How I Run Marketing Ops and Paid Media With iHarness

In short: We ran iHarness on our own marketing for months before we opened it to teams. It runs as background loops rather than a chatbot: it works on the goal while we sleep, then interrupts us only when a decision needs a human. Most mornings start with four items on a phone screen, not an open dashboard.

We ran iHarness on our own marketing for a few months before we opened it up to teams. I wrote this as an internal doc for the people joining our marketing ops and paid media work, and a couple of customers asked for it, so here it is with the internal bits cleaned up.

I used to spend my day inside ad platforms, channel-specific tools, and spreadsheets, and my evenings reconciling what the platforms claimed against what the CRM showed. Now most of that runs while I sleep, and my day is spent approving decisions instead of assembling reports.

The thing people get wrong when I describe it: they picture a chatbot with our data plugged in. It is not. A chatbot waits for a question. iHarness is already working on the goal when I wake up, and it comes to me. Most days I do not ask it anything. It asks me.

Before you start

Connect the iCustomer Platform to everything the audience touches. For us that is HubSpot, Instantly, Snowflake, the site pixel, Meta, Google, LinkedIn, our DSP, Slack, and the sales team's call notes. iHarness runs on its own compute inside your data cloud, around the clock. Close your laptop and the loops keep observing, deciding, and learning. Nothing waits for you to log in.

One thing to be clear on up front: iHarness does not make creative and it does not run campaigns. Your martech builds the campaigns. Your adtech runs the media. iHarness watches the audience, works continuously toward your goals, KPIs, and targets, decides what to do with both, and measures whether it worked.

Three things are different from the personal AI setups most of us have used:

• One brain for the team, not one per person. Everyone talks to iHarness, in the app or as @iHarness in Slack. iHarness runs on your Growth Brain, the shared memory of the audience. When it learns something about a segment while working on my goal, the sales lead's goal knows too.
• The Growth Brain remembers the audience, person by person, account by account and cohort by cohort: who is showing intent, which segments convert, what a household or buying group responded to last quarter. It learns how you work too, but that is the smaller half.
• You do not drive it. It has goals, thresholds, and a loop running against each one. It decides within the limits you set and interrupts you when a decision needs a human. The chat window exists. It is the exception, not the interface.

What iHarness actually does

The team picks the goals, the KPIs, and the targets. iHarness takes it from there. It holds the goals and the rules, runs each goal as a loop (observe, orient, decide, act, learn), and does the work at each step. It is also the one thing I talk to when I do have a question, which is less often than you would think.

The work breaks into five jobs. None of them is a channel. That took me a while to accept and it changed how I work.

• Signals. Observes the pixel, the warehouse, the CRM, intent feeds, and named accounts for anything that changes: a cohort heating up, a segment going quiet, a buying committee forming at an account we care about.
• Audience. Turns signals into cohorts, resolves the people in them to identities the walled gardens will accept, builds suppression lists (current customers, open opportunities, people who unsubscribed last week), and hands the matched audiences to the platforms we already use.
• Decisions. Our current goal is qualified pipeline from mid-market fintech under a target CAC. iHarness recommends where spend and attention should go next, which cohorts to lean into, which to suppress, and when to stop. It grades on audience outcomes in our warehouse, never on what Meta or Google report about themselves. Execution happens in the ad platforms, by the media team.
• Handoff. Watches every lead that came from paid as it moves through sales. Writes the audience story into the CRM record so a rep knows what someone responded to before they call. Tells me which segments close and which only click.
• Measurement. Runs holdouts and geo tests, keeps the experiment calendar, and produces the weekly incrementality readout.

A Tuesday

6:45, on my phone. Pulse. This is where iHarness brings a human in: anything it has been asked to watch that now needs a decision, with the evidence and a recommended move. I did not open anything. It is there when I pick up the phone, like a text from someone who worked the night shift. Four items this morning: the CTV geo holdout reached significance overnight, a cohort whose CPA doubled on Meta (evidence: frequency over 6, response rate falling for three days), a LinkedIn micro-campaign segment that started converting to demo requests, and a buying signal at an account sales has not seen yet. Each item comes with the evidence and a recommended move. I approve two, say no to one, and ask a question about the fourth.

The routine behind it: every morning at 6:30, review what changed since the last run across all active loops. Report only what needs a human: audiences heating up, performance drifting past the thresholds I set, buying signals from named accounts, experiments that reached significance. For each item, give the insight, the evidence, and one recommended move, short enough to read on a phone. Never take a budget action above $5,000 without approval. If nothing needs a human, send nothing.

8:30. Measurement first, because it changes everything after it. The CTV holdout says lift is real but smaller than the platform reported, provisionally about 18% smaller pending finance sign-off. iHarness has already posted the readout to Slack, updated its model of that channel, and drafted the note for finance. I edit the note and send it. Before this, that readout took an analyst two weeks and a fight with the platform rep.

