Clean Inbound Routing With Claygent, HubSpot Dedupe, and Required Fields

Clean Inbound Routing With Claygent, HubSpot Dedupe, and Required Fields

The first time I wired Clay into a HubSpot inbound queue in 2026, the enrichment worked almost too well. A demo request came in from “A. Patel” at a Gmail address, Clay found the company from the website field, Claygent read the site, and HubSpot had a neatly filled company record 90 seconds later. Then the record routed to the wrong SDR because there were already two Acme Analytics companies in the portal, one from a 2024 webinar upload and one from a 2025 paid search form.

That is the trap. Clay’s newer workflow stack can fill gaps fast. HubSpot can assign leads fast. Speed is not the scarce thing once the plumbing is in place. The scarce thing is a clean handoff that does not dump half-guessed records into someone else’s morning.

Clay’s own 2026 HubSpot sync guide frames the loop correctly: import HubSpot records as a live source, enrich and score them in Clay, then write back to the same HubSpot records without creating duplicates. Clay’s workflow docs also show why this has gotten easier in 2026. Workflows can trigger per record, branch on rules, run Claygent for judgment work, and call HubSpot write actions from the same graph. HubSpot added its own moving parts too. As of the July 23, 2026 HubSpot duplicate-management docs, the duplicates tool scans contact and company records daily and can use custom duplicate rules on Data Hub Professional or Enterprise.

Put those together carefully and you get a useful inbound gate. Put them together casually and you get shinier duplicates.

Start With The Record You Already Have

For inbound, I start in HubSpot, not Clay. The first object I care about is the contact created by a HubSpot form, Chili Piper handoff, Webflow form, or Segment event. On most B2B sites I audit, the form has 4 to 7 fields: email, first name, last name, company, website, job title, and one free-text “What can we help with?” box. That is enough to create a lead, but it is not enough to assign a rep.

HubSpot’s dedupe rules matter here because they decide whether that new form fill updates an existing record or creates clutter. HubSpot automatically deduplicates contacts by email address and companies by company domain name, according to its June 22, 2026 deduplication docs. If a contact imports or submits a form with the same email, HubSpot updates the existing contact. If a company is created with the same primary domain, HubSpot can match it to the existing company. If the email is personal and the domain field is blank, you are driving without headlights.

So my first workflow branch is boring on purpose. If email is missing, hold the record. If email is a personal domain like gmail.com, yahoo.com, or icloud.com, require website or company name before enrichment runs. If company domain name exists on the associated company, use that as the anchor. In one 2026 paid search account with 312 monthly demo requests, this gate held 41 records in the first week. Twenty-nine were students, vendors, or obvious junk. Twelve were real buyers who had typed personal emails, and Clay had enough company clues to salvage 9 of them.

That is a good trade. An SDR should not be the fallback parser for a lazy form.

Use Claygent For Judgment, Waterfalls For Fields

I split Clay work into two lanes. Waterfalls fill known fields. Claygent makes calls that need reading and judgment.

Clay’s Waterfalls docs describe the core behavior: a waterfall runs multiple data providers in a chosen sequence and stops when one returns the field you asked for. For inbound lead cleanup, my field waterfall is narrow. Company domain, employee count, industry, LinkedIn company URL, headquarters country, work email verification, seniority, department, and direct phone if the go-to-market motion can use it. I do not enrich 30 fields because a rep will never read 30 fields before the first call.

The Claygent lane answers 3 questions. What does this company sell? Who does it sell to? Does this buyer match the ICP? Clay’s Workflows docs list Run Claygent as the step for work that needs judgment, with Claygent versions that can be pinned as v# or set to Current. I pin versions for routing. A prompt tweak on August 14 should not quietly change how the August 15 inbound queue gets assigned.

A practical Claygent prompt for this workflow is short. “Visit the company website and LinkedIn page. Return JSON with business_model, primary_customer, icp_fit, fit_reason, confidence, and routing_notes. Use icp_fit values high, medium, low, or unknown. If the website is unavailable, return unknown and explain the missing evidence.” That last sentence saves pain. Without it, agentic enrichment has a way of sounding certain while standing on fog.

Write Back Into Holding Fields First

Never let enrichment overwrite the handoff fields directly on the first pass. I use staging properties in HubSpot with names that make their job obvious: clay_company_domain_suggested, clay_employee_count_suggested, clay_icp_fit, clay_fit_confidence, clay_last_enriched_at, and clay_enrichment_status. The production properties stay separate: company domain name, industry, number of employees, lifecycle stage, lead status, and HubSpot owner.

HubSpot’s 2026 object API docs say companies require at least domain or name for creation, while contacts have no hard creation requirement, with email, firstname, or lastname recommended. Leads are stricter in the 2026-03 API. The Lead object requires fields such as hs_associated_contact_email, hs_lead_name, hs_associated_company_domain, contact names, and company name. I treat that as the real bar even when the record came from a form and already exists in the CRM.

My required-property check before SDR assignment is this: contact email, contact last name, company name, company domain, buyer role or seniority, country, consent basis where required, ICP fit, and enrichment timestamp under 30 days old. For enterprise routing, I add employee count and region. For partner-heavy businesses, I add channel conflict status. In a founder-led motion with 2 reps, you can simplify this to email, company domain, ICP fit, and owner region.

The point is that the required check happens after Clay and before owner assignment. If you assign first and clean later, the rep gets the task, the Slack ping, and the bad data. You have already spent their attention.

Let HubSpot Dedupe Interrupt The Route

HubSpot’s duplicates manager compares default contact properties such as first name, last name, email, IP country, phone number, zip code, and company name. For companies, it compares company domain name, company name, country or region, phone number, and industry. The July 2026 docs say Professional and Enterprise portals can review up to 10,000 duplicate pairs in the duplicates manager, while Data Hub Professional can show up to 30,000 and Data Hub Enterprise up to 100,000. Custom duplicate rules need Data Hub Professional or Enterprise, with up to 2 rules per object and up to 9 comparison properties per rule.

