I have seen this 30 times in 20 years: you walk into the quarterly review, the CMO projects a beautiful slide where “AI has improved engagement by 40%”, and the CFO, without looking up from their spreadsheet, asks: “So why has direct channel EBITDA fallen 3%?”.
An awkward silence. Someone mutters about “attribution models”.
If you are in that room, I have news for you: you have been sold high-tech smoke. Data Innovation, which manages more than a billion emails a month from Barcelona, has put its finger on the sore spot: in 60% of cases, the AI in your CRM is not failing; it is amplifying your incompetence in data management.
Here is what your Salesforce or Adobe Account Manager is NOT going to tell you: their AI is agnostic to the quality of your business. If you feed it garbage, it will return garbage at supersonic speed.
The real cost of the “black box”
In 2021 I saw a European retailer activate a purchase propensity model. On paper, an investment of €150,000 in licenses and six months of implementation. The AI decided to reactivate “dormant” users. Learn more about Why You Have AI CRM Insights But Zero Revenue Growth.
The real result: * Open rates rose to 40% (artificially). * Deliverability crashed because the AI fired at thousands of spam traps and dead users. * Hidden cost: 4 weeks of crisis consulting to clean up IP reputation and an estimated loss of €200k in lost sales from emails that ended up in Junk during the Christmas campaign.
The AI knows nothing about deliverability. It only looks for patterns. If the pattern is security bots opening emails, the AI will optimize to please the bots, not your P&L.
Business diagnosis: cut the cord or invest more?
Before approving another consulting invoice, do this reality check. If you fail here, turn off the tap on software investment until you fix the pipes.
- The open rate myth: If your dashboard shows Open Rates >40% but your Click-to-Open is <2%, you are celebrating bot traffic (Apple MPP, Barracuda). Your AI believes there is interest; the reality is that nobody is reading.
- Latency kills the sale: If your “abandoned cart” offer arrives 4 hours late because you depend on nightly batch processes instead of real-time APIs, you are throwing money away. The customer already bought on Amazon three and a half hours ago.
- Fantasy attribution: If Marketing claims a sale from an email, but that customer would have bought anyway (organic), you are cannibalizing your margin.
Your action plan: the 3 C Framework
You do not need to change CRM. I have seen Data Innovation fix this for clients like Nestlé without migrating platforms, simply by fixing data ingestion. This is what I would order implemented tomorrow: Learn more about Stop Overpaying for GPT-4: Optimizing AI Model Selection for ROI.
1. CONTEXT (Clean Input) Install a cleaning layer before the AI. Tools like IPQS or in-house validation solutions. * Cost: Marginal compared to your CRM license. * Timeline: 2 weeks. * Instruction: “I want all machine opens filtered out before they enter the predictive model”.
2. CONTROL (Business Rules) The AI suggests, the business decides. If a customer has an LTV (Lifetime Value) of >€500 and buys at full price, forbid the AI from offering them discounts. * What happens now: The AI sees that the discount increases conversion by 0.5%, so it gives it to everyone, destroying your gross margin. * The decision: Set financial “Guardrails”. Conversion is vanity; margin is sanity. Learn more about Fixing Email Domain Reputation Loss.
3. CLOSING (Incrementality) Stop measuring last-click attribution. Demand measurement of incremental lift. * The test: “Turn off the AI for 10% of the audience for two weeks”. If sales in that control group barely drop, your AI tool is a cost, not an investment.
What to do on Monday at 9:00 AM
Stop accepting more “engagement” reports. Do this:
- Week 1 (Forensic Audit): Ask for a list of the 50 “Top Tier” customers the AI has identified this month. Review them manually. I bet you dinner that you will find duplicates, inactive users or bots.
- Week 2 (Ghost Control Group): In the next automated campaign, exclude 10% of the audience from the action recommended by the AI. Measure the difference in net cash revenue, not in the marketing dashboard.
At Data Innovation they are technology agnostic. They do not care whether you use a six-figure Enterprise suite or a stack put together with Mautic and dedicated engines to optimize costs. The premise is the same: Technology without a clean data strategy is just an expensive way to accelerate chaos.
If your numbers do not add up, stop blaming the algorithm and start looking at the foundations. You have a structural problem, and that is fixed with business decisions, not with more software.
→ datainnovation.io/contacto
Let’s talk today https://datainnovation.io/contacto/!
Inspiration: The CrunchFREE DIAGNOSTIC – 15 MINUTES
Does your ESP take more than 25% of what your email marketing generates? Do many of your emails not reach the Inbox? Is your team losing hours on tasks that intelligent automation would solve on its own?
We review your real sending cost, your domain reputation and your automation opportunities – and we tell you exactly where you are losing money and what you can recover with managed infrastructure, proactive deliverability and agentic automations.