Your company knowledge, answered and checked.
Your own internal AI agent, configured around the knowledge and business data your team already has: product documents, exported correspondence, structured spreadsheets and the business rules your team saves.
No ERP replacement and no retrained model. Answers show their sources and a person reviews them before anyone relies on them.
Works from documents and spreadsheet or CSV exports, including exports from ERP systems.
The internal agent layout
The internal agent delivered for a client, reproduced with its layout and navigation. Branding and content are replaced; no client data appears. Read the customer story.
Ten-second illustrative film for the internal assistant: records become one reviewable answer with its context.
Try the guided demo01 / What it is configured around
Your knowledge. Your rules.
- Sources your team chooses: product documents, price lists and specifications, exported correspondence, competitor notes and structured business spreadsheets such as sales and stock exports.
- Saved business rules and corrections: when the team corrects an answer or records how something is done, that rule is kept and applied next time. The underlying model is not retrained and no accuracy is guaranteed.
- Reviewable answers: every answer shows the document, row or rule it relied on, so anyone can check it against the source.
- Business data analysis: questions across spreadsheets and exports, such as which accounts are late against their usual interval or how one product family is moving across codes.
- Team responsibilities: who owns which knowledge, who reviews outputs and who can save a rule, agreed before build.
02 / How your information gets in
Three routes in. One proven, two scoped.
The agent is only as good as what it can read. Here is what has been done, what would be scoped for your systems, and what you would need to provide.
Upload the exports and documents you already have
Your team exports from the systems it uses and uploads the files. The agent reads them, shows what each file can answer, and asks a person to confirm before the figures are trusted.
- Excel workbooks, multi-tab files and SAP exports, CSV
- PDF and Word: product sheets, price lists, guidance
- Emails exported from your mail system
You provide: the exports and documents, and a person who can confirm what they contain.
Scheduled refresh of the same exports
If your team already runs a weekly or monthly export, the same files can be picked up on a schedule from an agreed shared folder or mailbox instead of being uploaded by hand. The confirmation step stays.
You provide: a regular export your team is happy to keep running, and a folder or mailbox it lands in.
Not yet delivered for a client; scoped per business. It does not read your systems directly.
A custom connection to a source system
Where a system offers an interface and your IT team can grant access, a bespoke connector can pull agreed data on an agreed schedule, so the agent works from the system rather than from files.
You provide: the source system’s interface details, access approved by whoever owns it, and the fields you want read.
Subject to the source system, its access rules and a separate scope. No native connectors exist today.
03 / Reporting use case
Three workflows. Start with one.
Each is a defined piece of work with agreed rules, agreed exclusions and a check against your source records before handover. The tables are sample data.
01
Customer-order review
Compare each account’s days since last order with its own usual ordering interval, not a blanket cut-off. Returns, cancellations and one-off machinery lines are treated separately.
02
Stock-movement review
Read dispatches alongside receipts and on-hand quantities so a fresh delivery is not mistaken for slow stock. Agreed periods, agreed exclusions, totals reconciled to the source report.
03
Product-code grouping
Equivalent products sold under several codes are mapped to agreed families, with matching units and specifications signed off by your team. Review the family as a whole while keeping every source line visible.
04 / Acceptance checks
Checked before anyone relies on it.
These are the checks we agree with your team and run against sample exports before handover. They come from the kinds of problems that surface in real reporting work.
- 01Scope confirmed, rows reconciledConfirm the export’s filters, period and scope first, then reconcile the loaded row count with the file. Truncated files are caught here.
- 02Correct documentThe report type, period and site are verified against the question before anything is calculated, because a wrong export can look plausible and even reconcile internally.
- 03Totals reconcileQuantity and value totals in the output match the source report for the same period, and any difference is explained before handover.
- 04Exclusions behaveRecent arrivals are not flagged as slow stock. One-off machinery sales do not make a consumables customer look lapsed. Seasonal pauses are treated as agreed.
- 05Spot-checks against sourceYour team picks accounts and products and checks the output against the original records. Only then is it handed over.
A few practical questions
Before we start.
Do we need to change our ERP or give you access to it?
No. The starting point is the exports your team already runs. Any direct integration would be a separate, scoped piece of work.
How much data do you need for the first conversation?
None. We use sample data to talk through the question you want answered. Real exports come later, once we have agreed what to look at.
What does the output look like?
A short review list or table your team can read in a few minutes, with the rules written down beside it.
How is it priced?
A fixed price per agent, agreed in writing, half to start. Scope and acceptance checks are agreed in writing first. See pricing.
Is this the same as the website agent?
No. This agent works on internal knowledge and exports for your team. The website agent answers product questions for visitors on your website. Either can be taken up on its own.
A 15-minute conversation
Worth a conversation?
Tell us which reporting question comes up most often, and we will talk through how it could be answered from the exports you already have. No preparation needed. We can use sample data.