Three products from one estate

Take the four use cases from session 2 that remove hours and need no outside payer: the content engine (420 hours a week off the studio's 1,200), customer service reply drafts (273.6 off 912), ad pre-flight and pacing (108 off 360) and the store ops copilot (120 off 480). On Serai's constructed block that is 921.6 hours a week, worth $15,188.80 a week to Serai after run costs, and none of it depends on a brand buying anything.

That independence is the point. Session 2's bench could not rank hours first above brand intelligence first, because the gap between them was made of attach fees somebody typed. What it could say is that hours first is certain: it needs nobody's permission but Serai's and no buyer at all. A product whose only customer is you is the one product you can ship without winning an argument outside the building.

So product 1 is not "a set of tools". It is one system, run by one team, with one backlog, whose job is to take repeatable work off the rota and keep it off.

Before any listing, video or ad ships, somebody checks that every claim on it is one the brand has approved for that market. A skincare line can say one thing on its home-market store and a narrower thing two borders away, and the brand's own team keeps that list. Today the check is a person reading a shared sheet and a queue of assets waiting for them.

In the house system the check is a lookup: the draft's claims are compared against that brand's approved list for that market, anything on the list passes, anything off it goes back to the brand's named approver. Serai does not decide what is allowed; the brand's list does. That keeps it where an operator can run it: operations, not interpretation. Serai's block does not size it as a separate row, so no number is claimed for it here. Count your own queue.

Each module needs data to run. The content engine needs the master SKU and every listing's current copy. The reply drafts need orders and their status. The ad pre-flight needs store-level spend. The copilot needs the campaign calendar. Build them properly and you end up with a warehouse of orders, listings and content keyed on Serai's own store and SKU ids, the grain session 4 showed holds.

Each module also emits something. If every asset leaves with a tag (brand, SKU, market, claim, and for content its format and creator), then in a year Serai owns an archive that says which content ran where and for which SKU. Nobody asked for the tags. Product 2 cannot exist without them.

And each module changes how people work: the studio starts from the engine's draft, service starts from the reply draft, both start from the same table. That habit is what session 3's eight teams needed on Pasar Live. It is also the hardest of the three to buy later.

Total hours removed is a vanity number: it grows when the rota grows. The measure that tells you whether product 1 is working is hours per brand served, the hours a week the four teams spend divided by the brands they run. On Serai's constructed figures, 2,952 hours across 48 brands is 61.5 hours a brand; with 921.6 hours removed it is 42.3.

Watch it when Serai signs new brands. If hours per brand stays flat as brands rise, the system is absorbing growth and each new brand needs fewer new hires. If it climbs back, the modules are being worked around. Either way it is a number the finance lead recognises, and it does not depend on anyone's GMV.

• Which creator actually sells? Answered on settled GMV, not attributed: what the creator's orders were worth after cancellations and returns. The mechanics, and why the ranking changes, are analyst session 6, Content to creator yield. Serai's bench block carries a $60,000 a quarter attach for the creator scorecard (constructed).
• Which content converts? Answered from the tagged archive: every asset product 1 shipped, with its SKU, market, format and creator, joined to what that SKU sold while it ran. Live stream scoring sits here, $40,000 a quarter on the block (constructed). The livestream methods behind a score belong to the live commerce analytics course.
• Where does each SKU stand, per platform? Price, sell-through, returns and content for the brand's own SKUs on every marketplace Serai runs. This is the grain the bench says holds: store grain at 100 percent, SKU grain at 97.5 percent (0.9748), because the keys are Serai's. $75,000 a quarter on the block (constructed).
• What sells next month? A demand forecast per SKU, on the same keys. $20,000 a quarter on the block plus a small GMV uplift (constructed).

Notice that none of the four needs to know who the buyer is. That is not an accident of the list; it is the list. Session 4 closed the customer grain, and product 2 is everything that survives it.

Brands already receive these answers, inside the retainer, as slides somebody spent a week building. Product 2 does not invent a new service; it turns an unpriced one into a line on the invoice. That is why it comes second: the demand is proven by every review meeting, and the supply is what product 1's warehouse and tags make cheap.

The attach fees above are the bench's stipulated inputs, typed by somebody, not measured anywhere. No page here claims what a brand will pay. The pricing decision belongs to a commercial owner, a named person who talks to brands, tests a price and brings back a signed number. Until then the data team's job is to make the answer good enough to charge for, and to say which grain it rests on.

Not the customer. "Your buyers on one platform also buy from you on another" is a customer-grain claim, and the best key combination on the bench reaches 0.23 at that grain. Any module that needs it is off the list, however often a brand asks.

Not the comparison, yet. Showing a brand its own SKU on three platforms is its own data, returned to it. Showing it how its category does across Serai's other 47 brands is a different product: every number in it came from data another brand or a platform handed Serai to do a job. Session 4 gave that line a commercial owner and a conversation in month one. Cross-platform comparison waits for that conversation to end in a yes, a yes with conditions, or a no.

Not a clean SKU view without upkeep. The 2.5 percent gap on the bench is 219 lines on listings missing from the mapping table. A product built on an owned key is only as good as the weekly check on that key.

Three forms, in rising order of effort. Playbooks: how to open a store, run a campaign week, set up a live room, written from what Serai's own tables showed worked. Agents on top of the operating knowledge: the house system's prompts, checks and claim lists turned into something another team can run. Training: the same knowing delivered to people, which is the cheapest to test and the slowest to scale.

The first buyers are people who gain from the knowing without taking Serai's brands: smaller enablers in other markets Serai does not serve, brands entering the region that want to learn before they hire an enabler, and adjacent verticals that sell through the same marketplaces with different products. Competitors in Serai's own markets come last, and only through a delayed, aggregated tier, later, if at all: nothing current, nothing that identifies a brand, nothing that hands away the lead the first two products built.

Serai knows the asset is real, because the house system runs on it every day. It does not know who pays for it, how much, or in which of the three forms. So there is no revenue line for product 3 on this page, and there should be none on your roadmap yet. The next step is three conversations, one with each buyer type, each with one question, before anyone writes a price list. A buyer who says "I would pay for the playbook, not the agent" has just saved a quarter of build.

Each product pays for the next, and each needs something only the previous one builds.

• 1 funds 2. The house system pays in hours from week one, with no buyer. It also builds the warehouse on Serai's store and SKU keys and the tagged archive. Without those, product 2 is a consultant with a spreadsheet, rebuilt for every brand.
• 2 funds 3. Brand intelligence turns an unpriced service into attach revenue, and every review meeting tests which answers brands value enough to pay for. That test is what tells you which knowing is worth packaging.
• 3 last, on purpose. It has the least certain buyer and the most exposure: it sells outside the building what the first two use inside it. Doing it last means the permission conversation from session 4 has had months to land, and the knowing has been proven on Serai's own brands first.

Run the order backwards and every step is harder: a packaged product with no operating system under it is a slide deck, and brand intelligence with no tagged archive is the free review deck again, with an invoice attached. The way the three land on a single roadmap is the playbook's roadmap on one page.