For inbound routing, I use 2 custom rules when the subscription supports them. Company rule 1 is domain + normalized company name. Company rule 2 is company name + country + phone. Contact rule 1 is email + company domain. Contact rule 2 is first name + last name + company name + country, mostly for events and list uploads where email quality is messy.

This is where the route pauses. If HubSpot flags a possible duplicate company above an internal threshold, set clay_enrichment_status to possible_duplicate and send the record to an ops queue named Inbound Data Review. Do not rotate it. Do not create a lead object. Do not stamp MQL. The ops task should show the fields needed to decide quickly: source form, email, company domain, suggested domain, existing company owner, last engagement date, lifecycle stage, and Claygent routing notes.

A daily manual merge review sounds old-fashioned until you compare it with bad assignment churn. On a 50,000-contact HubSpot portal I cleaned in May 2026, the inbound queue averaged 73 new hand-raisers per week. The duplicate gate caught 8 to 11 records each week for the first month. After the old webinar imports were merged and the domain rules were tightened, that fell to 2 or 3.

Build The Routing Branches After The Gate

Only after dedupe and required fields pass do I branch into routing. HubSpot workflows can rotate records to owners, and the current workflow action docs describe round-robin style assignment for records and leads, with options like Round Robin, Balanced, and Random for leads when the right Sales Hub or Service Hub tier is in place. Keep Clay out of the final owner decision unless the territory logic genuinely lives outside HubSpot.

My base branches look like this in HubSpot: enterprise North America, enterprise EMEA, commercial North America, commercial EMEA, partner or agency, existing customer, student or vendor, and nurture. The branch criteria use HubSpot properties, not raw Clay outputs. That means number of employees must be accepted from staging, country must be normalized, and clay_icp_fit must clear the confidence bar before it touches HubSpot owner.

For example, a contact from ramp.com with a director-level RevOps title, 1,000-plus employees, United States country, and high ICP fit can go to the enterprise North America rotation. A 12-person agency in Dublin with medium fit goes to partner review. A Gmail lead with no domain and unknown fit goes to nurture or manual review, even if the message says “urgent demo.” Urgency in a text box is not qualification.

I also keep a bypass for named accounts. If the company domain matches a target-account segment owned by a specific AE, HubSpot stamps that owner and skips round robin. The list needs maintenance. In one Series B SaaS portal, 187 named accounts lived in a static HubSpot list called FY26 Target Accounts - Owned, and sales ops reviewed it every Monday before 10 a.m. Pacific. That tiny ritual prevented the classic fight where paid search creates a “new” inbound lead from an account sales has worked for 6 months.

Add The Failure States Reps Can Trust

The ugliest workflows are the ones with only 2 states: routed or not routed. I prefer 5 enrichment states because support and sales can read them without asking ops to decode a workflow log.

ready_to_route means dedupe passed, required properties passed, and the record has a valid branch. possible_duplicate means HubSpot found a contact or company conflict. missing_required_fields means Clay could not fill the minimum fields. low_fit means the record is real but should not hit SDR capacity. enrichment_error means a tool failed, timed out, or returned unusable data.

These states should live on the contact and the associated company. HubSpot associations get messy when teams only stamp the contact. A second person from the same company can arrive 3 days later, and your workflow needs to know whether the account already failed the company gate.

Clay’s workflow docs mention that node outputs are referenced from upstream steps and that branches can split and converge. Use that. Send every failure branch into a shared HubSpot update step that stamps status, timestamp, and failure reason. Then send only ready_to_route into owner rotation. It feels mundane when you build it. It feels luxurious when a founder asks why 17 demo requests did not reach sales and you can answer from one HubSpot view.

Test With A Mean Little Batch

Do not test this with 10 perfect Salesforce contacts. Pull 50 messy inbound records from the last 90 days. Include 10 personal emails, 10 duplicate-looking companies, 10 missing-title records, 10 target accounts, and 10 obvious junk submissions. That batch will tell you whether the workflow can survive Tuesday.

I use 4 pass metrics before turning on live routing. At least 90% of records with work emails should preserve or find the right company domain. Zero known duplicates should route automatically. At least 95% of routed records should have owner, lifecycle stage, lead status, company domain, employee band, and ICP fit populated. Manual review should stay under 15% of inbound volume after the first cleanup week.

Those numbers are achievable for paid and organic programs with clean forms and 500 to 5,000 monthly inbound contacts. They are harder with event scans, co-marketing lists, and freemium signups. Fine. Split those sources into separate intake paths. A booth badge from SaaStr Annual and a pricing-page demo request should not share the same trust level.

The Setup I Would Ship

If I were building this today, on August 26, 2026, I would keep the stack plain. HubSpot form creates or updates the contact. Clay imports the record from HubSpot or receives a webhook. Clay waterfalls fill company and buyer fields. Claygent reads the company site and returns structured ICP judgment. Clay writes only staging fields back. HubSpot checks duplicate rules and required properties. HubSpot rotates only the records that pass.

The source docs are worth bookmarking while you build: Clay’s HubSpot two-way sync guide, Clay’s Waterfalls docs, Clay’s Workflows docs, HubSpot’s duplicate records guide, HubSpot’s deduplication docs, and HubSpot’s workflow actions reference.

Claygent can make inbound routing smarter in 2026, but the quality comes from the gates around it. Dedupe first. Required fields next. Routing last. That order keeps reps out of data cleanup and gives founders a cleaner read on which channels are bringing real buyers into the room.