Every CRM platform has gone all in on AI in the last 18 months. HubSpot has Breeze. Salesforce has Agentforce. Zoho has Zia. Pipedrive has its own AI assistant. The marketing on all of them sounds roughly the same: smart automation, predictive insight, agentic everything.
The reality, when you actually use these features in an SME, is far more uneven.
Some are quietly brilliant. Some are demo-friendly and useless. And some actively make your team's job harder while charging you a premium for the privilege.
Here's what I've found after rolling these out and ripping them out across multiple SME accounts.
After running these tools across multiple client accounts, here's where I've consistently seen real value.
Data enrichment and deduplication. AI is genuinely better than humans at spotting that "Sara Smith" and "Sarah Smith" at the same company are the same person, especially when you scale to thousands of records. This alone can save a CRM admin a day a week. When I was setting up the unified CRM at Goodwood, this kind of pattern matching was the difference between a clean database and a mess of half-duplicates that nobody trusted.
Email and meeting summaries. Auto-generated notes that capture what was actually agreed in a sales call, populated back into the deal record. Useful, time-saving, low risk. The salesperson reviews it, edits anything wrong, and saves twenty minutes per call. Across a team of three, that's a couple of working days a week reclaimed.
Forecasting and pipeline scoring. Not magic, but the pattern recognition is good enough to flag deals that look stuck or contacts who've gone cold. A useful prompt for sales managers in the Monday review. Not a replacement for judgement, but a sensible filter.
Content drafting for sequences. Not the final copy. But a starting point that saves 20 minutes per email is genuinely valuable. The trick is treating it as a first draft, not a final output. Marketers who use it that way get value. Marketers who use it to skip the writing process entirely get the bland, generic AI prose everyone has learned to ignore.
Auto-generated reports nobody reads. "Here's a summary of your week" emails that get ignored after the first three. The format never quite matches the question the user has, and the insights are generic enough to be useless. These are the AI features the platform demos love but the customer ignores within a month.
Chatbots pretending to be salespeople. Customers can spot these in seconds, and the conversion data shows it. There are good use cases for chatbots (FAQs, simple routing, account self-service). Pretending to be a human salesperson is not one of them. It damages trust before the conversation has even started.
Predictive lead scoring with no historical data. If you've got 200 leads a year, the model has nothing to learn from. The "AI score" is essentially random. SaaS platforms will happily turn it on for you anyway, because it makes the demo look impressive.
Generative AI writing your customer comms unsupervised. I've seen this go wrong in expensive ways. AI doesn't know your tone, your context, or what you promised the customer last week. The horror stories I've heard from my network involve AI-drafted apologies for the wrong issue, AI-generated price quotes that bear no relationship to the real pricing, and AI customer service responses that quietly accept liability the business never agreed to.
Most SMEs are buying AI features without first asking the boring question: what slow, repetitive task in our business actually deserves automation?
The answer is usually not "write our marketing emails". It's usually something like "match incoming form submissions to existing contacts" or "flag deals that haven't moved in 14 days" or "summarise this week's customer support tickets into a single readable digest". Less glamorous. Far more useful.
If you start with the task and work backwards to the tool, AI can be transformative. If you start with the tool and look for tasks to apply it to, you end up with a list of features running in the background that everyone has quietly switched off.
Before you turn on any AI feature in your CRM, run it through these four questions:
If you can't answer all four, leave the feature switched off. The AI features that survive contact with a real business are the ones you can defend with these answers. The ones that fail tend to be the ones where the answers are vague.
AI in your CRM should be invisible. If you can feel it, it's probably annoying you.
The good implementations free up time you didn't know you were losing. The bad ones add another layer of complexity to a system that was already too complex.
Before you turn on the next AI feature your CRM offers, ask: what would I stop doing if this worked? If you can't answer that, leave it switched off.
What AI features have actually made a difference in your business? And which ones are you quietly ignoring?