For every active loop, the routine keeps at least one holdout or geo test running against the biggest spend line. When a test reaches significance, it posts the readout to #growth with lift, confidence, and the gap versus platform-reported numbers, updates the channel model, and drafts a one-paragraph note for finance in plain English. It never reports platform-attributed conversions as outcomes.

10:00. Audience refresh. iHarness rebuilt three cohorts from last night's signals, matched the identities, and handed them to Meta, LinkedIn, and the DSP. One of them was not my request. Our demand gen lead asked iHarness at 11pm for a cohort of accounts that hit the pricing page twice this month, and it was built, matched, and handed off before either of us logged in. Same Growth Brain, same audience memory, no handoff meeting. Suppression ran first. I check the one cohort it was not confident about and drop two accounts by hand.

Nightly, the routine rebuilds cohorts from new signals against each loop's goal, resolves people to platform identities inside the data cloud, and runs suppression before any push: current customers, open opportunities, anyone who unsubscribed or opted out, anyone touched by sales in the last 14 days. It logs match rates per platform, and holds a cohort to ask if confidence falls below the threshold set for it.

11:30. Audience read for the content team. iHarness does not write ads, and we do not want it to. It tells the content team who the cohort is and what they have been responding to. Taste stays with humans. The fastest way to burn a good audience is to point AI slop at it.

1:00. Sales handoff. This is the channel nobody measures. Marketing loses the plot at the lead. The CRM has no idea what produced it. iHarness closes that gap. Today it flagged that leads from one segment book demos at twice the rate but close at half. I take that to the pipeline meeting instead of a lead volume chart.

3:00. Spend decision, in iHarness, not a dashboard. With this morning's CTV readout in the model, I ask what it wants to do with the fatigued Meta cohort. iHarness recommends moving 30% of that spend to a cohort built from last month's converters and holding the CTV line flat until the next test. I approve. The media team makes the change in Meta. iHarness logs the decision, who approved it, and what evidence it was made on, so measurement can check it later.

For each active loop, the routine compares platform-reported conversions to warehouse-verified outcomes and flags any gap over 20%. It spots audiences drifting off target and recommends reallocation with the evidence, but it never moves budget between goals without asking.

5:00. Weekly learnings. The pipeline loop learned this week that accounts who engage with a comparison page convert on a shorter cycle. The retention loop picked that up and now flags existing customers who hit the same page as expansion candidates. Nobody wrote a ticket for that. I skim the log. Most weeks I add one rule, usually because it got something wrong. This week it flagged a partner-sourced cohort as fatigued when the dip was a long weekend. New rule: check for seasonality before calling frequency the cause.

The same day looks different for a D2C team, and the loops are the same. One D2C clothing brand on the platform runs a repeat-purchase loop: household-level cohorts in Meta, suppression of anyone who bought in the last 30 days, a holdout on every retention push, and a Pulse alert when a high-value cohort goes quiet.

What I'd tell someone setting this up

• Onboard it like a new hire. The first time we do a task together, walk through it and have iHarness turn the walkthrough into a routine.
• Write rules every time you correct it. My rule file for spend decisions is now about 40 lines. It has not repeated a corrected mistake.
• Set the approval gates early. Budget moves, audience pushes to a new platform, anything touching a named account. Loosen later, not first.
• One loop per goal. Resist the urge to build one per channel. The channel view is how every platform ends up grading its own homework. We run four loops today (pipeline, expansion, retention, brand). Most teams start with one and add the second in month two.

If you are deploying it for a team, our FDEs stand it up inside your data cloud in weeks. Bring your existing tools. Nothing gets ripped out.

What it has done for us so far, provisional numbers pending a full month's log: roughly 12 hours a week back for me and the demand gen lead, mostly the evening reconciliation work. About 20 budget decisions a month that used to wait for the weekly meeting now happen the morning the signal shows up. And one CTV line we would have kept funding on platform numbers alone. Replace these with your own numbers after month one. They will be different, and they will be yours.

Count it up and iHarness initiated most of today. The holdout readout, the cohort rebuild, the buying signal at a named account, the expansion flag from the retention loop. I initiated two things: the spend question at 3:00 and one rule at 5:00. That ratio is the product. If you find yourself prompting it all day, something is set up wrong.

I still spend time in the platforms. Less of it. The campaigns and the creative are still ours to make, in LinkedIn for example. What changed is that the audience knowledge, the decisions, and the proof of what worked now live in the Growth Brain, where the whole team can use them, and it gets sharper every week instead of resetting every campaign.

Your competitors can rent the same AI. They cannot rent what you have learned about your audience.

FAQ

Does iHarness run ad campaigns directly? No. iHarness decides where spend and attention should go and hands matched audiences to the platforms already in use. The media team executes the change inside the ad platform itself.

How does iHarness measure results differently from the ad platforms? It grades performance on warehouse-verified outcomes rather than platform-reported conversions, and runs holdouts and geo tests to check the gap between what a platform claims and what actually happened.

What decisions still require a human? Anything above the budget threshold set for a loop, any audience push to a new platform, and anything touching a named account. iHarness flags these and waits for approval rather than acting on its own.