Blog

Transformation17 Nov 20265 min read

One customer, five names

The overdue invoice is filed under one name, the credit limit under another and the new order under a third. A note on how a buyer goes over their limit without any system noticing.

BoLondon

At twenty to five on a Thursday, a trader closes a deal in a chat window. A processor near Parma wants 22 tonnes of frozen pork shoulder for next week, at €2.65 a kilo. The buyer found the offer on a marketplace, asked two questions about the plant and agreed: €58,300. The trader saves the order, books a truck for Monday and goes home pleased.

The buyer is Carni Brenzola. The back office knows them well: two loads in October, both delivered. The invoicing tool knows them as CARNI BRENZOLA SRL, with €92,400 still open and one invoice 38 days late. The credit insurer’s list has them too, with a limit of €120,000. The marketplace knows a user called Brenzola, from Parma, buying for the first time.

That is one company under several names, and none of the systems said no.

The trader is invented, and so is Carni Brenzola. The several names aren’t unusual. Ask a credit controller how many times their biggest customer appears across the company’s systems, and they will usually need a moment to count.

Every system answered its own question

Nobody was careless. The back office was asked whether it could save an order for Carni Brenzola S.r.l., and it could. The invoicing tool holds invoices, and the new order isn’t an invoice yet. The insurer’s list holds limits, and nobody looks at it when a deal is going well. The marketplace was asked whether the buyer was verified, and the buyer was. Each system answered the question it was given, and each answer was correct.

The question that mattered was one no system could answer on its own: if the truck leaves on Monday, how much will Carni Brenzola owe? The answer is €150,700, against a limit of €120,000, with one invoice already more than a month late. Credit insurance covers a buyer up to the limit and no further. On many policies, deliveries to a buyer who is already late past a set number of days aren’t covered either.

Here are the records. Each one gives its answer about Monday’s truck. Then match them.

One buyer, six recordsA new order for 22 tonnes of frozen pork shoulder. Each system is asked about Monday’s truck. Carni Brenzola is invented. The records are the kind every desk has.
  1. Back office

    Carni Brenzola S.r.l.

    VAT IT02741980345

    Two loads in October, both delivered.

    About Monday’s truckOrder saved. Truck booked for Monday.

    The first record: a new company

    org_01a7c3e9…

  2. Invoicing

    CARNI BRENZOLA SRL

    VAT 02741980345

    €92,400 open. One invoice, €47,100, is 38 days late.

    About Monday’s truckNo new order under this name.

    Sure match: the same VAT number, written without the IT

    org_01a7c3e9…

  3. Credit insurer’s list

    CARNI BRENZOLA S.R.L.

    VAT IT 02741980345

    Insured limit €120,000.

    About Monday’s truckNobody asked.

    Sure match: the same VAT number

    org_01a7c3e9…

  4. Sales team’s sheet

    Brenzola Carni

    VAT IT02741980345

    Good customer. Ask for Marco.

    About Monday’s truckFine. Marco is happy.

    Sure match: another name, the same VAT number

    org_01a7c3e9…

  5. Marketplace

    Brenzola

    Parma, Italy · no number

    First deal today: 22 t of frozen shoulder, €58,300.

    About Monday’s truckNew buyer, verified. Deal agreed.

    Maybe: the same name in the same country. Held for a person.

    A person said: the same company

    A person said: a company of its own

  6. Back office

    Salumi Brenzola S.r.l.

    VAT IT02741990347

    Buys sliced bacon. Same phone number.

    About Monday’s truckPays early. Never late.

    Kept apart: a different VAT number. A company of its own.

Can Monday’s truck go?22 t · €58,300
Back office
Yes
Invoicing
Not asked
Credit insurer
Not asked
Sales team
Yes
Marketplace
Yes

Three yeses and nobody saying no. The truck is booked.

Names mislead in both directions

The obvious fix is to match on the name. In this example it gets it wrong twice. “Brenzola” on the marketplace and Carni Brenzola S.r.l. in the back office share one word, and they are the same company. Salumi Brenzola S.r.l. shares two words and a phone number, and it is a different company: the family’s slicing business, with its own VAT number, its own limit and a habit of paying early. Merge it in and its good record hides its sister’s bad one.

So a name can only ever raise the question. What answers it is a number: the VAT number, or a national register number, such as an Italian REA number or a German HRB number together with its court. The invoicing tool writes the VAT number without the IT in front, and the back office writes it with. That is still the same number, so the two records are the same company, and nobody has to be asked. Two records with different numbers of the same kind are never joined, however alike their names. Everything in between is held until a person who knows the buyer decides.

Four answers, and one of them needs a person

That is how One Identity, the registry we built for our own desk, treats every record a system sends it. Each record gets one of four answers:

  • The same record. It has been seen before, and its company is updated.
  • A sure match. The same registered number, or for a person the same e-mail address. It joins the company that already exists.
  • Nothing similar. It becomes a new company.
  • Maybe. The same name in the same country, the same phone number or website, or the same number under another name. It is held until a person says which company it belongs to, or that it is a company of its own.

The answer is kept. The next time the marketplace sends Brenzola, it goes straight to the company the trader chose and isn’t held again. Every company gets one ID that never changes. Each system keeps its own records and its own screens, and nothing is migrated or switched off. The systems simply learn that they were talking about the same buyer.

Where we kept the model out

This looks like a job for AI, and we kept AI out of the decision on purpose. A sure match is a rule about numbers. Anyone can check it in a few seconds, and it gives the same answer every time. Deciding whether two companies are one is cheap to review and expensive to get wrong, so it goes to the person who knows the buyer. A model is more useful earlier: reading the number off a letterhead or an invoice in the first place, so the rules have something to work with.

It is the same split Ivan wrote about in Models read. People decide.

Count yours

This takes an hour. Take your twenty biggest customers. Search for each one in every system you have: the ERP, the invoicing tool, the CRM, the web shop and the sheet the sales team actually uses. Write down every name you find and whether it carries a number. Then, for each customer, add what they owe in one system to what they have on order in the others, and compare the total with their limit.

If your list has a Brenzola on it, it is better to find out on a Thursday afternoon than from the insurer.

Send us two customer lists

Written by BoMakeBetter · London

Let’s make it better.

Tell us where the day goes in your business. In thirty minutes we will tell you what we would change first.

[email protected] Sofia · Zagreb · Berlin · London · Warsaw · Osaka