type.com team guide
AI for ecommerce: a practical guide
Seven Shopify workflows, from the daily revenue number to stockouts and promo margin, each with an owner who approves before anything changes.
What does this guide set up for a Shopify brand?
This guide sets up one shared Ecommerce Space where Claude runs seven recurring store workflows, from the daily revenue number to review mining, while a person approves anything that changes the store or spends money.
This guide is for the founder, ecommerce manager, or head of growth at a Shopify brand doing roughly $2M to $50M a year, where a handful of people own the daily number, the ad accounts, and the reorder sheet.
By the end you will have seven workflows posting into channels your team reads, each with instructions, an owner, and a review rule. Together they typically replace 10 to 20 hours a week of exports and spreadsheet joins, depending on your channels and SKU count.
A common setup is one AI bot per job, plus a manager bot to route between them. type.com takes the other route: one Space per department, with channels for workflows. Every channel shares the same Shopify connection, cost sheets, and instructions, so the stockout check and the promo plan read the same numbers. Routines become scheduled automations, and drafts wait in shared threads for approval.
Everything here works with plain instructions you paste in. Where a ready-made skill fits, it appears at the end of a workflow as an optional shortcut. For the full list, see Best AI skills for ecommerce teams.
The examples use Fernlow, a fictional skincare brand doing about $14M a year on Shopify. Nora is the founder, Theo runs growth and paid social, Ines runs operations and purchasing, Ruth leads CX, and Dev is the part-time finance lead.
| Workflow | When | Channel | Owner | Time it replaces |
|---|---|---|---|---|
| Daily revenue check | Daily, 7:30 AM | #daily-numbers | Dev | 2–3 hours a week |
| Stockout and reorder risk | Monday, 7:00 AM | #ops-inventory | Ines | 2–3 hours a week |
| Weekly marketing efficiency | Monday, 8:00 AM | #growth | Theo | 3–5 hours a week |
| Creative fatigue and budget review | Tuesday, 9:00 AM | #paid-social | Theo | 2–4 hours a week |
| Feed, product page, and AI search checks | Friday, 8:00 AM | #catalog | Theo | 1–2 hours a week |
| Review and support insight | First Monday, 10:00 AM | #customer-voice | Ruth | 3–4 hours a month |
| Promo plan and readouts | 4+ weeks before each promo | #promos | Nora | 4–8 hours a promo |
How do you set up an Ecommerce Space in 30 minutes?
Create one private Ecommerce Space on Claude, connect Shopify, Meta Ads, your cost sheets, and Slack with read access first, add a channel per workflow, and write your revenue definitions into the Space instructions.
You need Shopify admin rights, admin access to the Meta Business Portfolio, and 30 minutes. The Shopify setup guide and the Meta Ads setup guide cover each connection screen by screen.
- 1
Create the Space and make it private
Choose Create, then Space, describe the work or start from the Data template, and choose Claude. Make it private, since it holds margins and supplier terms, and add the owners.
- 2
Connect Shopify
In Space settings → Connections, choose Shopify, then Connect to Shopify, and approve the type.com app. Test with a read: ask for the store name and the five newest products.
- 3
Connect Meta Ads with a reporting-only token
Create a Meta system user, assign it the ad account, and generate a token with ads_read only. Paste it into the Meta Ads connection, never into a message.
- 4
Add the cost, supplier, and promo files
Connect Google Drive and select only the COGS sheet, supplier terms, and promo calendar. Bring Google Ads, Merchant Center, and review data in as CSV exports at first.
- 5
Create the channels and map Slack
Add daily-numbers, growth, paid-social, ops-inventory, promos, catalog, and customer-voice. If the team lives in Slack, map each to its Slack channel with Mentions only.
- 6
Write the Space instructions
Paste the starter below and have Dev confirm the definitions. type.com includes Space instructions with every message, so every workflow uses the same net sales.
