Contact data an AI SDR can trust

Step 1List the fields each message needs

List every fact one message rests on. At Example Software, an invented company that sells to finance teams, a first email rests on nine. The agent finds five through enrichment: the person’s name, role and company, a verified work email, and the signal the message is about, such as a job post. The CRM already holds four: is this a customer, is a deal open, who owns the contact, and did they unsubscribe? Note where each comes from, and how old it is.

These four change on your side, when a rep opens a deal or a contact unsubscribes, so the agent reads them from HubSpot as it writes (part 6). HubSpot says that when “Unsubscribed from all email” is true, its email tools, “including prospecting agent, won’t send emails to the contact”. Customers and open deals need a rule of your own (part 8), and the owner decides whose inbox the message comes from (part 9).

Step 2Enrich them, provider by provider

No single provider leads on both coverage and quality, Clay’s guide says, so enrichment tools ask several in turn. Clay’s waterfall “lets you search sequentially across multiple tools until you find a valid match”. For a work email it first guesses an address from the name and domain, for free, then “cascades across more than 100 email providers in sequence and charges credits only for the one that returns the match”. Once one returns a valid result, the run ends.

Each field is its own lookup, with its own credits, so enrich the fields your messages use and no others. Then set the order: “The order of a waterfall is a budget decision, not a coverage one”, Clay’s guide says, because the first provider runs on every contact; put the cheapest one you trust first. And set the match strict: it sends more contacts down the chain and costs more, which Clay calls “the right trade for cold outreach where a wrong address is worse than a missing one”.

Step 3Verify each address before the first message

A found address is a guess until a verification service checks it; Clay’s waterfall uses ZeroBounce by default. Decide in advance what the agent does with each result.

So in Clay, choose. Turn the toggle on, and catch-alls count as invalid. Or keep the default and send the verification result to the CRM with the address (Clay hides its validation columns by default), so the agent can hold a catch-all for a person.

A bounce costs more than one email. Gmail’s sender guidelines say to “Automatically unsubscribe recipients who have multiple bounced messages”. Outlook asks domains that send over 5,000 emails a day to “Remove invalid addresses regularly”, and may filter mail from senders with serious lapses in hygiene. Part 11 covers the providers’ rules in full.

Step 4Check old fields again before writing

People change jobs. In January 2026 the median US wage and salary worker had been with their employer for 4.1 years, the Bureau of Labor Statistics reported in September: half had been there for less. It isn’t a yearly decay rate for your data, and it misses new titles at the same employer, but a role found last year may be wrong today. So store the date each field was found or last verified, and give each field a limit. In HubSpot, that means a custom date property per field: the built-in “Last modified date” changes when any property on the record does. Before writing, the agent checks again any field past its limit: as an example, 90 days for role and company, 30 for the email’s verification, 14 for a signal.

If the check finds a change, such as a new employer, the agent writes nothing from the old record: it updates the record, and the contact starts again at step 1.

Next in this series

Part 5: your ICP and messaging, written for an agent. See all 13 parts.

Why we write these guides