· 6 min read

Retention Consultant vs AI: Who Decides, Who Writes

AI and a CRM can write messages and run journeys. A retention consultant coordinates your teams, decides who gets what and proves it worked. Here's the split.

Retention consultant vs AI: AI writes the messages, a consultant decides who gets them, coordinates the teams and proves the result against a holdout.

Updated October 2026.

AI and your CRM can write retention messages, suggest segments and run journeys. What they can't do is get your teams to share the data that decides the next move, judge which customer gets which message on which day, or prove the result against a holdout. That's the work of a retention consultant. Use AI to do the work faster, and keep a person responsible for the decisions.

I'm Raviraj Ramavarapu, founder of Compound Lab, a retention agency for D2C brands and consumer apps. I use AI every day in client work, for drafts, variants and personalisation rules. I don't use it to decide anything. AI and a CRM are tools, and someone still has to know how to run them.

Can you run retention marketing with just AI and a CRM?

Yes, for execution. AI can draft a WhatsApp message, suggest a segment or write a Jinja rule in seconds, and a CRM will send whatever you build. If your journeys are simple and someone on your team owns the results, that can be enough for a while.

The gap shows up in three places. Nobody checks whether the journey is built on the right data. Nobody questions the first answer AI gives. And nobody is accountable when the numbers move, up or down, for the wrong reason.

What can a retention consultant do that AI can't?

Four things, and none of them is writing copy.

  1. Coordinate your teams. The data that decides a retention plan sits with different people. Orders sit with ops, ad creatives with performance, returns with support and the catalogue with merchandising. AI can't sit in those meetings and get it all into one place.
  2. Read the data before writing a message. The gap between first and second order, which first products lead somewhere, which cohorts are slipping. I cover the numbers I start with in retention metrics.
  3. Own the result. Every journey carries a holdout, and decisions are taken on orders against it, not on opens. I explain holdouts in global control groups.
  4. Catch decay early. A person reading campaigns every week spots a message going stale in the week it happens, not at month end.

Why isn't AI's first answer good enough?

Because it's the most common answer, not the one your data supports. AI learns from what people like me have already written and repurposes it. Unless you push back, give it new inputs and test what it says, you get the average.

Ask AI how to cross-sell to clothing buyers and it will suggest colour-contrast matching: a shirt that contrasts with the trousers they bought. That's a reasonable idea. But your own orders might say customers don't want contrast at all. They want the same t-shirt in another colour. You only find that out by pulling the order data, which means working with the people who hold it. I wrote about that exact pattern in fashion cross-sell.

The first answer is rarely the best one. A consultant spots what's missing in it and fixes that part. AI can't do that until someone feeds it the next set of inputs, and knowing what those inputs should be is the job.

What does weekly campaign tracking catch that a dashboard doesn't?

The week a message stops working. Most teams, and most AI summaries, read campaigns monthly on open rate and revenue. That's too late and too blended.

Here's real WhatsApp data from a D2C home brand I ran, over 12 weeks earlier this year. I've withheld the brand and the volumes and kept only the rates.

What the report showed Monthly view Weekly view
July click rate 1.7%, which read as a soft month Fell from 6.2% to 0.7% in a single week
Read rate Blended into the month Fell from 48% to 33% over two weeks
When it's found After the month closes The week it happens

Caught in week one, that's a fix to frequency, audience or copy. Caught on the first of the next month, it's a lost month.

The same brand ran 38 subject-line tests on email. 22 found a clear winner on opens, but only 5 found one on clicks. Most subject lines that win the open don't win the click. A tool that picks winners on opens, which is the default in many CRMs, will roll out the wrong one most of the time. Read tests on orders.

Monthly versus weekly campaign tracking at one D2C home brand. The monthly WhatsApp report showed a 1.7% click rate for July. The weekly view showed the click rate falling from 6.2% to 0.7% in one week, and the read rate falling from 48% to 33% over two weeks. Of 38 email subject-line tests, 22 had a winner on opens and only 5 on clicks.

What do tools miss when nobody coordinates them?

The connections between them. Most D2C brands now run a store platform, a checkout, a CRM, a WhatsApp provider, an app and ad accounts. Each works. The gaps are where they meet.

When I tested a fashion brand's cart recovery recently, six messages reached me across two abandoned carts. Only one of them knew what was in the cart. Each tool was sending what it had been set up to send. The cart details just weren't reaching most of the messages, and nobody had noticed. That's a coordination problem, and AI won't find it from inside a single tool.

Where should you use AI in retention work?

For speed, not for judgement. This is how I split it.

AI does it A person decides
First drafts of messages and subject lines Which customer gets which message, and when
Variants for A/B tests What the test is trying to learn, and how it's read
Jinja and personalisation rules Which data is clean enough to personalise on
Summaries of campaign exports Whether a drop is a blip or a trend
Ideas for reasons to come back Which reasons fit the brand and its customers

I write more about the personalisation side in Liquid and Jinja personalisation.

When don't you need a retention consultant?

When your CRM has someone who reads the numbers weekly and acts on them, your events are clean, and your journeys already carry holdouts. Then AI and the CRM are enough, and a consultant would mostly add cost.

You also don't need one before the basics exist. If you have no CRM yet, or your order and event data isn't reaching it, start there. My guide to who should run retention compares an in-house hire, an agency and the vendor's own team.

What should a retention consultant deliver in the first 90 days?

A plan built on your data, journeys live, and proof that they worked. This is the blueprint I take into a new D2C brand.

  1. One number. The median days from first to second order, and the repeat rate it drives.
  2. One grid. Months since first order against orders placed. Every customer sits in one cell.
  3. Data per cohort. Only what changes the next message, starting with naming the visitor before the first order.
  4. Two tracks. Push the customers who came back, and nurture the ones who didn't with reasons that aren't a discount.
  5. The brand's own reasons. About ten, taken from the founder's words and the ads that already work.
  6. A personal destination. One page per customer that changes as they buy, and every channel points to it.
  7. Proof. A holdout in every journey, six 15-day sprints with a review after each, and campaigns read weekly.

The first two weeks are an audit: which events are tracked, which tools are running and how they talk to each other. No journey works on events that never arrive. The month-by-month messaging that follows is in cohort analysis.

Where I fit

Compound Lab works with D2C brands and consumer apps on everything after the first order: the audit, the plan, the journeys and the weekly reading that proves them. AI is part of how we work, never the thing we sell. If you're deciding between a tool and a person, book a 30-minute call and we'll look at what your current setup is missing.

Next postFashion Cross-Sell: Sell Colours Before Categories

Your customers already paid once. Make them pay again.

Book a Consultation