AI-Assisted Contact Enrichment & Account Matching

A recurring challenge has been losing more useful CRM contacts than we gain. People change jobs and addresses go stale, while new contacts arriving through web forms, webinars, email and social media often come with a name, an email address, maybe an organization and a title, and not much else. Before a new contact is useful to Sales or Marketing, someone has to work out which organization it belongs to, whether that organization is already a customer, which account record applies and who owns the relationship.

The company operates on a custom-built internal CRM developed and maintained over many years. Because it predates modern CRM ecosystems, standard third-party integrations are limited. There was no off-the-shelf integration to lean on, so the workflow had to be designed around the systems we actually had: the CRM, Klaviyo and Claude.

I designed a Claude-assisted matching workflow. Claude works from periodic exports of CRM contact and account data that I provide, and from a connector that gives it access to relevant Klaviyo information through Klaviyo's API. There is no live connection into the CRM itself.

For each new contact, the workflow uses the information the contact supplied, mainly email address, email domain, organization and job title, and compares it against existing CRM context. One important matching signal is whether existing contacts share the same email domain. If the domain is already associated with existing contacts and an account, confidence in the proposed match goes up.

• Each proposed match receives a 1-to-5 confidence score. High-confidence matches have strong supporting account signals, while lower-confidence matches are flagged for additional research.
• For each contact the output can include the likely organization and account, a likely or possible account number, the market or segment, the sales representative, the account status and a short explanation of why the match was suggested.
• I review the output. Nothing is written to the CRM automatically.
• In many cases the first practical step is to place the contact in the right Klaviyo audience so they receive relevant messaging. If later engagement makes the relationship clearer, the contact is added to the CRM with more confidence.

The workflow is currently triggered manually when contacts need processing. I review the output before anything is added to the CRM.

The conceptual flow is below. Not every contact travels through every step; it shows how the workflow supports the decision rather than a fixed pipeline.

The demonstration below shows the shape of a single result. Every value is fictional.

I designed the workflow end to end, from identifying the business problem and organizing the available CRM context to defining the matching logic, confidence scoring, output and review process. That included working out how CRM exports and Klaviyo data could be used together, writing and refining the Claude workflow, and fitting it into the day-to-day contact-management process. Before this, no process combined those data sources in this way.

• Reduces the manual research that used to happen before a new contact was useful.
• Gets contacts into the right Klaviyo audience sooner, so they receive relevant messaging.
• Gives Sales context on who is engaging inside their accounts.
• Helps Marketing connect new contacts to existing accounts for attribution and segmentation.
• Makes it practical to keep a custom CRM current without a modern integration layer.

The scores are a working judgment, not a validated statistical model. A 5 means the supporting signals are strong. A low score means a person needs to look, and that is what the workflow is for.

This page describes the design of a private workflow at a high level. It does not reproduce prompts, data, schemas or connector details, and it is not connected to any company system. Every value in the demonstration is fictional and uses the reserved .example domain.