Talking to your data, also called natural-language analytics or conversational analytics, means asking a question about your business in plain English and receiving an answer computed from your own records. The software translates the question into a query, runs it against structured data, and returns a figure with the products, periods, and causes behind it.
Business intelligence tools of the 2000s and 2010s put a dashboard on top of a database; the questions were fixed by whoever built the dashboard, and anything else required someone who wrote SQL. Vendors then added search bars that mapped typed phrases to prebuilt metrics, which worked when the phrase matched and failed when it did not. Large language models changed the translation step: a model can now read "which products lost money last quarter after fees" and produce the query, the comparison, and the explanation without a prebuilt report for that question. The database underneath did not change. The interface did.
A seller's profit lives in six places Amazon never joins: ordered sales, referral and fulfillment fees, ad invoices, refunds and reimbursements, storage charges, and purchase orders with freight. A question like "why is margin down on this SKU" needs all six on one line, per product, per period. Nobody builds a dashboard tile for every SKU and every question. Natural-language analytics is the way a seller with 300 products asks about product 187 without opening a spreadsheet.
The catch is the word "computed." The answer is computed from what the data holds. Ask about margin and the data has to hold margin's ingredients at the SKU level: landed COGS on FIFO layers, fees per order, ad spend per SKU, refunds per SKU. Ask the same question over data that lacks them and the software either declines or, worse, computes a confident number from what it has. The general version of this problem is laid out in the data quality problem behind AI accounting tools.
Illustrative numbers. A seller asks, "What was my margin on the walnut cutting board in June?"
Over raw Seller Central exports. The Business Report shows 300 units and $8,997.00 in ordered product sales at $29.99. The seller pastes it into a general-purpose chatbot with "my cost is about $6 a unit." The referral fee is 15 percent, the published rate for Home and Kitchen on sell.amazon.com. The chatbot computes: $8,997.00 minus $1,349.55 referral minus $1,800.00 COGS equals $5,847.45, a 65.0 percent margin. It reports that figure with no hedge.
Over reconciled books. The same 300 units and $8,997.00. Referral fee $1,349.55, from the settlement lines. Fulfillment fee at an illustrative $4.40 per unit, $1,320.00, also from settlement lines. Ad spend attributed to the SKU from the June advertising invoice, $1,480.00. Refunds, 21 units, $629.79. FIFO landed COGS at $7.15 per unit, because the May PO landed with duty and a freight surcharge and the older $6.00 layer sold out in May: $2,145.00.
Contribution: $8,997.00 minus $1,349.55 minus $1,320.00 minus $1,480.00 minus $629.79 minus $2,145.00 equals $2,072.66, a 23.0 percent margin.
Same question. Same seller. Same month. One answer is 65 percent and one is 23 percent, and the whole difference is what the data held. The chatbot did nothing wrong with the numbers it was given. It was given a third of them.
Crunch is the natural-language analytics layer inside ConnectBooks. Sellers ask about profit, margins, advertising, inventory, fees, and cash flow, and the answer is computed from data ConnectBooks has already reconciled from Amazon, Shopify, Walmart, TikTok Shop, and eBay into QuickBooks or Xero. Each Amazon settlement is split into its fee, refund, and reimbursement lines against the SKU; COGS runs on FIFO with landed cost; ad spend is allocated by product. The profit reports built on that layer are what Crunch reads, and every answer follows the same chain: what changed, over which periods, on which products, why, and what to do next.
The reason it starts with settlement reconciliation rather than with the chat window is the worked example above. The chat window is the easy part.
What Crunch cannot do: fill in a landed cost that was never recorded, know that a supplier raised prices before the PO posts, or act in Seller Central. It answers and recommends. Actions stay with the seller.
The interface is similar. The data is not. A spreadsheet holds what you exported; reconciled books hold every transaction line with cost attached. The chatbot answers from the export, and the export is incomplete on cost by construction.
Because the questions are. "Why is margin down" has an answer only at the product level: which fee band changed, which PO layer sold through, which campaign overspent. Account-level data can say margin fell. It cannot say where.
Yes, in two ways. It can compute correctly on incomplete data, which is the worked example above. And a general-purpose model can misstate totals on a long paste. The defense against the first is reconciliation; against the second, a system where every figure traces to a source line.
No. It replaces the pivot table you were going to build to answer one question. The reconciliation still has to happen, the close still has to be reviewed, and tax questions still go to a CPA.
Ones with a product, a period, a metric, and a comparison. "Which SKUs had a contribution margin under 10 percent in August versus July" works. "How is business" does not.
Running an e-commerce business comes with plenty of challenges, but ConnectBooks is here to make your life easier. With real-time insights, seamless integrations, and detailed tracking of your profitability and inventory, you can stay ahead of the game. Whether you’re selling on Amazon, Shopify, Walmart, TikTok or eBay, ConnectBooks helps you manage your finances with 100% accuracy and confidence, so you can focus on growing your business.
Ready to level up? Start making smarter, data-driven decisions every step of the way. Try ConnectBooks Free Today or Schedule a Demo