Starter Space instructions (Space settings → Details → Instructions)
You support Fernlow's ecommerce team: Nora (founder), Theo (growth and paid social), Ines (operations and purchasing), Ruth (CX), and Dev (finance). Store facts - Shopify store, reporting timezone America/Los_Angeles, currency USD. - DTC sales run through Shopify. Orders tagged "wholesale" are not DTC revenue. Definitions (owned by Dev; change them only with his approval) - Gross sales: product price × quantity before discounts. Excludes tax, shipping charged, and gift card sales. - Net sales: gross sales − discounts − returns. Returns count on the date the refund is processed. - New customer: a customer whose first non-cancelled order is in the period. - Contribution margin: net sales − landed COGS − shipping and fulfillment − payment fees − expected return costs. "After ads" also subtracts all marketing spend. - MER: net sales ÷ total marketing spend. Platform ROAS is always labeled with its platform and never added together. Sources of truth - Sales, orders, inventory: Shopify. Costs: the "Fernlow COGS 2026" sheet. Lead times, MOQs, case packs: the "Supplier terms" sheet. Spend: Meta Ads, the weekly Google Ads export, and the influencer log. Rules - Read and analyze freely. Draft in the thread. - Never change prices, inventory, discounts, products, ads, budgets, or the feed, and never message customers, unless the named owner approves that exact change in the thread. - If a source is missing or stale, name it and stop. Do not estimate around it. - Show every number with its period, source, and data-through time. - Refer to orders by order number, never by customer name or email.
| Connection | Start with | Upgrade later, if at all | Why |
|---|---|---|---|
| Shopify (type.com app) | Reads only, per the Space rules; test any write in a development store | Owner-approved edits to unpublished pages or draft discounts | Orders, refunds, discounts, inventory, products |
| Meta Ads | System-user token with ads_read | ads_management for one approver's numbered changes | Ad, creative, and insights data |
| Google Drive | Selected files: COGS, supplier terms, promo calendar | Edit access to one readouts folder | Landed cost, lead times, MOQs |
| Google Ads and Merchant Center | Weekly CSV exports | A read-only connection | Search spend and feed diagnostics |
| Analytics tool or warehouse (for example Polar Analytics or BigQuery) | Read-only, if you already use one | None needed | Blended spend, sessions, cohorts |
| Reviews and helpdesk (Judge.me, Okendo, Gorgias) | Monthly CSV export | A read-only connection | Review text and first customer messages |
| Slack | Mapped channels, Mentions only | Channel messages for #daily-numbers | Where the team already reads results |
How do you run a daily revenue check the whole team trusts?
Agree the definitions once, have AI rebuild yesterday's net sales from Shopify each morning, run a short list of checks, and post the number only when every check passes.
The daily number breaks in predictable ways. One dashboard shows gross and another shows net. A wholesale order inflates Tuesday. A refund for an August order lands today and reads as a bad day. Fix the definitions first (they are in the Space instructions above), then automate the check.
It runs daily at 7:30 AM Pacific into #daily-numbers, owned by Dev, from Shopify orders, refunds, and discounts.
Shortcut: the Daily Revenue Check skill returns PUBLISH or HOLD across Shopify, Amazon, and wholesale using finance's revenue definitions file. The Ecomm Starter Skills collection adds four skills for revenue definitions, sales reporting, marketing measurement, and reconciliation.
Automation instructions: daily revenue check
Daily revenue check for yesterday, 12:00 AM to 11:59 PM Pacific. 1. From Shopify, pull every order created yesterday and every refund processed yesterday. 2. Remove test orders, cancelled orders, and orders tagged "wholesale". List what you removed and its value. 3. Using the definitions in the Space instructions, calculate gross sales, discounts, returns, net sales, orders, AOV (net sales ÷ orders), and new vs returning orders. 4. Compare with the same weekday last week and with the 28-day daily average. 5. Run these checks and mark each pass or fail: a. Data is complete: the latest order is within 2 hours of midnight. b. No single order over $2,000 is unexplained. c. Refunds on orders older than 30 days are listed separately with their original order month. d. Net sales match the Shopify sales report total within 1%. If you cannot read that report, ask Dev to paste the total and mark the check pending. 6. If every check passes, post PUBLISH, the bridge (gross → discounts → returns → net), and up to three bullets on what moved. If any check fails or is pending, post HOLD, the check, and what Dev must confirm. Never post a net sales figure on a HOLD. Keep the post under 120 words.
- Gross and net get mixed up: show both on every post with the bridge, so nobody quotes the wrong one.
- Late refunds look like a bad day: list refunds on old orders separately, with the month of the original order.
- Partial data at 7:30 AM: the freshness check turns an incomplete day into a HOLD instead of a low number.
- Review rule: Dev reads every post for the first two weeks, then only HOLDs. Nobody forwards a number marked HOLD.
