal (a one-off) and the rest is energy on the new unit coming online. Underlying overhead is flat. Sourced from Xero · the nominal ledger · prior-year compare.

"Which supplier is quietly eating our margin?" Meridian's landed cost is up 6% since January, but it's the carriage, not the unit price, driven by smaller, more frequent deliveries. Consolidating to fortnightly orders recovers about 1.5 points. Sourced from purchase invoices · the freight ledger · six-month trend.

What makes this work

This works when the numbers underneath are right, reconciled and current, so that when you ask, the answer can be trusted and traced. Every answer above names its sources and its timestamp, giving you a figure you can check before you take it to the board.

So the work isn't the asking. That part is easy now. The work is everything that makes the answer true: one governed source of truth, agreed definitions, and automatic refresh underneath. Get that right and the report stops being a wall of charts you interrogate, and becomes the colleague who already has the answer.

Frequently asked questions

Can you really ask a BI report a question in everyday words?

Yes, tools like Power BI Copilot and natural-language query let you ask in everyday words and get an answer back. The catch is that the answer is only as good as the model underneath: ask an ungoverned, unreconciled dataset and you get a confident guess. We get the data right first, so the answer you get is one you can trust and defend.

Is this ChatGPT for my business?

It's the same idea (ask, get an answer) grounded in your governed business data. A good answer cites where each figure came from and when it was last true, so you can check it. That provenance is the whole difference between a useful answer and a plausible one.

What do we need in place for this to work?

One governed source of truth with agreed metric definitions and automatic refresh. Get that foundation right and answers become reliable when you ask in everyday words.

Want this set up and handled for you?

Start with a fixed-price Trusted Numbers Review: two weeks, written findings on why your figures disagree, and one fixed price to put it right.