The AmplifierReal cases · Founder · D2C
Kabir, 26: a snack brand, three people, and every job at once
Ads, listings, packaging, a co-packer who is late, and a founder who answers customer emails at midnight because nobody else can.
Planned for release phase 2. The page is complete as a draft; interactive parts run in the browser and do not yet save to an account.
In brief
The situation. Kabir runs a millet-snack brand out of Jaipur with two colleagues. Revenue is real and small. He does everything: ad copy, marketplace listings, supplier chasing, customer emails, the accounts. He uses AI for all of it, in bursts, and has a folder of forty half-finished outputs. A mentor asked him what the business would look like in a year and he described the next month.
Step 1 — Define the End State
"In twelve months a customer in Pune reorders without a discount because the product is worth it and the experience was easy. The three of us each own something. I spend my mornings on product and suppliers, not on listings. The accounts close monthly without me." Audiences: one reordering customer, his two colleagues, his accountant, and the co-packer.
Step 2 — Map the Flow
His list is the forty half-finished things. AI review, asked to sequence rather than add, tells him that thirty of the forty are one task — "make the product findable and orderable" — and that nothing in the list is about why the customer reorders. It adds: a reorder-reason survey, and a weekly supplier call. It suggests hiring an agency; the budget says no.
Step 3 — Assign Tasks
What the brand is and who it is for: his. Listing copy for three marketplaces: AI drafts from his master description; he edits. Ad variations: AI generates twenty; he chooses two and tests. Customer emails: a triage by AI into routine and not; routine answered from a bank he writes; not-routine, him. Monthly accounts: his colleague plus a bookkeeping tool, not a chat tool. The supplier relationship: him, in person, monthly.
Step 4 — Select the Tool
A conversational tool with a project holding the brand document; an image tool for listing graphics (never for the product photo — that is real); the marketplace's own analytics; a bookkeeping product for accounts, because a chat tool with numbers is where founders get hurt.
Step 5 — Prepare and Direct
Criteria for listing copy: the first line names the ingredient and a real benefit; nothing a competitor could not claim is removed; under the platform's character limit. For ads: two variables only per test; a result is a result only after a week. For emails: anything about a refund or a complaint goes to me.
Steps 6 and 7 — Execute and Evaluate
Listing drafts pass after one round. Ad variants mostly fail the "competitor could claim it" test; two survive and one of those wins the week. The email triage misroutes a complaint into routine — the criterion catches it because the bank has no refund answer, so the junior escalates. He tightens the triage rule.
Step 8 — Integrate and Stress-Test
Ninety days in, assembled against the twelve-month end state. Small fixes: listings inconsistent across platforms. Structural: reorders are not rising, and nothing in the ninety days touched the product or the unboxing — everything was findability. A Step 1 to Step 2 trace: the customer he named was never served by a task. He reallocates his mornings to product and supplier, as the end state said, and hands listings to a colleague with the criteria.
What changed
The forty half-finished outputs become six finished systems, three of them owned by other people. His mornings are his. The reorder rate moves in month five, after packaging changes that came from the survey, not from a tool.
Ask the tool to sequence your list before you ask it to add to it. Keep numbers out of chat tools. If the end state names a customer, check that a task actually serves them.
A composite case, not a client engagement; the person is invented, the situation is not.