How do you measure weekly marketing efficiency without trusting platform ROAS?
Lead with blended MER, new-customer efficiency, and contribution margin after ads, all from your own sales and costs, and show platform ROAS only as labeled context that is never added up.
Meta and Google each count the same order when both touched it. In Fernlow's week of September 28, they claimed $283,400 of revenue against $268,400 of actual net sales, so the scorecard starts from Shopify and the cost sheet.
It runs Monday at 8:00 AM Pacific for the prior Monday to Sunday, into #growth, owned by Theo. Inputs: Shopify, Meta Ads, the Google Ads export, the influencer log, and the COGS sheet, which Dev checks monthly.
The same pattern works for any recurring report; how to automate a weekly team report covers it end to end.
Shortcut: the Marketing Measurement skill writes down your attribution windows, cost pool, and new-customer rule before it calculates MER, ROAS, and CAC. For monthly cohort payback, the Cohort LTV Analyzer builds repeat rate and contribution-margin LTV by first-order month from an order export.
Automation instructions: weekly marketing efficiency
Weekly marketing efficiency for last week, Monday to Sunday, Pacific time. 1. From Shopify: net sales, orders, and new vs returning customers, using the Space definitions. 2. Total marketing spend: Meta Ads, the latest Google Ads export, and the influencer and affiliate log. If any source is missing or more than 2 days old, name it and label every total that uses it as partial. Missing spend is unknown, not zero. 3. Calculate and show each with its formula: - MER = net sales ÷ total marketing spend - New-customer MER = first-order net sales ÷ total marketing spend - New-customer CAC = total marketing spend ÷ new customers - Contribution margin after ads = net sales − COGS − shipping and fulfillment − payment fees − expected returns − marketing spend 4. Compare each with the prior week and the 4-week average. Add the week's totals first, then divide; never average daily ratios. 5. In a separate table, show Meta and Google platform ROAS with their attribution windows. Never add platform revenue together or call it sales. 6. Note the date of the COGS sheet you used. 7. End with up to three observations, each tied to a number, and one question for Theo. No budget recommendations here.
- Averaging ratios: add a week's sales and spend first, then divide, or one small day skews the result.
- Missing spend counted as zero: an absent influencer log makes MER look better than it is, so mark the week partial.
- A stale cost sheet: contribution margin is only as current as landed cost, so date the COGS file on every scorecard.
- Review rule: Theo brings three numbers from the scorecard to the Monday growth meeting; nothing in it changes a budget.
How should AI review Meta creative fatigue and ad budgets?
Have AI compare each creative concept week over week, label it fatigued, watch, or scale, and propose numbered budget moves that a named owner approves before anyone touches Ads Manager.
Fatigue is rarely one metric. A concept is tired when frequency rises, link CTR falls, and CPA climbs together while spend keeps flowing to it, and the offer, landing page, tracking, and stock did not change.
It runs Tuesday at 9:00 AM Pacific into #paid-social, owned by Theo, from Meta Ads (ads_read) and the Monday scorecard.
Review rule: Theo approves by number in the thread. With ads_read, AI cannot apply anything, so Theo makes the change in Ads Manager and the next run reads budgets back. If you later grant ads_management, keep one named approver, only numbered moves, and a read-back after each change.
Shortcut: the Meta Creative Fatigue skill packages this review: it groups ads by concept, compares the last 7 days with the 7 before and the last 30, checks format and placement coverage, and drafts refresh briefs without launching anything.
Automation instructions: creative fatigue and budget review
Weekly Meta creative and budget review. Read-only: do not change any ad, ad set, budget, or status. 1. Pull ad-level results for the last 7 days and the 7 days before, using the account's attribution setting. Group ads by creative concept, not by crop or placement. 2. For each concept with at least $1,000 of spend in the last 7 days, show spend, frequency, link CTR, CPM, CPA, and purchase ROAS for both periods. 3. Label each concept: - FATIGUED: frequency up, link CTR down 20% or more, and CPA up 25% or more, while still spending over $500 a day. - WATCH: two of those three signals. - SCALE: CPA at least 20% below the account target for 7 days, with frequency under 2.5. Before calling anything fatigued, check for an offer, landing page, tracking, or stock change that explains it better. 4. Propose budget moves as a numbered list: ad set, current daily budget, proposed budget, reason. No single increase above 20% of that ad set's budget. Keep total daily spend flat unless Theo asked otherwise. 5. For each FATIGUED concept, draft a two-line refresh brief: the hook to replace and two new angles drawn from top reviews. 6. Read the latest #ops-inventory check. Never propose scaling a product with under 21 days of cover. End by asking Theo to approve moves by number.
