AI & Automation · Live demonstration

An AI that actually understands your business data.

Ask your business a question in plain English: "how is revenue tracking?", "which customers are slipping?", "what is about to run out?" A data agent reads your own systems and answers with the real figures in seconds. No dashboards to build, no reports to wait on, no exports.

Live recordsanswers from your actual data, not a guess
Read-onlyanalyses everything, changes nothing
Traceableevery figure comes from a query you can see
Plain Englishno dashboards or query language to learn

The idea, in one line

A general chatbot guesses about your business. A data agent reads your actual records.

Ask a general AI how your sales are tracking and it cannot know: it will either decline or invent something plausible, because it has never seen your numbers. A data agent works the other way around. It connects to your live system, runs a real query, answers only from what it found, and can show you exactly what it read.

General chatbotguesses from training data
Your question Its training data A confident answer × Knows nothing about your numbers
Data agentreads your live system
Your question Your live ERP Runs a real query A grounded answer ✓ Your actual figures, traceable
01

The demo · an example use case

Ask your business a question, get a straight answer.

We built this demonstration on a realistic ERP for a trade and industrial distributor: two years of orders, 300 customers, 400 products and stock across three warehouses, so we can show live numbers safely. The industry is not the point. The same thing works on whatever your business runs on, and everything below is the actual demo, not slides.

$

Sales and margin

Revenue and gross margin by month, quarter, category or product, and what is driving the trend.

$30.5M last 6 months · +28%

Customers

Top accounts, who is growing, who has gone quiet, and who is tracking below their usual run-rate before a call.

153 growing · 110 below run-rate

Products

Best and worst sellers, strong-revenue-but-weak-margin lines, and what gets bought together.

a rank-3 seller on just 10% margin

Inventory

What is out of stock or at risk, and what is on order with expected arrival, across every warehouse.

35 at risk with nothing on order

Reps and targets

Target attainment by month against each rep's own book of customers, ahead of and behind plan.

7 of 10 reps above target

Returns and warranty

Return and warranty rates by product and category, and which items are warranty-prone.

worst category 1.4% · most under 0.5%

Read-only, by design. In this demonstration the agent can read and analyse everything, and change nothing. That is the safest place to start, and where it goes from there is a decision made deliberately, with the right controls.

02

See it working

From a plain question to a grounded answer.

The agent is connected to a live ERP and answering in real time. These screens are the actual demo.

1
Getting oriented

Ask what it can do

Someone asks, in plain English, what the agent can see and help with. It explains that it is connected to the ERP, read-only, lays out the ground it covers, then offers a live snapshot.

No training required · plain English
The agent explaining it is connected to the ERP database read-only and can analyse sales, customers, products, inventory, reps and returns
2
A real question

Ask a real business question

"How is revenue looking?" comes back with the actual figures, up 28% with June the biggest month by a wide margin, and the month-by-month table of revenue and margin behind it.

Real numbers, straight from the records
The agent answering a revenue question with a six-month table of revenue, gross margin and margin percent
3
Proactive analysis

It surfaces what you would have missed

Ask who is slipping and it surfaces the accounts that have gone quiet and the ones running below their usual rate, the kind of thing a dashboard will not volunteer. The same works for stock at risk, weak-margin best-sellers, or reps behind plan.

110 below run-rate · 28 accounts gone quiet
A table the agent produced of accounts that have gone quiet, with each customer's rep, six-month activity, last order date and days quiet
4
Grounded

You can watch it work

It explains which records it needs, reads them, and answers only from what it finds. Every figure is traceable back to your records, not generated.

Reads the data · never invents it
The agent's reasoning, explaining which database views it needs to examine before pulling the relevant data
03

What to think about

An agent that can read your business has to be governed.

Connecting an AI to a live system is powerful, which is exactly why the controls matter. These are designed in before anything goes near production, and it is the reason the demo is deliberately read-only.

What the agent does

  • Reads your live records, read-only in this demonstration
  • Answers in plain English with the actual figures
  • Shows the queries behind every answer, so it is traceable
  • Uses agreed metric definitions, so numbers stay consistent
  • Sees only the data and tables you grant it

What it will not do without deliberate controls

  • Write, change or delete anything until the process is agreed
  • Reach data beyond what it needs
  • Invent a figure: answers come from a real query, or not at all
  • Act without logging: every question and query is auditable
  • Go to production without testing on real questions first

Capability first, then permission

Reading is the safe first step. Whether the agent should recommend, act with approval, or automate is a ladder you climb deliberately. Each rung adds value, and each rung adds controls.

1Ask

Read-only. Answers questions from your live data. Changes nothing. This is the demo.

2Recommend

Drafts the reorder, the call list, the follow-up. A person still decides.

3Act with approval

Takes the action once a person approves it, inside agreed limits.

4Automate

Handles routine, well-bounded actions on its own, fully logged.

Write-back is a process decision, not a technical one. The same connection can do more than read. We start read-only, prove the value, then decide together what it is allowed to do next.

04

Your version

The same pattern, on the systems you already run.

The architecture underneath does not care what the business is. Point it at your own ERP, finance package, CRM, inventory or job system and your team gets answers from your live data, with the same grounding, traceability and guardrails.

Sales and margin cockpit

Ask where revenue and margin are really coming from, by month, product, rep or region, without waiting on a report.

Account intelligence

Pre-call snapshots: who is growing, who has gone quiet, who is below run-rate. Your team walks in prepared.

Inventory and supply

What is short, what is at risk, what is inbound and when, across every warehouse, answerable on the spot.

Performance against target

Attainment by rep, branch or region, live, without anyone building the spreadsheet first.

Quality and returns

Which products and categories drive returns and warranty claims, and what they are costing you.

Executive "what changed"

A plain-English brief on what moved this month, pulled straight from the data instead of assembled by hand.

The systems it can connect to

Whatever your team runs on, a data agent can connect through its database or its interfaces. A few examples:

Logos illustrate the kinds of systems a data agent can connect to. All trademarks belong to their respective owners.

How we approach building one

We build the secure connector and the accurate process around it, so it works with the systems you already run. A useful data agent is a governance and curation job as much as a technical one.

1Scope

The questions worth answering, and the systems that hold the answers.

2Connect securely

A read-only, least-privilege connection to your data. Nothing more.

3Guardrail

Agree metric definitions, access scope, and what, if anything, it may act on.

4Test

Score it on real questions, including the ones it should refuse or flag.

5Deploy

Into the tools your team already uses, with logging and monitoring.

Then it loops. Real questions feed back into the configuration, new data sources get added, and once the controls are ready, safe write-back actions. That ongoing loop is what AI Manage is for.

What a data agent gives you

A grounded, careful assistant that answers your team's questions from your own live systems: traceable, governed, and living in the tools you already use.

The answers Grounded From your live data, traceable to a query you can see.
The place In your tools Chat, Teams, the web, or your own applications.
The controls Governed Read-only to start, least-privilege, logged and tested.

A read-only data agent is bounded, provable and useful on day one, which makes it exactly the kind of first build an AI Blueprint scopes and de-risks before you commit to a bigger programme.

A small, provable first build

Imagine asking your ERP a question.

Tell us what your business runs on and we will show you the demo live, then talk honestly about what a data agent would do with your data.