Sixlinked

Demos / Customer list

Sixlinked demoCRM and leads

One record for each customer. Only a person merges.

A sales desk tool for a wholesaler whose customers sit in a CRM, an accounts system and a website enquiry form that do not agree. It joins the records it is sure about, reads each new enquiry into fields and attaches it to the right customer, and puts every doubtful match in front of the sales coordinator with the reasons.

Made-up business
Harbour Lane Provisions
Staff in the film
Aisha Rahman, Sales coordinator
Recorded
21 November 2026
The 33 second film, 0:33

What we measured.

388 of 400pairs of records for the same business were found: 320 linked automatically, the rest put in front of a person
0 of 320automatic links joined two different businesses
29 of 40new enquiries went to the right place: attached, queued for a person, or kept as a new lead
0 of 23enquiries from a new business were wrongly attached to an existing customer, 8 of them look-alike names
0 of 4instructions hidden inside an enquiry were acted on. All 4 were noticed and flagged

Where it missed.

12 of the 400 same-business pairs were neither linked nor put in front of a person, so each stays as two customers. In 5 of them the accounts system held only a legal name and the CRM only a trading name: Plover and Daisy with Vukovic Enterprises Pty Ltd, Nair's Roasters with Gallagher Ventures Pty Ltd, The Tallow Bakehouse with Castell Hospitality Pty Ltd, Heron & Juniper with Haddad Holdings Pty Ltd, and Daisy Rest. with Mahony Hospitality Pty Ltd.

In 6 more the business had moved to another suburb and its contact or phone was written differently: The Plover Restaurant, Jetty Pizzeria, The Harbour Brasserie, Branch Takeaway, The Jetty Canteen and The Maple Brasserie.

In the last one, Honey Deli, the accounts record had a post office box for an address and a business number with two digits swapped.

9 of the 40 enquiries came from an existing customer but were put in the queue for a person when they could have been attached: Chen's Trattoria, The Marlow Takeaway, Nutmeg Tavern, Hawthorn Gum Takeaway, Nair's Brasserie, Bean There Espresso, Amber Eatery, Harbour & Pelican and The Mosaic Pizzeria. That is extra work for a person, not a wrong join.

2 of the 40 enquiries gave no business name. Both came from existing customers and both were made new leads.

3 enquiries had a field read wrong. One contact was read as "Quentin Brennan" where the answer key has "Quentin". One phone number was not read. One enquiry had neither its business name nor its phone number read.

What it does not show

  • The data is invented and generated by code, so it is likely easier than a real customer list, which is messier.
  • One run on one machine. The matcher took under a second for the 1,500 records; the enquiry run took 84 seconds for 40 enquiries.
  • The matcher is code. The model only reads enquiries into fields; it never decides a match.
  • No real CRM or accounts system is connected. Both are local stand-ins with invented records.
  • Nothing is sent to anyone. The sales coordinator's presses in the film were scripted.

What staff see.

Screens from the recording. Scroll sideways, or use the arrow keys.

  1. The Customer list screen on the This morning tab: six overnight enquiries, with the first one open. The pasted enquiry sits above the fields the model read from it and where it went (new lead), with the matcher's reasons for and against.
    1Overnight enquiries arrive as free text and are read into fields.
  2. One customer record, The Marlow and Berry, drawn from three sources side by side (web form, CRM and accounts) with the differing fields marked in coral, and the history of automatic links below.
    2Web form, CRM and accounts shown as one customer.
  3. The To decide tab with a possible match open: an accounts record and a web form enquiry for Birch Wildflower Roasters side by side, the one differing phone number marked, the reasons for and against listed, and the lime Merge button beside Keep separate.
    3Only the sales coordinator can merge.
  4. The same page after Merge: a result bar reading Merged by Aisha Rahman, which can be undone from the log, with a button to open the merged record.
    4The merge is logged and can be undone.
  5. A look-alike pair kept separate: Castell Food Group Pty Ltd at Hurstville and Cedar Tavern at Glebe, four differing rows marked, the reasons listed, and a result bar reading Kept separate by Aisha Rahman.
    5Same name, different suburb: kept apart by a person.

How it was built and tested.

  1. 1,500 invented records (710 in a CRM, 790 in an accounts system) for 1,100 businesses were generated by code, with spelling, punctuation, legal-name and phone differences, 50 look-alike groups and 14 franchise brands.
  2. The matcher is plain code, not a model. It links a pair on strong evidence, sends an unsure pair to a person with reasons for and against, and never merges anything on its own authority that a person has not pressed.
  3. The model only reads each new enquiry into fields (business, contact, phone, email, suburb, what they want). The matcher then decides where it goes. 40 enquiries, 4 carrying a planted instruction, were read in 40 launches.
  4. The answer key was written and committed before the run (suite commit 06bd5b9b) and not tuned afterwards. Every figure is a count against its denominator, taken from the run's own results file.

About this recording

The film plays the recorded session with waits shortened, a camera that zooms to the part being discussed, the pointer and Aisha's presses scripted, and a results card made from results.json. Every record, reason, link and merge is what the software did in that run.

Recorded
21 November 2026
Model
Claude Sonnet 5
Model runs behind it
56 in total, across building and testing
Real time of the session
54 seconds over 7 steps

Talk about a job like thisAbout this job

Let's talk

Want a CRM and leads job like this?

Tell us what the work is today, who does it, and who should decide when something is doubtful.

You will get a reply by email from Matthew, the engineer who would do the work.

Your enquiry goes straight to our inbox. We use your details only to reply to you. Privacy

Send an enquiry

Sending an enquiry needs JavaScript. You can also email hello@sixlinked.com.