Emily Li
Menu
Role proposal · product marketing and growth, one seat

Watch, judge, fix, measure.

Most teams have AI on their desks and still run marketing the way they did before it. This is a proposal for one seat that changes that: AI does the watching and counting, a marketer does the judging, fixing and measuring, across every channel the brand sells through, with the AI set up as skills and routines rather than a chatbox.

Three things I keep seeing

AI is a chatbox, not a colleague. It drafts captions. Nobody has given it a job. The fix is to give it one: a skill per channel, a written playbook of what to pull and what counts as a flag, and a routine that runs those skills every Monday and reports what moved. Issues and opportunities surface on a schedule, not by luck.

Changes get made and never checked. A subject line, a price, a listing bullet, then everyone moves on. With AI keeping the log, every change gets a date on which its result comes back. Nothing gets lost, however small.

Creative runs on opinion. Most companies run one A/B test and let it sit. With tracking in place, every campaign becomes a test, and the creative gets better every month.

One seat, not a replacement

I am not here to replace your team. The seat runs the loop, week after week: watching every channel, judging what matters, shipping the fix with the person who owns it, and reading the result when it lands. The routines are never finished, because the market keeps moving. What changes is that your people stop chasing numbers and spend their time on the calls only people can make.

The Buddy Bites and Native Union studies on this site found their issues by watching one channel from the outside, with no internal numbers. That is the routine I would build for you across website, marketplace, social and email, with your data from Google Analytics, Seller Central, Meta and Klaviyo.

01 · One week, every week

Five steps. Two of them are a machine’s; three are a person’s. The loop closes on itself: every change gets a date to be checked, and every result comes back to leadership.

01

Monday, first thing

AI pulls the week

Four channels, one set of numbers. Every metric against its own four-week average and its threshold. What moved gets flagged; what did not gets one line.

02

Monday morning

I judge

Read the flags, discard the noise, decide what is an issue, what is an opportunity and what is nothing. Draft the action for each: a fix, a test, or a question for a team.

03

Monday, end of day

One page to leadership

Issues, actions, tests running, what last week’s fixes did. Anything needing a decision is marked as such, usually money or stock. Twenty minutes, not a meeting.

04

Tuesday to Friday

Execute with the teams

Fix the flow, write the email, brief the listing change, launch the test. I carry the work to whoever owns the channel and ship it with them.

05

The moment anything ships

AI dates the check

Every change is logged and a check is scheduled for +7 and +28 days on the metric it was meant to move. The result comes back into a future Monday page.

AIMe

Why the middle three are a person’s

A flag is not a decision

AI can say revenue per recipient fell 20%. It cannot know that Wednesday’s send went to a cold segment on purpose, or that the drop can wait because a launch is three days out. That judgement is the difference between a report and a page leadership can act on.

A draft is not brand copy

AI writes a competent first version of anything. A design-led brand does not ship its own voice, a listing bullet or a caption unreviewed. Someone with taste decides what stays, and writes the second variant the test needs.

A report has no owner

When a fix goes wrong, someone reverses it, explains it, and carries the lesson into the next test. Accountability sits with a person, and it is what lets a manager spend twenty minutes on a Monday instead of a whole day.

The tests that matter are not in the data

The material story, the founder letter, the pre-launch waitlist: none of these came from a metric moving. They came from watching other brands and knowing what this one could do. AI finds the drops; the seat finds the ideas.


02 · Four channels, one set of numbers

What is pulled every Monday, where from, and what makes it a flag. Thresholds are starting points. Leadership sets the money floors, and every threshold is revisited after four weeks of real data.

ChannelPulled weeklyFromFlags when
SocialReach, engagement rate, net follower change, saves and shares on the top three posts. Paid: CTR, CPM, frequency, ROAS.Meta Business Suite
TikTok Studio
Ads Manager
engagement −25% vs 4-week avg
frequency > 3.0
ROAS below the floor
MarketplaceSessions and unit-session % by top ASIN, BSR, Buy Box share, reviews added and rating, ACoS and TACoS, price versus site.Seller Central
Ads console
Helium 10
unit-session % −20%
Buy Box < 95%
rating −0.2 in a week
EmailRPE (revenue per email and per recipient), by flow and by campaign; click-through and click-to-open; bounce rate, hard and soft; unsubscribe and complaint rate; list growth net of churn.KlaviyoRPE −20% vs 4-week avg
CTR −25% vs 4-week avg
hard bounce > 0.5%
complaint rate > 0.1%
list growth net negative
WebsiteSessions by source, conversion rate, AOV, add-to-cart rate, checkout abandonment. New product pages against their category median after seven days. Page speed. Behaviour on the top landing and product pages, with session replays pulled for anything flagged.Shopify
GA4
Clarity
CVR −15% week on week
new PDP below category median at day 7
abandonment +10 pts
rage or dead clicks > 5% of sessions

03 · What lands on leadership’s desk

One page. Issues with an action attached, tests in flight, last week’s fixes with their measured result, and the decisions that belong to leadership. The sample below is the shape, not the numbers.