- Judging new ads too early: set a minimum spend before any label, or every launch looks like a failure on day two.
- Spending into a stockout: the inventory cross-check stops AI from scaling a serum that runs out in 14 days.
- Chasing platform ROAS: tie every scale proposal to the scorecard's new-customer MER, not only to Meta's own number.
How do you catch stockouts before they cost revenue?
Compare each SKU's days of cover with its supplier lead time every week, rank the gaps by revenue at risk, and let the buyer decide whether to reorder, expedite, or pull ad spend.
A stockout on a hero SKU costs twice: the orders lost while it is out, and ad spend still sending shoppers to a sold-out page. Run the check before the weekly purchasing call.
It runs Monday at 7:00 AM Pacific into #ops-inventory, owned by Ines, from Shopify inventory and sales, the supplier terms sheet, and landed cost.
Review rule: Ines decides every quantity and places POs herself. Products flagged for less ad spend go to Theo's Tuesday review.
Shortcut: the Inventory Stockout Risk skill computes days of cover from sales velocity, ranks stockouts by revenue at risk, sizes reorders against lead time, MOQ, and case pack, and shows the capital frozen in overstock.
Automation instructions: stockout and reorder check
Weekly stockout and reorder check, as of last night's close. 1. For every active SKU, compute average daily units sold over the last 28 days and the last 7 days. Exclude promo days from the baseline. If a promo is on the #promos calendar in the next 60 days, use the higher of the two rates. 2. Days of cover = available units in Shopify ÷ daily units. Count open POs only from their expected arrival date. 3. Compare days of cover with lead time plus 14 days of safety stock, from the "Supplier terms" sheet. 4. Rank every SKU that will run out before new stock arrives by revenue at risk = days out of stock × daily units × average selling price. 5. For each, suggest one action: reorder now (round up to the case pack and at least the MOQ), split or expedite an open PO, or reduce ad spend on that product. Say what Ines needs to confirm with the supplier. 6. List SKUs with more than 180 days of cover and the landed cost tied up in each. 7. If an inventory figure looks wrong (negative, or unchanged for 14 days while selling), flag it instead of using it. Do not create purchase orders, change inventory, or edit products.
- Velocity taken from a promo week: excluding sale days keeps a Black Friday spike from triggering a six-month reorder.
- Invoice cost instead of landed cost: freight and duties are real cash, so use landed cost or the overstock figure looks smaller than it is.
- Lead times in someone's head: put every supplier's lead time, MOQ, and case pack in the sheet, or every SKU gets the same default.
How do you plan promos with margin guardrails and read them out honestly?
Before a promo, have AI solve the deepest discount each SKU can carry and draft exclusions for approval; afterward, compare contribution margin with the plan, including the returns that arrive weeks later.
Promos look good on day one because revenue arrives at once and costs arrive later: returns, reships, and repeat purchases that would have happened at full price. A guardrail makes the sale a decision per SKU.
It runs at least four weeks before each promo (for Fernlow, Black Friday to Cyber Monday, November 27 to 30), with readouts 3 and 45 days after, in #promos. Nora owns it; Ines and Dev sign off.
Cannibalization is the quiet cost: compare the two weeks before and after with the same weeks last year, so demand pulled forward from December is not counted as new.
Shortcut: the Promo Margin Guard skill prices a promotion after COGS, shipping, payment fees, refunds, and expected returns, names the codes that lost money, and solves the breakeven discount for every product.
