What we learned from Outform's data: every past job joined into one intelligence database, and what each workbook family can teach. About 2.5 minutes.
From a new request to a recommended price, the kits experiment, and how every scenario is scored. About 4 minutes.
An update of an earlier display: the past kit cut to the new scope, priced and scored against the real cost. About 2 minutes.
A two-bay gondola with one reference job: the cost from the look-alike, the margin from the client’s own wins. About 2 minutes.
A repeat order with two changes on a short run: what the agent can price, and what it leaves open. About 2 minutes.
We started from the whole history: 1,296 jobs. Each job folder was joined to its deal in Sugar and to its actuals in D365, by job number, read-only on every source.
| 531 | jobs estimated in the Bottom-Up sheet: the margin and the outcome on most, the kit breakdown still in the workbooks |
| 209 | in flat sheets |
| 142 | in the 714 estimating tool: the cost broken down kit by kit on nine jobs in ten |
| 140 | in R37D |
| 106 · 87 | in Breakout and in Clarity; 63 jobs had no workbook at all |
Three levels of quality follow. High: 129 jobs with the cost by kit, plus price, margin and outcome. Medium: 730 with total cost, price, margin and outcome. Low: 437 still open or with no margin on record. Checked against each other: what D365 invoiced matches the last quote in Sugar, and D365 holds only direct cost, so its margin reads about ten points above the quote.
Counts as measured on 14 October 2026, and shown in Film 1.
A request arrives: a drawing of a display, a quantity, and a retailer. Before anything else, it raises two questions. What will it cost us? And what should we charge?
Each kit's cost is split into Outform's own categories, from the 714 cost sheet: material, labour, packing, freight, tariff and duty, and prototypes. Only kits split the same way can be compared.
Is it the same job? A host, that this one updates? Or a different display from the same family? Each one is just another comparable: the agent reads what it shares and what differs, and weighs it. Then it scales by quantity, checks air or ocean, and separates one-off design from per-unit cost. Tools only fetch and calculate; the thinking lives in the agent and its skill, and the skill improves with every test.
The new kit starts from the past kit's own split. The design blocks carry over unchanged. Freight is spread again, over the new quantity and the new route.
Take Haus Labs, job 270499. Thirty-one past wins, between 32% and 70% margin. The recommended start is 60%: $319.67 a unit, on a cost of $127.87. Outform won this job at 57%.
Four checks decide it. Today job 270499 reads low, zero of four, and it says why. A low that tells you why beats a confident wrong number.
Eight real past jobs, with known cost and outcome. Before each round, the expected answers are sealed. Each job runs twice, without seeing the answer. The prediction is scored against reality. Every miss becomes a change to the skill or a tool, and then every scenario runs again.
A kit is one thing Outform ships to a store, made of parts. Its workbook row gives the units and the cost per unit. Of the 900 jobs we read, 254 hold only parts and totals, so they cannot serve as comparables. Normalising writes the kit row from the job's own numbers: units, material, cost per unit.
| 5 | refused because the job was priced as one lot |
| 1 | refused because its recorded cost was for another item |
| 3 | jobs with several kits: one splits by its own recorded material per kit; in the other two the recorded cost belongs to one item only, so the parts outside it were refused |
Never by adding up the parts: they match the recorded material on only 8 of 57 jobs. Whether the comparison agent finds and uses the new kits is being measured tonight.