Reads
Invoice lines
It reads every outgoing invoice line, credit notes included. It needs no other data.
An autonomous sales agent for wholesalers. It works the accounts nobody has time for, and charges only for measured results.
349 customers with no sales rep, ranked by margin
Synthetic data
The customer base
Above are a wholesaler's 349 customers, one dot each. Out of 520, these are the ones with no sales rep behind them. Small one by one, two thirds of the customer base together.
349 customers
out of 520. 171 have a sales rep, these have nobody.
The gap
A rep's day is finite, and the big accounts fill it. Everyone else orders when they happen to remember. When they do not remember, they order from someone else. The hollowed-out dots have already fallen out of their own rhythm.
HUF 33.1M
of latent margin a year in this group. Latent, not incremental.
The turn
Every account nobody owns runs into one thread: someone knows its rhythm and writes when an order is due. A human approves the messages. Half of the drifted ones are left alone on purpose. They are the control group: no line leaves them, and what came back is measured against them.
93 customers
that drifted get messages from Again. Another 92 stay untouched, and we measure against them.
Reads
It reads every outgoing invoice line, credit notes included. It needs no other data.
Learns
It learns each customer’s ordering rhythm and notices when one falls out of it.
Suggests
It suggests a specific item, checked against stock, price floor and credit limit.
Books and measures
It books the order into the ERP and measures its effect against an untouched control group.
It does not look at averages. Each customer is measured against their own past pace.
Alföld Irodatechnika ordered every 31 days. It has now been 84.
If it fails a single check, it does not go out.
The order lands where all your other orders live. Not in a separate system nobody looks at.
It breaks a free-text reply down into item, quantity and condition. When it is unsure, it calls a human.
“Make it eight, but only if the wall mount comes with it.”
The first analysis runs on your own machine, with no network connection.
Purchase prices and margins never reach a language model. The first phase needs no names, emails or phone numbers either.
Each steps in at a different point of a customer's order rhythm and writes one type of message. So each can be measured on its own, and each has to earn its own autonomy.
due
Writes to accounts that have slipped out of their own order rhythm.
Your usual order, ready to go: toner
2 items, all in stock
Autonomy: Can be earned
dormant
Rebuilds the usual basket for a dormant account.
We have put together your earlier items
At your current partner prices
Autonomy: Can be earned
lost
Writes to a lost account, with a discount on the first order back.
An offer on your earlier items, with a discount
4% off the first order
Autonomy: Never autonomous
on rhythm
Suggests a new category once similar accounts already buy it.
A dock to go with your usual order
Starter quantity: 4 pcs
Autonomy: Can be earned
reply
Reads the customer's reply and drafts the order for the ERP.
Re: Your usual order, ready to go: SSD
SSD-NVME-1T, 20 pcs, delivery Thursday
Autonomy: Can be earned
Every message runs into the approval queue
During the pilot, not one offer goes out without approval. A message type can only leave human hands once it has performed well for a sustained period. Discounted offers never do.
Kecskemét
Ordered toner every 31 days. It has now been 84, so their stock is likely running out.
Total, net
HUF 259,800
What the control group produces by itself is the baseline. Again charges only for the difference between the two groups.
Again works with them
Gross margin over the 12-week pilot
Nobody contacts them
Gross margin over the 12-week pilot
The difference Again madeHUF 4.8M
Illustrative figures, not a measured result
You pay nothing until there is a measured gain.
The fee is a share of the measured extra margin, plus a modest platform fee for the integration. We do not sign without a control group, because then there is nothing to invoice.
Group AAgain works with them
Group BNobody contacts them
01
A CSV or Excel file of your invoice lines. The analysis runs on your own machine, with no network connection.
02
Within a week you see on one page which customers have drifted away, and what they should be ordering now.
03
We split the customer base in two at random. Again works with one half, and nobody contacts the other.
There are no customer references yet. When there are, they will be here.
Then there is no way to measure what Again delivered, and no way to invoice a results-based fee. We do not sign on those terms.
The first phase runs on your own machine, with no network connection. Purchase prices and margins never reach a language model.
No. A CSV or Excel export is enough for the first analysis. Writing into the ERP comes later and needs its own agreement.
No. Again works the accounts nobody has time for today. Your key accounts stay exactly where they are.
The list of drifted customers is ready within a week. A measured result needs one full ordering cycle, which for most product groups is between 8 and 12 weeks.
During the pilot a human approves every message before it goes out. The system learns from what gets rewritten, and the rewrite rate decides when a message type can earn autonomy.
The first step is a one-page list built on your own data. No contract needed for that.
Get the analysis