Instructions: promo plan with margin guardrails, plus readouts
Promo plan for [promo name, dates, planned discount]. 1. For each SKU, take landed cost, shipping and fulfillment per order, payment fees, and the 90-day return rate from the COGS sheet. Compute contribution margin per unit at full price and at the planned discount. 2. Solve the breakeven discount: the deepest discount that keeps contribution margin per unit above $0. Show it beside the planned discount. 3. Flag SKUs to exclude or cap: breakeven below the planned discount, under 45 days of cover in the latest inventory check, or launched in the last 30 days. 4. Find the last two comparable promos and report units, net sales, contribution margin, and 60-day return rate for each. 5. Draft the plan as a document: discount structure, exclusions with reasons, projected orders and contribution margin per order, and approval lines for Nora, Ines, and Dev. 6. Do not create or edit discounts in Shopify. After approval, list the exact discount settings for Theo to enter. Readouts, 3 and 45 days after the promo ends: actual vs plan for orders, net sales, discount cost, contribution margin, new customers, and returns to date. On the day-3 readout, label returns as incomplete.
- One sitewide depth on thin-margin SKUs: a travel kit with an 18% breakeven loses money on every order at 25% off.
- Day-3 victory laps: returns on skincare often arrive weeks later, so the day-45 readout is the one that counts.
- Discounting what is about to sell out: a promo on a SKU with thin cover just moves the stockout forward.
How do you keep the product feed, product pages, and AI search visibility healthy?
Run a weekly feed and product page check that lists exact item ids and the system that owns each fix, plus a monthly check of whether your product pages answer the questions shoppers ask AI assistants.
Feed problems are silent: Merchant Center disapproves a product and it stops serving, often days before anyone notices. Product pages drift too, with sizes or claims the reviews contradict.
Shoppers also ask AI assistants what to buy, and pages that state size, ingredients, and who a product is for answer them plainly. Treat it as a monthly check, not a new channel.
In Fernlow's latest check, 23 items were disapproved for a missing GTIN, all travel kit bundles from one template: one fix, not 23.
It runs Friday at 8:00 AM Pacific into #catalog, owned by Theo, from a Merchant Center export and Shopify products. Review rule: Theo assigns fixes, and a person publishes any rewrite.
Shortcuts: the Product Feed Auditor skill checks a feed export for missing attributes, policy problems, broken URLs, and price conflicts, and never edits the feed. The Shopify AEO Checker skill takes one product link, searches Shopify's cross-store Global Catalog with shopper-style queries, and shows how often your brand appears against competitors.
Automation instructions: weekly catalog check
Weekly catalog check. 1. From the attached Merchant Center export, list disapproved and warning items grouped by cause (missing GTIN, price mismatch, image problem, policy). Give the exact item ids for each cause. 2. From Shopify, check every active product: a title that names the product type, a description that states size, key ingredients, and who it is for, at least 3 images, and a price and compare-at price that match the feed. 3. Flag product pages whose claims conflict with themes in the latest #customer-voice report, such as a "fragrance-free" claim alongside reviews that mention scent. 4. For each problem, name the fix and the system that owns it: the Shopify product, the feed app, or a Merchant Center setting. 5. Draft rewrites for the five weakest product descriptions in the thread. Do not publish or edit products. First Friday of the month: for the five best-selling products, write five questions a shopper might ask an AI assistant (for example, "best vitamin C serum for sensitive skin") and say whether the product page answers each one plainly.
- Fixing items one at a time: group by cause, because a broken bundle template or image sync is one fix.
- Rewriting copy the reviews contradict: read the latest #customer-voice themes before changing a claim.
- Writing for the AI instead of the shopper: plain answers to a shopper's real question serve both.
How do you turn reviews and support tickets into product and copy fixes?
Once a month, have AI group reviews and first support messages into themes, split product defects from expectation mismatches, and route each theme to the person who can fix it, with verbatim evidence.
Reviews and tickets are a brand's cheapest research. The useful split: is the product failing (operations), or is the page promising the wrong thing (whoever writes pages and ads)?
Fernlow's September run found 41 complaints about the Barrier Cream pump jamming, a defect for Ines, and 28 saying the Vitamin C Serum smells odd despite its "fragrance-free" line. It has no added fragrance, just a natural vitamin C scent, so that fix is one sentence of copy.
It runs the first Monday of each month at 10:00 AM Pacific into #customer-voice, owned by Ruth, from a CSV of reviews and first support messages.
Review rule: Ruth checks the themes and assigns the top five. Check your review platform's terms before quoting reviews in ads.
Shortcut: the Review Insight Miner skill clusters reviews and tickets from a single CSV, separates defects from expectation mismatches, ranks themes by the stars they cost, and attaches a fix and an owner to each.