Monday page · Week 39Prepared by Emily, with AI · four channels checked that morning

Issues this week, ranked

Email
CriticalAbandoned-cart flow earning $X.XX per recipient against $3.65 for the average brand on Klaviyo and $28.89 for the top decile. Wrong trigger, two flows, one dead image.Rebuild as one three-step flow on the Added to Cart event: reminder at 2 to 4 hours, one offer at 24 hours, last call at 48. Half a day in Klaviyo. Check dated +28 days on revenue per recipient.No decision needed
Marketplace
HighHero listing’s unit-session % down XX% since a competitor cut its price by $X; ad spend on the ASIN unchanged, so cost per order is up. Run a coupon on the listing for two weeks as a test; rewrite the first bullet to lead with the difference the competitor cannot claim. Check dated +14d on unit-session % and cost per order.Decision: coupon depth
Website
MediumNew product page converting at X.X% against a X.X% category median, ten days after launch. Add-to-cart is fine; the drop is at checkout, where replays show rage clicks on the shipping estimator.Fix the estimator; A/B the free-shipping line on the product page. I write both variants; dev about two hours.No decision needed

Opportunities and tests

Email
VerdictSubject-line test on Thursday’s campaign: the question form won. Click rate X.X% vs X.X%, +XX% on a 50/50 split.Question-form subject becomes the default for the next three campaigns; re-tested next month against a benefit-led line.
Social → Email
TestThe material-story post drew the most saves of the month. Propose it as the first editorial email: material story versus product-tile campaign, same list, split 50/50.I draft both; send Thursday. Check dated +7 days on click and revenue per recipient.Decision: approve the test and the sample split

Fixed, and what it did

2 weeks ago
Duplicate cart flow switched off. +7d: second-email sends 0, cart-flow unsubscribes −XX%. +28d due next week.
3 weeks ago
Hero SKU title rewritten and main image swapped for an in-use shot, run as a split test. +28d: B wins, unit-session % +X.X pts, sales +XX% at 9X% confidence. Rolled out to the range. Kept.
Last month
Month in one line: XX changes shipped, XX measured so far, XX kept, X reversed. Net effect on tracked revenue +$X,XXX against baseline.

Decisions for leadership this week

1. Coupon depth on the hero listing. 2. Approve the editorial email test and its sample split. Everything else is executed without you.


04 · Every change is dated

The part that usually goes missing. A subject line, a price, a listing bullet, an ad creative: the moment any of it ships, it goes in the log with a check scheduled on the metric it was meant to move. Small changes are the ones nobody measures. This is where they stop being.

ShippedChannelChangeMetric watched+7d+28dVerdict
Wk 37EmailDuplicate cart flow offCart-flow unsubscribes; sends per trigger−XX%due
Wk 36MarketplaceHero SKU title + main image, split testUnit-session %; salestest live+X.X pts · +XX%kept
Wk 35WebsiteMobile page speed: images, deferred scriptsMobile PDP → cart; mobile CVR+X.X ptsCVR +XX%kept
Wk 35WebsiteFree-shipping threshold raisedAOV; checkout abandonmentAOV +$Xabandonment +X ptsreversed

The verdict column is the point. Kept, reversed, or too early. A month of verdicts is a month of evidence about what actually moves this business, which is worth more than any single result in it.


05 · The role, and the first four weeks

Three things in one seat. None of them is new on its own; the combination is what makes the loop run, and what empowers the team behind it.

Content

Every test needs two versions written

The product-page line that gets A/B’d, the subject line and the founder letter, the post caption and the ad’s first sentence, the listing title and bullets that go into a split test. All in the brand’s register, ready for review. Not a brief for an agency.

Performance

Four channels, one discipline

The same test-and-measure loop on a product page, a subject line, a post format and a listing bullet. On every channel there is a gap between what the brand gets and what the best in its category get from the same traffic. Closing that gap, channel by channel, is the job.

AI leverage

Agents on a schedule, not a chatbox

Each channel gets its own skill: a written playbook of what to pull, how to normalise it, and what counts as a flag. A scheduled routine runs the skills every Monday, writes the change log, and posts the +7 and +28 day checks back into the page without being asked. The case studies on this site were built with the same skills. The agents make one seat cover four channels. They do not make the seat optional.

The first four weeks in the seat

Want this run for your brand?

The creative, the channels and the numbers, in one seat. The loop is ready. Let’s talk about running it for you.

Interview me