The AI workforce that runs your store inside Focus ERP.
Stackbirds records how your team works in Focus, then does the data entry itself — supplier shipments, product masters, goods receipts and daily reports — with a human approving the moments that matter. One playbook, everything in one place.
See what's in the playbook.
A two-minute look at how a Stackbirds agent watches a Focus workflow once and then runs it for you.
▶ Tap the video for sound — it plays muted on load; unmute for the voiceover.
Watch every workflow in action.
One short demo per workflow — tap any tile to play. The Focus ERP pipeline below (inventory-in, GRN, invoice OCR, AR collections, daily brief) is shown here as live agent demos. Same gallery as /ai-workflows/ and Stackbirds.
A store that runs on re-typing.
Le-Pearl Fashions runs Focus ERP across three-plus outlets. Every day, three data-entry staff and an accountant re-key the same things by hand — 17 procedures, ~12 hours a day, roughly $5,500 a month on work a recorder-agent can do.
People on entry
Three data-entry staff (~$4,000/mo) plus an accountant (~$1,500/mo), doing keystrokes, not judgement.
Focus procedures
Mapped end to end as SOPs and rebuilt as a live web simulator — prospects watch the workflow without a Focus licence.
Per year, per store
The status-quo cost of the manual way. The alternative isn't cheaper staff — there are none. It's an agent.
Customer zero is the founder's own store. The open question is whether other Focus retailers will pay — see Status.
One pipeline first: supplier shipment → stock on the system.
Not all seventeen procedures. The two fattest tasks are Product Master creation (120 min) and Goods Receipt posting (90 min) — about 3.5 hours a day in two tasks. v1 is exactly this, and you approve the pricing.
| Focus task | Manual | With Stackbirds | Saved | In v1? |
|---|---|---|---|---|
| Product (Item) Master creation | 120 min | ~5 min | ~96% | Yes |
| Goods Receipt Note posting | 90 min | ~4 min | ~96% | Yes |
| Receipt of stock | 30 min | ~3 min | ~92% | v1.1 |
| POS sales report | 25 min | 1 min | ~96% | Later |
| GP% report | 20 min | 1 min | ~95% | Later |
| Store → head-office sync | 30 min | 1 min | ~97% | Later |
Why narrow: win the fattest, most repeated task at one store, prove it, then widen. Six modules at once is how a one-person company dies.
Observe → extract → validate → flag → write.
The same loop on every record, bound for v1 to two screens — Item Master and GRN — with one hard human gate on pricing and a read-back audit so you can trust it with the live store.
Read the shipment
Ingests the supplier sheet or invoice and authenticates to the real Focus session.
LP-01 / LP-02Identify & extract
Parses each line — description, barcode, category, size/colour/season, cost. Anything below 90% confidence goes to a human review queue.
LP-04 / LP-05Validate & enrich
Business rules check barcode format and quantities; category and supplier codes resolve against Focus master data. Anomalies are flagged in plain English.
LP-06 / LP-07Human approves pricing
The one hard gate. You confirm cost and retail price before anything is written. Nothing posts without you.
Human gateWrite to Focus
Creates the Product Masters, then posts the GRN lines — each write guarded by an idempotency key so a retry never duplicates a record.
LP-08Read back & report
Re-reads the posted records, asserts written equals intended, freezes on any mismatch, and reports: products created, GRN total, time taken.
LP-09 / LP-11This is the actual Focus ERP screen the agent drives.
Two minutes of the real Focus ERP screen at Le-Pearl Fashions — the same Item Master and GRN screens described in the loop above, played silent so you can read along while the agent works. No mock-up, no Loom polish — the real session, recorded.
▶ Silent by design — Focus ERP screens carry the story; tap for sound if you want the voice-over. See also a live fashion-retail AI journey on Perplexity Apps: fashionlretail-ai-journey.pplx.app →
What we build this week — and what we don't.
In scope
• Inventory-in only: shipment → Product Master + GRN.
• One store (Le-Pearl), the real Focus instance.
• One human gate: pricing approval before posting.
• Read-back audit on every write.
Out of scope (for now)
• Finance/AP, Procurement, HR, Sales/CRM, Payroll agents.
• The KPI dashboard, OKRs and sprint board.
• Multi-store, multi-currency, self-learning loop.
The three risks that actually matter
R1 · Session / auth
Focus times out mid-post and leaves a half-written GRN. Refresh the token before each write batch; never start a GRN you can't finish.
R2 · Idempotency (P0)
Focus can return success on a failed write; a retry then duplicates live inventory. Check existence by natural key (barcode, invoice no.) and write exactly once.
R3 · Write-back audit
After posting, re-read and assert written equals intended. This single step is what lets you trust the agent with a real store. Never skip it.
Record → Train → Deploy → Monitor. Then widen.
The full delivery method: discover the manual workflow, record it in Focus, train the agent, validate to a 95% sign-off, deploy with approval gates, then monitor and retrain. v1 runs this for one pipeline; the rest is roadmap.
The four phases
Record — capture the real Focus session. Train — model infers intent, maps fields, asks clarifying questions on edge cases. Deploy — schedule, triggers, approval gates. Monitor — accuracy, cycle time and SLA, with a retrain loop.
The expansion path (later)
Once inventory-in is proven and paid for, the same loop extends to the other Focus modules — Finance, Procurement, HR, Sales, Payroll — each a new recording, not a rebuild. The moat: a shop owner can run this when they can't run a general agent.
Make one cell true.
The operating model imagines seven agents with accuracy scores and SLAs. None of those numbers are measured yet — one founder, one store, one pipeline. The whole game is to replace a single invented number with a real one.
| Milestone | State |
|---|---|
| 17 Focus SOPs mapped + web simulator built | Done |
| Per-task time savings measured | Done |
| Inventory-in pipeline writes to real Focus | Open — the #1 unknown |
| One real shipment run end-to-end, filmed before/after | This week |
| A second Focus retailer says "I'd pay $X/mo" | Not yet tested |
When one cell is a real measurement, Stackbirds stops being a model and becomes a company.
The assignment this week
Run the Product-Master + GRN automation against the real Focus at Le-Pearl on one actual supplier shipment, with you approving pricing at the end. Film the manual way and the automated way side by side. That clip is both your product proof and your sales asset.
Try Stackbirds free