Automation instructions: monthly review and support insight
Monthly review and support insight for last month. 1. Use the attached reviews and support export. Keep only the first customer message from each ticket. Remove names, emails, and addresses before analysis. 2. Group reviews and tickets into themes by what customers describe. For each theme give the count, the SKUs, the average star rating of its reviews, and three short verbatim quotes. 3. Label each theme: product defect, expectation mismatch (the page or an ad promised something else), shipping and delivery, or praise worth reusing. 4. Rank themes by volume and by how far their rating sits below the product's average. 5. For each of the top five, propose one fix and an owner: Ines (product or supplier), Theo (product page or ad copy), or Ruth (help center or macro). 6. List phrases customers repeat in praise that do not yet appear on the product page. Do not reply to reviews or customers.
- Mining years of reviews at once: window it to last month or since a supplier change, or old issues bury the current one.
- Whole ticket threads: agent replies and signatures swamp the customer's words, so use first messages only.
- Themes without owners: every top theme gets one name, or it is still on the list next month.
What does the weekly operating rhythm look like?
Seven automations post into seven channels on a fixed calendar, and each result has one person who reads it and decides what happens next.
Each workflow above becomes a scheduled automation. Scheduled runs use the creator's identity and connection access, so the person who owns the data should create each one: Dev creates the daily check, Ines the inventory check. If a run cannot reach a connection, check the creator's access first.
Promos run on demand: Nora starts the plan in #promos at least four weeks out and asks for each readout in the same thread on its date.
- 1
Choose Create, then Automation
Pick the Ecommerce Space. New automations start paused.
- 2
Paste the instructions and pick the channel
Choose the destination channel and set the schedule in Pacific time.
- 3
Save, then Test
Test runs privately as you. Compare it with a day you already know and fix the instructions until it matches.
- 4
Enable
Check the first live run. Three failed runs in a row pause a schedule.
| When | Workflow | Channel | Owner | What the person does |
|---|---|---|---|---|
| Daily, 7:30 AM | Daily revenue check | #daily-numbers | Dev | Reads every HOLD and confirms the fix |
| Monday, 7:00 AM | Stockout and reorder | #ops-inventory | Ines | Decides reorders before the Tuesday supplier call |
| Monday, 8:00 AM | Weekly efficiency | #growth | Theo | Brings three numbers to the growth meeting |
| Tuesday, 9:00 AM | Creative and budget review | #paid-social | Theo | Approves moves by number and applies them |
| Friday, 8:00 AM | Catalog check | #catalog | Theo | Assigns feed and product page fixes |
| First Monday, 10:00 AM | Review and support insight | #customer-voice | Ruth | Routes the top five themes |
| 4+ weeks before, then days 3 and 45 after | Promo plan and readouts | #promos | Nora | Approves depth and exclusions with Ines and Dev |
What guardrails keep AI from changing the store or spending money?
Write down what AI may read, draft, and do only with approval for each tool, enforce it in the source tool where you can, and keep customer data out of prompts and exports.
In type.com, AI acts with the effective access of the person who started the work, and a read-only setting tells the agent not to make changes but does not block them at the connected service. So enforce limits at the source where you can. A Meta token without ads_management cannot edit ads.
Shopify needs the most discipline. The type.com app requests write permissions for tools such as inventory, discounts, and pages, so the Space instructions carry the rule, and any write is tested first in a development store. Nothing in this guide needs AI to change inventory.
The type.com Shopify app does not request direct access to Shopify customers, and protected customer fields on orders stay unavailable until Shopify approves that access. Work with order numbers, strip names from exports, keep margins in a private Space, and never paste a token into instructions or messages.
For patterns that keep a person in charge of every consequential step, see AI agent approval workflows.
| Tool | AI may read | AI may draft | Only with named approval | Never |
|---|---|---|---|---|
| Shopify | Orders, refunds, products, inventory, discounts | Product copy, promo plans, discount settings | Edits to unpublished pages, tested in a development store first | Change live prices, inventory, discounts, or orders |
| Meta Ads | Ads, creative, insights (ads_read) | Numbered budget moves, refresh briefs | Apply approved moves with an ads_management token, then read back | Launch ads or raise total spend |
| Google Ads and Merchant Center | Exports | Fix lists by item id | None | Edit the feed or campaigns |
| Google Drive | Selected cost, supplier, and promo files | Plans and readouts as documents | Updates to the COGS sheet (Dev) | Open files it was not given |
| Reviews and helpdesk | Exports with names removed | Themes, macro drafts | Help center updates (Ruth) | Reply to customers or reviews |
| Slack | Mapped channels | Posts in internal channels | None | Message customers, suppliers, or creators |
What does a 30-day rollout look like?
