The Situation
CRM was generating significant traffic, but more traffic was not translating proportionally into more revenue.
The traditional answer would have been to increase campaign frequency, expand audiences and keep pushing more customers through the channel. I saw the opposite opportunity.
The issue wasn't that we weren't communicating enough. It was that too many customers were receiving the same campaigns regardless of their value, engagement, product interests or likelihood to buy.
I wanted to move CRM away from a volume-led model and answer a different question.
How do we generate more revenue from the deployments we already have?
What I Found
The database was being treated too much like one audience.
A highly engaged repeat customer, someone who had browsed yesterday, a customer who hadn't interacted for six months and a high-value customer interested in a completely different product category could all receive the same campaign.
That created two problems. We were using valuable CRM inventory on customers with limited intent, while customers showing strong buying signals weren't necessarily receiving more relevant treatment.
It also meant traditional campaign metrics could be misleading. A campaign could generate a large amount of traffic while converting inefficiently simply because the audience was too broad.
The opportunity was to progressively add intelligence to audience selection rather than to add more sends.
Adding intelligence to audience selection
Starting point Broad database send — everyone sees the same campaign
Engagement segmentation
Should we be contacting this person at all?
RFM and CLTV
How much is this relationship worth, now and in future?
Product affinity
What is this customer actually interested in?
Behavioural intent and propensity
What do they want right now, and how likely are they to act?
Personalised treatment
What is the single best thing to do for this customer?
Result A customer-level decision — including the decision to send nothing
My Approach
Each layer answered a question the previous one couldn't.
1 Broad audiences to engagement segmentation
The first step wasn't AI. It was stopping every campaign from behaving like a database send.
I separated customers based on meaningful engagement using CRM interaction, onsite activity and purchase recency. Customers moved between four states, with different contact expectations for each.
Highly engaged customers could receive regular commercial activity. Lower-engagement audiences needed stronger reasons to be contacted, while dormant customers were better handled through dedicated reactivation activity.
This immediately turned frequency into a customer-level decision rather than simply a campaign-calendar decision.
2 Engagement to RFM and CLTV
I then added Recency, Frequency and Monetary value to understand the commercial relationship behind that engagement.
RFM told us where someone was today. CLTV added another dimension: how much future customer value might be worth protecting or investing in.
That created very different treatments. A recent, frequent, high-value customer didn't need constant discounts. A previously valuable customer beginning to lapse deserved more attention than a historically low-value inactive customer. A new customer with strong predicted lifetime value could justify more investment despite having limited purchase history.
Instead of treating every reachable customer equally, CRM investment started reflecting customer economics.
3 RFM to product affinity
Value alone didn't tell us what someone wanted.
I therefore introduced product and category affinity using purchase history, browsing behaviour, CRM clicks and product interactions. Rather than assigning customers one permanent label, the aim was to build affinity scores around categories, styles, price points and products.
A customer could simultaneously be high value, highly engaged and strongly associated with a particular category. That meant a single commercial campaign no longer needed to show everyone the same merchandising.
The campaign stayed the same. The customer's version of it changed.
4 Affinity to behavioural intent and propensity
Affinity showed long-term preference. Behavioural signals showed what mattered right now.
Recent searches, repeated product views, wishlist additions, basket activity, session frequency and engagement acceleration were layered onto historical value and affinity data. AI-driven propensity scoring could then estimate outcomes.
How likely is this customer to purchase within the next seven days?
The model could consider transaction history, RFM, CLTV, category affinity, browsing behaviour, discount dependency, previous CRM response and current contact pressure together rather than relying on one signal.
Importantly, high propensity didn't automatically mean "send more". Someone already extremely likely to purchase may need very little intervention. The bigger opportunity may be a medium-propensity customer whose behaviour suggests the right message could materially increase their likelihood of converting.
5 Propensity to personalised treatment
The final step was moving from segmentation towards customer-level decisioning. Instead of asking whether someone belonged in a campaign, CRM could determine the most appropriate treatment.
- Which product?
- Which creative?
- Which offer?
- Which channel?
- Which moment?
- Or should we send nothing at all?
A high-value customer showing strong new-collection intent might receive personalised newness without a discount. A customer with strong category affinity and increasing browsing behaviour could receive products from that category. A discount-sensitive customer with positive intent could receive an incentive. A low-engagement customer with weak intent could simply be suppressed.
Not sending became an optimisation decision rather than a missed opportunity.
One campaign, many treatments
One campaign New season launch
What Changed
Broadcast CRM stopped meaning broadcast audience.
Commercial teams could still run the same launches, newness campaigns and trading messages, but CRM decided which customers were worth prioritising and what version of the proposition they should see.
Core engaged, value and recent-browser audiences became the engine of regular BAU activity, while lower-engagement groups were contacted more selectively.
Within the wider CRM strategy, this approach contributed to significant growth in CRM's commercial importance.
More importantly, growth no longer depended simply on adding another campaign to the calendar.
Growth without more volume
Commercial impact
↑120%
Revenue uplift
Year on year
↑23%
Average conversion
Across CRM campaigns
↑15%
Repeat purchase rate
Customers ordering again
CRM efficiency
↓40%
Deployment volume
Fewer sends — lower is better
↓13%
Unsubscribes
Lower is better