Retail planning on an appliance in your building

You're losing 4.1% of sales to inventory you don't have.

And another 3.2% to inventory you can't sell. Most systems can't tell the two apart, so they fix neither.

Source: IHL Group · Corsten & Gruen. On R500m turnover that is roughly R20.5m a year. The maths, on your turnover →
Offline · a box on a desk in Johannesburg Live demo

The demo box is on a desk in Johannesburg, not in a data centre.

That isn't a limitation. It's the product. RetailPulse runs on an appliance in your building, with nothing leaving it. So the demo runs on ours the same way, and right now it's offline.

Everything below is the real thing, captured. And we’ll show you the live one, on your own data, within one business day of you asking.

See it on your own data A personal reply within 24 hours. From Rob, not a sequence.
The Recommended Orders queue. One row expanded: a 16-unit order for store 213, Bramble Ridge, with the reasoning (9.2 days cover against a 10-day lead time, 16 short over 17 days, earliest arrival 09 Sep) and Approve 16 / Reject.
Look at the open row. A 16-unit order for one store, and underneath it the reason: 9.2 days of cover against a 10-day lead time, 16 short over 17 days, earliest arrival 09 Sep. For you: every recommendation comes with the sentence you'll need on Monday. It does the first pass. You do the second, with one click. demo estate · names, prices and geography invented; sales, inventory and gaps real · money masked
Bad data is Tuesday

Your system orders in packs of 12. Your supplier changed to 6 in March.

Nobody told the system, so every order since has arrived half short. Bad master data isn't the edge case in retail. It's Tuesday.

A line that was out of stock for six weeks looks exactly like a slow seller. Most systems read the flat sales line and cut the order, so it stocks out again, so they cut again, quietly strangling the line they should be growing. Nobody notices, because the report says demand fell.

RetailPulse is built to say "I don't have enough here to give you a credible number" rather than to guess confidently. Where the data can't carry an answer, it tells you on the screen, and tells you what to fix.

Four stores at the top of the ranking, each carrying the chips stocked out and thin history, and beside each the sentence: top of the list despite being unavailable 54% of the time; the true number is higher. Keep it in stock.
Look at the chips beside each rank. Store 899 is top of the list, and it was unavailable 54% of the window. The engine says so before you act on the rank. For you: the number you're about to defend in a trading meeting comes with its own health warning. 217 items on this estate carry no cost price; their profit reads "no cost price on file", never zero. Analytics · demo estate · quantities and time; money masked
Eight questions your system can't answer

Tick the ones you couldn't answer by Friday.

Nobody is scoring this. Tick a box if the honest answer is "I don't know" or "about four minutes".

Nothing about you is sent. Only which boxes.
Guard-railed autonomy

It starts by asking permission for everything.

Three settings: Review, Guarded, Autopilot. It ships on Guarded: it tells you what it would have done on its own, and waits for your click.

It does the first pass. You do the second. It doesn't decide; it gets you to the decision faster, with the reasons attached, so a lean team can cover a week's worth of data before lunch. You watch it be right for a few weeks. Then you move the dial, one notch, when you're ready. There's a kill switch, and it takes effect instantly. No restart, no run, no waiting.

Every competitor sells you autonomy. We think you should make us earn it.

Four tiles: 1,412 open recommendations; 447 would push itself; trust dial Guarded; nightly runner 02:00 local. Under them the daily digest: 447 of the 447 Green were within every rail and would have pushed themselves on Autopilot; preset Guarded, so nothing was pushed; 47 outside norms need sign-off.
Look at the second tile and the digest. 447 orders were inside every rail and would have placed themselves on Autopilot. The dial reads Guarded, so nothing moved. 47 sat outside norms and waited for a signature. For you: it tells you what it would have done, and waits. You watch it be right for a few weeks before you move the dial one notch. Orders · demo estate · trust dial, kill switch, every change audit-logged
Adoption

Every recommendation arrives as a sentence.

Order 200DC01 holds 500, but 300 are allocated to a higher-priority line this run — 200 left. Packs of 4, rounded up. Recovers in 23 days.

Every recommendation comes with the sentence you'll need on Monday. A planner can act on that in ten seconds, and defend it the moment someone asks why. A number with no reasoning behind it is a number nobody follows. An unfollowed system is a cancelled system.

Sovereignty

Nothing leaves your building.

The intelligence runs on the appliance in your building: your store, your server room. Local inference. Zero open inbound ports. Not our cloud, not anyone's.

And the tunnel you'll ask about is how you reach it remotely. It isn't how it runs. Cut it and the system keeps trading on your LAN; it just vanishes from the internet. The nightly run still fires at 02:00 local, tunnel or no tunnel.

How this is built
Inbound ports

Zero

The server is invisible to the internet. There is nothing to scan and nothing to knock on.

Databases

One per customer

No shared data pool. A worst case touches one customer and never cascades.

Outbound connections

One

One outbound HTTPS connection. No firewall changes, no static IP, no port forwarding at your site.

Kill switch

Seconds

Drop the tunnel and the system vanishes from the internet, while carrying on in-store on the LAN.

The anchor

Live in 48 hours, not next winter.

Enterprise planning suites run $60,000–$180,000 a year and six to twelve months to go live.

RetailPulse is a fraction of that, live in 48 hours, on an appliance in your building. And the price doesn't go up because you grew. Add SKUs, add stores, add users. Everyone else charges you more for succeeding.

What stockouts actually cost you
Enterprise suiteRetailPulse
Time to live6–12 months48 hours
Annual cost$60k–$180kA fraction of that
Runs onTheir cloudAn appliance in your building
When you growThe bill growsThe price is the price
ImplementationTime and materialsA fixed number
Origin

Built by a retailer, for retailers.

Built by retailers, for retailers. Thirty years from every side of the table: the vendor shipping in, the distributor in the middle, and the retail head office deciding what goes on the shelf. The first version of this engine ran across hundreds of stores for a major retailer. What runs today is a ground-up rebuild: the same understanding of the problem, on hardware and models that did not exist when it was first written.

And it runs in your building, which means it isn't a service anyone can switch off from somewhere else.

The whole story

See it on your own data.

Twelve weeks, one category, anonymised by you. A ten-minute job, not a project. Send it Thursday and you'll have the findings the following Wednesday.

  1. 1 · The extract"Twelve weeks, one category, anonymised by you." A ten-minute job, not a project. No NDA drama, no IT ticket, nothing to approve.
  2. 2 · The findings"Here is what it found in your data." The engine has never met a real dataset it did not find something embarrassing in. Evidence about you, not persuasion.
  3. 3 · Run-alongside"Eight weeks. It doesn't touch anything." Zero operational risk; your own scorecard.
  4. 4 · Founding agreement"One of three places." By then you have your own proof, and the scarcity is real.
Send the extract

A personal reply within 24 hours, from Rob, not a sequence.