Turn on one or two workflows a week, starting with the daily number, and widen access only after a workflow has matched your own numbers for two weeks.
When an output is wrong, fix the instructions, not just the thread: memory carries context in the background, but instructions are what you control. More in shared AI memory for teams.
- 1
Week 1: definitions and the daily number
Create the Space, connect Shopify and Drive, and write the instructions with Dev. Run the daily check by hand for five known past days until each matches, then enable it.
- 2
Week 2: the scorecard and the inventory check
Run the scorecard for the last four weeks against your spreadsheet, and the stockout check beside Ines's reorder sheet. Enable both once they agree.
- 3
Week 3: paid social and the catalog
Connect Meta Ads with ads_read, run the creative review twice by hand, then enable the Tuesday and Friday automations.
- 4
Week 4: promos, customer voice, and an access review
Plan the next promo and run the first review mining. Then review Space members, connections, and whether anyone needs write access. Most brands do not yet.
How do you know it is working after 30 days?
Record a baseline in week one for time spent, corrected numbers, stockout days, and promo margin, then compare at day 30 instead of relying on impressions.
Judge the setup by whether owners make the same decisions faster, with fewer corrections. If Theo still declines most budget moves after three weeks, the thresholds are wrong; fix the instructions.
If marketing and finance run in their own Spaces, see AI for marketing teams and AI for finance and operations for the same approach applied to campaigns and the monthly close.
| Metric | Baseline in week 1 | Good after 30 days |
|---|---|---|
| Time to a trusted daily number | When the number was posted on each of the last five days | Posted by 8:00 AM Pacific most days, with every HOLD explained |
| Numbers corrected after posting | How many were restated last month | None restated; misses caught as a HOLD instead |
| Hours on weekly reporting | Each owner logs one week of pulling and formatting | Owner time goes to reading and deciding, not exports |
| Proposals approved | Not applicable before launch | Most moves approved as proposed, with declines explained |
| Hero SKU stockout days | Days any top-10 SKU was out in the last 90 days | Every gap flagged at least one lead time ahead |
| Promo margin vs plan | The last promo's contribution margin | Next promo's day-45 readout within plan, no SKU below breakeven |
| Feed disapprovals | Today's count in Merchant Center | Falling, with every cause assigned to an owner |
Frequently asked questions
How can a Shopify brand use AI to run ecommerce operations?
Give AI the recurring checks with fixed rules: the daily revenue number, weekly MER and margin, creative fatigue, stockout risk, promo margin, feed errors, and review themes. In type.com they run as scheduled automations in one Ecommerce Space, and an owner approves anything that changes the store or spends money.
Can AI change my Shopify store or Meta ad budgets on its own?
It can if you give it write access, which is why you should not at first. Start Meta with an ads_read token, keep Shopify read-only in your instructions, and have AI propose numbered changes that an owner approves in the thread. New type.com automations start paused until someone tests and enables them.
Is it safe to connect Shopify and ad accounts to AI?
It can be when access is narrow. Assign each connection only to the Space that needs it, use a private Space for margin data, never paste tokens into prompts, and work with order numbers instead of customer names. The type.com Shopify app does not request direct access to Shopify customers or theme files.
What is a good MER for a DTC brand?
It depends on your margin, so calculate your own breakeven instead of borrowing a benchmark. Breakeven MER is 1 divided by contribution margin before marketing as a share of net sales. A brand keeping 52% of net sales after COGS, shipping, fees, and returns breaks even at about 1.92x; anything above that funds overhead and profit.
Do I need an analytics tool or data warehouse before using AI?
No. Shopify, your ad accounts, and a cost sheet with landed COGS and supplier lead times cover every workflow in this guide. If you already use an analytics tool or a warehouse such as BigQuery, connect it read-only so AI can use blended spend and cohort data you have already cleaned.
How much does it cost to run these workflows in type.com?
type.com plans start at $50 a month for 2 members and $100 a month for 4, with included AI usage metered at the model provider's published rates. Members can connect their own Claude or ChatGPT subscriptions so supported runs don't draw down the workspace allowance.