Resolving contacts to companies when the badge says a domain you never saw
Resolving contacts to companies uses three signals in order: the email domain, the employer name the registrant typed, and the company address on the record. The domain resolves the largest share and fails on consultants, agency staff and contractors, whose domain belongs to a firm other than the account they are attending for.
An exhibitor sales director asks for every attendee from one target account before the show, so the team can arrange meetings. The list comes back with eleven names. She knows of at least three more people from that business who attend every year, and none of them are on it.
Two of the three registered with an email at a consultancy that runs the account's procurement programme. The third used a personal address and typed the employer name with an ampersand where the account record has the word and. Resolving contacts to companies had worked exactly as configured, and the configuration was matching on email domain.
Two entity types, and only one of them is a person
A registration record carries evidence about two different things: a human being and an organisation. Getoor and Machanavajjhala's 2012 tutorial at the VLDB conference categorises entity resolution by the type of input, separating the single entity case, where all mentions are of one entity type, from the multi entity case with potentially linked mentions of different types. Contact to company assignment is the second kind. You are resolving people, resolving organisations, and asserting a link between the two, and each of those three has its own error rate.
The reason to keep them apart in the data model is that they fail differently and change at different speeds. A person is stable and their employer is not. A company account outlives every contact on it. The separation itself is company identity versus person identity in J40, and what matters here is the link between them.
Talburt and Zhou's 2015 book on the entity information life cycle, published by Morgan Kaufmann, frames identity as something managed over time rather than computed once, and the framing earns its keep here. The link from a contact to an account has a start date, a source, and a reason, and treating it as a fact you recompute silently on every load is what produces a CRM where nobody trusts the account rollups.
Three signals, applied in order
The cascade that works is short, and the discipline is recording which step made each decision.
The email domain. Split the address, drop the local part, and match the domain against domains already attached to accounts. This is exact, cheap and covers most of a business audience.
The typed employer name. Normalise it and compare it against account names. This picks up the rows where the domain was useless, and it inherits every problem in company name matching.
The company address on the record. Postcode and building number against account addresses. Low coverage, useful at the margin, and prone to putting people on the account of whoever occupies the same serviced office.
Below that is the review queue, and nothing else should be automatic.
How much of the file does the cascade actually resolve?
Numbers make the shape clear. Take a show with 8,400 registered contacts.
Of those, 5,120 carry a corporate domain that matches an account already in the system, which is 61 per cent of the file. Another 1,880, or 22 per cent, use a consumer mailbox provider, so the domain tells you nothing about their employer. A further 940 carry a corporate domain nobody has seen before, and 460 have no usable email at all.
Run the second signal over the 1,880 free mailbox rows and suppose 1,240 type an employer name that normalises onto an existing account. That leaves 640. Run the address check and 210 more resolve. What is left is 430 contacts, about 5 per cent of the file, that no rule assigns, and they belong in a queue rather than in a default bucket called Other.
The 940 with an unknown corporate domain deserve separate handling. Each one is either a new company account or an existing account with a domain you have never recorded, and the second case is common with recent acquisitions and rebrands. Cluster them by domain, and a domain carrying nine contacts is worth two minutes of somebody's time, because resolving it moves nine people at once.
When does the email domain send you to the wrong account?
The rule's failures are quiet, which is why they persist for years.
Take the sales director's missing attendees. Both work for a procurement consultancy engaged by the target account, and they attend on that account's behalf, sit in its meetings and influence its buying. Their email domain is the consultancy's. Every automatic rule available will assign them to the consultancy, correctly by the letter and wrongly for every commercial purpose the assignment exists to serve.
The same shape recurs across an exhibition audience. Agency staff attending for a client brand. Contractors on a long engagement using the client's domain, which puts them on the client account when their own firm is also a customer of yours. Franchisees whose domain is the franchisor's. Staff of a shared services company that runs the back office for a group of operating businesses.
Measure it before you decide how much it matters. Take 200 contacts assigned by domain, look at the employer each one typed, and count the disagreements. If 9 of the 200 typed an employer that is a different organisation from the domain owner, the domain rule is misassigning about 4.5 per cent of the contacts it resolves, which across 5,120 rows is roughly 230 people sitting on an account they do not buy for.
