Are we making money?
Sales, labour % and food cost, on the same business date.
Case study · Magnolia Synergy
One view of profit and operations for a counter-service store.
A counter-service franchise location in New Jersey ran its day across four or five separate systems. Magnolia Synergy brought them onto one business date. The store closed its first actual food cost period at 27.8%, from two complete counts of 114 of 114 items.
IRIS (Integrated Reasoning & Intelligence Steward) is our planning assistant. The scoping page saves your answers to your own computer as a file, then offers a link to email them to us. The location is not named, and its figures are shown as percentages and counts.
The problem
Four questions, with the answers in four or five places.
An owner needs four answers every day. At this store each one lived somewhere different: the point-of-sale system for sales and tips, a scheduling app for the rota and time clock, a payroll service keyed by hand, and workbooks and PDFs for inventory and invoices.
Where is food cost going?
Counts, purchases and waste, turned into a real cost per period.
Is the team set up for the shift?
Station checklists, sign-off, and who can do what.
Are standards being kept?
Temperatures checked against food-code limits, every shift, on record.
What Magnolia Synergy brought together
Every source on one business date.
Sales arrive every 30 minutes, timecards every hour, and counts and waste every 15 minutes. The food distributor's invoices are read automatically each week. Each lands on the store's business date and is checked before it is shown. When a source has not arrived, the app says so instead of showing a zero.
Sales and labour, through the day
Net sales, orders and average ticket beside labour hours and labour %, with a day in progress marked as in progress, never mistaken for a closed one.
Food cost you can trust
Every item is counted on the tablet and a manager approves the count. A period becomes actual only when the counts at both ends are complete; until then it says provisional.
Vendor invoices, read for you
Each invoice line is matched to an item, so purchases count toward the right cost. Food is kept apart from packaging and supplies, so each has its own cost.
Every shift runs from the tablet
Opening, mid-shift and closing checklists for each station. A lead or manager signs them off, and nobody signs off their own. Temperatures are checked against food-code limits as they are entered.
Waste, logged when it happens
Item, quantity and a reason, costed from the item master, so waste shows up in each food cost period and points to the fix: portioning, prep size or training.
The right access for each person
From a read-only view of today's figures to configuring the store, each person can do what their role allows. The server enforces it, and sign-in is by invitation.
The outcomes
The first actual food cost period.
Actual, not an estimate: the first period backed by two complete inventory counts.
- Food cost, first actual period
- 27.8%
- Food cost as a share of net sales, 20 September to 1 October 2026.
- Items counted, at both ends
- 114 of 114
- Both counts were complete, which is what makes the period actual.
- From count to food cost
- Same day
- The closing count was approved in the afternoon, and the period's food cost was ready that night.
- Invoice totals matched
- 22 of 22
- The automatic reader matched the main food distributor's invoice totals on its July and August invoices.
- Morning checklist, signed off
- 41 items
- Completed by a team member on the tablet and signed off in the app by an admin.
Measured at the store from its own data, served 2 October 2026. Food is costed apart from packaging and supplies.
Week to week
A simple weekly rhythm.
The store runs to a rhythm, and each step tells the owner something.
| When | Do this in the app | What it tells you |
|---|---|---|
| Every shift | Run station checklists; a lead signs off | Standards kept, temperatures in range |
| As it happens | Log waste with a reason | Waste cost, and where it comes from |
| Mid-day and close | Check sales and labour % | Whether staffing matches the day |
| Same day each week | Count every item; a manager approves it that day | A new actual food cost period the next morning |
| Every delivery | Make sure each vendor invoice is in | Complete purchases, so packaging and supply cost turn actual |
| Each period | Review food cost %, waste and purchases by vendor | Where to act: portions, ordering, pricing |
The biggest lever is the weekly count on a fixed day, approved the same day. Each complete count closes another actual food cost period.
How we deliver
Built with Magnolia BluePrint: governed AI delivery, on the record.
AI agents do much of the building. Magnolia BluePrint is the discipline around them, and it is the same for a store's software as for everything else we ship.
Decisions on record.
Significant decisions are written down as numbered decision records, each naming what enforces it.
Gates on every change.
Every proposed change runs the same suite of automated checks, and runs it again just before it joins the main line.
Independent review.
A change is approved by an account separate from the one that wrote it, so nobody approves their own work. Code changes also get an independent AI code review, in up to three rounds, unless the operator waives it on the change; changes to records alone are exempt.
One way in.
A merge queue is the only path to the main line. Nothing is pushed straight to it.
Delivery, measured
What that discipline looked like over eight weeks.
Measured across every change merged into our main engineering repository, where this store's software is built.
- Pull requests merged
- 602
- From the repository's first merge on 11 August 2026 to 4 October 2026.
- Approved before merge
- 598 of 602
- Each by an account other than its author, 524 of them by the dedicated reviewer identity.
- Automated checks per gate run
- 48
- As of 4 October 2026, across 3,365 gate runs in the period on proposed changes, merge candidates and the main line.
- Merge candidates stopped
- 49 of 401
- Candidates whose checks failed in the merge queue did not reach the main line.
- Checks passed after merge
- 489 of 490
- Completed check runs on the main line itself, after changes landed.
- Decision records
- 84
- Written down and numbered, alongside about 3,630 automated test cases.
Measured 4 October 2026, from the repository's pull requests, reviews and check runs, 11 August to 4 October 2026 (UTC).
Where this fits
How a project like this starts.
IT and operations modernization
Assess what you have, secure it, connect your data and reporting, and automate the routine work, one phase at a time.
Fractional AI and automation architect
A senior architect for a set number of hours a week, who keeps the plan and builds or directs the implementation.
How engagements work
Start with a defined first step, and continue as a managed service or a Magnolia Synergy subscription only when that is useful.
Tell us how your business runs today.
Describe your systems, your reports and where the time goes on the IRIS scoping page. You keep a copy, and an engineer reads it before we talk.