A retail group asks Claude how its stores did yesterday

Sample data: Claude answering how each of four stores did yesterday against target

This week we connected a retail group's till figures to Claude, so its managers can ask how each store did and get the answer in their finance team's own layout. We also built an AI assistant that prices a builder's bill of quantities, tested Claude skills for a freight company against a sample month, and moved Jon's day onto one page that Claude writes each morning.

A retail group's managers ask Claude how each store did, in the finance team's own layout

After the shops close, a retail group's finance team sends out the day's sales against target for every store, and keeps weekly, monthly and stock sheets in the same shape. We have connected those figures to Claude. A manager opens Claude and asks how each store did yesterday, or which supplier's stock is running low, and the answer comes back laid out the way the finance team lays it out.

A small program on the group's own server reads the till system's database through a login that can only read, and sends the new figures every five minutes to a private store. Claude reaches them through a connector with a tool for each of the finance team's sheets, and a Claude skill for each kind of report tells it how to set the figures out. Each manager sees only the figures for their own part of the business.

Nobody watches a server, so this one is watched for them. If no figures arrive for 20 minutes, our team gets an email straight away, and another when the figures flow again. A report at 06:30 each morning says how the night went.

The connector, the skills and the alerts are built and tested end to end against a test database. The program goes onto the group's server next.

Sample data: a manager asks how each store did yesterday against target, and Claude answers with a table of day, week-to-date and target figures for four stores
Sample data. A manager's question and Claude's answer, in the layout of the finance team's daily summary.

Built with Claude, a custom connector, Claude skills, Vercel and Resend

An AI assistant that prices a builder's bill of quantities

A bill of quantities lists every material a building job needs, often in hundreds of lines of a quantity surveyor's shorthand. For a supplier, finding the lines it can fill means reading every one of them.

For a building products supplier, we built an assistant that reads the whole bill, picks out the lines the supplier can fill, matches each one to a product by what it does, and reads the quantity each line needs. A plain script then counts the packs and prices them from the supplier's own price list, so every total traces back to a listed price. On a 523-line sample bill it found every line the supplier can fill and left the look-alikes out.

We also taught the supplier's demo chatbot what to do when a customer asks for another brand's product by name: it answers with the supplier's own equivalent. The next step is running the quoting assistant on one of the supplier's real bills.

Sample data: lines from a bill of quantities, each matched to a product with the number of packs and a price, one line left out, and a total
Sample data. Each line of the bill matched to a product, with the packs worked out and priced from the price list.

Built with Claude and Python

Claude skills that cost, quote and track a freight company's jobs

A skill is a set of written instructions Claude follows for one job, with the files it needs. For a freight and logistics company, we wrote skills for job costing, quoting, shipment tracking and reconciliation. Each one reads the company's job records through a connector and follows the steps the company's own staff follow.

We tested them before anyone at the company relies on them. For a sample month, a script worked out the right answers from the records first: each job's revenue, cost and gross profit, and the charges that were never passed on to a customer. Claude's report was then marked against those answers, and the job-costing skill passed all 19 checks on both Claude Sonnet 5.5 and Claude Opus 5.5. The quoting skill works out the cost and selling price per container from the rate files and writes the quote email with the costs left out.

The skills are ready for the company's team to try in their own Claude.

Sample data: a month of closed freight jobs with revenue, cost and gross profit, each marked as checked, and one job flagged for a storage charge not billed
Sample data. A month of closed jobs, each checked against figures a script worked out from the same records.

Built with Claude skills, a custom connector, Claude Sonnet 5.5 and Claude Opus 5.5

One page Claude writes for Jon each morning

Jon's day now starts on a single page that Claude writes before he does: the day's meetings with their preparation, the tasks due, and replies waiting to go, already drafted.

Buttons on the page save a Gmail draft, post to Slack or open WhatsApp. Anything the page cannot do by itself, such as sending from the studio's email address or updating a task, is saved for Claude to carry out. Each morning's page starts as a copy of the day before, so earlier days are one click away in the list on the left.

We are setting up the same page for the rest of our team, each one built in that person's own Claude so its buttons act as them. The same works for any business owner: one page each morning, drawn from their own inbox, calendar and task list.

Sample data: a daily page with a list of days on the left, the day's meetings and tasks, and a drafted reply with buttons to save a Gmail draft, post to Slack or open WhatsApp
Sample data. The day's meetings, tasks and a drafted reply, with buttons that act on them.

Built with Claude, Claude Code, Gmail, Slack and Google Calendar

Also this week

  • Email from our AI waits until morningWhen Claude Code is about to email someone outside the studio between 19:00 and 05:00, it now asks first, and unless told to send, schedules the email for 06:00.
  • A sales dashboard that matches the finance team's sheetsFor the same retail group, we rebuilt its sales dashboard so each page is one of the finance team's own sheets, with their labels and column order. The new version passed 143 automated checks.

Next workshop: AI in Business on Wednesday 11 November in Harare

The garden courtyard at Sabre Business World, Harare

Leave the morning with AI working on your own business

When
Wednesday 11 November 2026, 08:00 to 12:00
Where
Sabre Business World, Harare
Price
USD 200 a seat, cash on the morning

A working session at Sabre with your laptop open. You set up Claude on your own documents, your inbox and a daily brief, and take home an AI policy for your company. Seats are booked in the chat on the workshop page.

Book a seat

Studio news

  • Book a workshop seat in the chatPress Book a seat on the 11 November workshop page and register in the chat with your name, phone, email and how many people you are bringing. A copy comes to you by email.
  • Short films from the 21 October morningThe workshops page now has clips filmed in the room at Sabre, so you can see how the morning is taught before you book.

New podcast episodes

Our podcasts are written by AI and read by AI voices. They are free in any podcast app.

Written with AI from Executive AI Studio's work records. Clients are not named, and the pictures of our work show sample data.