A message we received recently, lightly paraphrased:
"We're a nonprofit looking for support clearing our Salesforce data of duplicates and making sure everything is up to date — basically a wee declutter. We'd also love to know how to use AI with our Salesforce and Pardot systems."
If that sounds like your organisation, you're in good company. Almost every team that's used a CRM for more than two years has the same problem: duplicate contacts, dead records, fields nobody fills in, and a nagging feeling that the data can't be trusted. And now everyone's asking the second question too — "where does AI fit into this?"
Here's how we'd approach both, in plain language.
Why your CRM got messy (it's not your fault)
CRM data decays naturally. People change jobs, emails bounce, someone imports a spreadsheet without checking for duplicates, a form on your website creates a new contact instead of updating an existing one. Industry estimates put B2B contact data decay at 20–30% per year. You didn't do anything wrong — you just didn't have a system fighting the decay.
The mistake most teams make is treating cleanup as a one-off event. You "declutter", it looks great for three months, and then it's a mess again. A proper fix has two parts: the cleanup, and the plumbing that stops it from happening again.
Part 1: The declutter
Here's the order we work in, and the order we'd suggest even if you do it yourself:
1. Decide what "clean" means before touching anything
Which fields actually matter to your team? Which record types do you genuinely use? Half of most cleanups is deleting fields and processes that nobody has used since 2021. Less structure, maintained well, beats lots of structure ignored.
2. Back everything up
Before any merging or deleting, export a full backup. Deduplication done wrong destroys data — merged records can't easily be un-merged.
3. Handle duplicates with rules, not eyeballs
Salesforce has built-in duplicate and matching rules, and tools like Cloudingo or DemandTools can bulk-merge based on criteria you set (same email, similar name + same organisation, etc.). The important work isn't running the tool — it's deciding the rules: which record wins? What happens to conflicting field values? Who's the surviving record's owner? Get those decisions written down first.
4. Deal with the dead weight
Contacts with hard-bounced emails, no activity in 2+ years, and no donations or open opportunities probably shouldn't be cluttering your reports (or your Pardot email limits — you may be paying for mailable contacts you can't actually mail).
5. Fix it at the source
This is the step everyone skips. If duplicates came from your website forms, fix the forms to update-not-create. If they came from event spreadsheet imports, build an import process with matching rules. Otherwise you'll be doing this again next year.
Part 2: Where AI actually helps with Salesforce and Pardot
Now the fun part. Once your data is trustworthy, AI has genuinely useful applications — and a few overhyped ones. Honest rundown:
Genuinely useful today
- Data enrichment and hygiene. AI tools can flag likely duplicates that rule-based matching misses ("Jon Smith, Acme" vs "Jonathan Smith, Acme Ltd"), standardise messy fields, and fill gaps from public information. This is unglamorous and very effective.
- Drafting donor and supporter communications. AI grounded in your supporter's actual history — past donations, events attended, emails opened — can draft genuinely personalised outreach in Pardot (now "Marketing Cloud Account Engagement", because Salesforce loves renaming things). A human still reviews before sending.
- Summarising relationship history. Before a call with a major donor or partner, an AI assistant can summarise every interaction across emails, notes, and giving history in seconds. For small teams where everyone wears five hats, this is a real time-saver.
- Smarter lead and donor scoring. Pardot's scoring is rule-based ("+10 points for opening an email"). AI-driven scoring looks at patterns across your whole history and often predicts engagement much better.
Approach with caution
- Fully automated AI responses to supporters. Letting AI reply to people without human review is risky for any organisation, and doubly so for nonprofits where trust is the whole product. Draft-for-approval, yes. Auto-send, not yet.
- AI on top of messy data. This is the big one. AI doesn't fix bad data — it amplifies it, confidently. If your records are full of duplicates and stale information, an AI assistant will give you confident answers based on garbage. Clean first, then add AI. That's why the two halves of that inquiry belong together, in that order.
DIY or get help?
Honest answer: if you have under ~5,000 records and someone on the team with a spare week, you can do most of the declutter yourself with Salesforce's native tools and the steps above.
Get help when: you have tens of thousands of records, complex objects (donations, memberships, grants), integrations feeding data in from multiple sources, or you want the "fix it at the source" plumbing and AI layer built properly. That's systems work, and it's what we do.
Either way, if you'd like a second opinion on the state of your CRM — or a realistic take on what AI could do with your setup — get in touch. We're happy to tell you if it's a DIY job.