My view is that the disagreement is the signal. When the domain and the typed employer point at two different known accounts, that row deserves a person rather than a default. Those pairs are where consultancy relationships, secondments and recent job changes become visible, and a queue of 230 of them is a fortnight of pleasant work for somebody in audience operations, with a direct commercial payoff at the end of it.
What to store on the link itself
A contact to account link written as a single foreign key throws away everything you need to audit it later. Five small columns beside it change that.
Which signal decided, from the cascade above. The value that matched, so a reviewer can see that the decision rested on the domain acmegroup.com and not on a name comparison. The date the link was asserted. Whether a person confirmed it. And the identifier of the rule version that produced it, so a change to the cascade can be replayed across the file.
The payoff comes the first time somebody disputes an account rollup. With those columns, the answer to why this person is on this account takes one query, and the fix is a targeted correction. Without them the answer is a shrug, and the usual response to a shrug is a manual override that nobody records, which is how a CRM acquires a layer of corrections that the next reload silently destroys.
There is a second payoff at review time. Links asserted by the weakest signal, the address check, can be listed and re-examined as a group when a serviced office turns up in the file, and links confirmed by a human can be protected from any automatic rule that would otherwise overwrite them.
Free mailbox domains, and the block list you need first
Before any of this runs, one list has to exist: the consumer mailbox providers whose domains must never become a company key.
Skip it and the effect is spectacular. Every gmail.com registrant is assigned to a single account, that account becomes the largest customer in the portfolio by attendee count, and it appears in board reporting as a genuine relationship. The same applies to the role addresses on a real corporate domain, since info@ and events@ belong to a function.
The list needs maintaining, which means somebody adds to it when a new provider appears in the file. A cheap trigger is a frequency check on every domain in a load, with any domain appearing above a threshold flagged for a human to classify as corporate or consumer once, permanently.
Where this stops
Three limits are worth stating, and none of them are solved by better matching.
A contact can genuinely belong to two organisations at once. A consultant on a two year engagement is a real employee of the consultancy and a real member of the client's buying team. Forcing a single account link is a modelling decision that throws away one of two true facts, and the alternative is a link table with a role on it, which costs more to build and more to query.
The link is also a point in time observation. Somebody who registered in 2024 with a corporate domain may have left in 2025, and their contact record still carries the old employer. A registration in the current cycle is evidence about now, an old one is evidence about then, and an account rollup that treats all five years of history as current will overstate every account with staff turnover. Date the link and count only what was true in the period you are reporting on.
The third limit is that the account itself moves. Companies are acquired, merged and renamed, and a contact correctly assigned in 2023 may need to move to a different account in 2026 without anything about the person changing. That is parent and subsidiary matching in J19, and it is the reason the link should point at a durable account identifier rather than at a name or a domain.
Take one show's contact file this week and produce four counts: contacts resolved by domain, by typed employer, by address, and not at all. Then sample 50 of the domain resolved rows and check the typed employer against the domain owner by eye. Two hours of that will tell you whether your account rollups are trustworthy, and it is a better use of the time than any unified data design meeting held without those numbers on the table. Where the two signals disagree, you also have a ready made set of conflicting values, and settling those is conflicting values across sources in J29.
Questions people ask about resolving contacts to companies
- How do you link a registration contact to a company account?
- Apply the signals in order of reliability and record which one decided. A corporate email domain matched against known account domains resolves the largest share. The typed employer name, normalised, catches most of the free mailbox rows. A company address on the record catches a few more. Everything left goes to review.
- Why does email domain matching put contacts on the wrong company account?
- Because a domain identifies who pays for someone's mailbox, which is not always who they are attending for. Consultants, agency staff, contractors and staff of a shared services company all carry domains belonging to a firm other than the buying organisation, and the domain rule assigns them confidently to the wrong account.
- What do you do with contacts using a free email provider?
- Fall through to the typed employer name. Keep a block list of consumer mailbox providers so that gmail.com or outlook.com never becomes a company key, since matching on those merges thousands of unrelated people into one account. In the file worked through here they are 22 per cent of it, and the share is worth measuring on your own audience.
Related reading
- Parent and subsidiary matching when one exhibitor books under four legal names
- Conflicting values across sources and how to pick one without guessing