Skip to content

Matching and the review queue

Matching is how a gift finds the right person. It is the piece of Relae doing the most work and the piece you should have to think about the least.

The same donor arrives from different channels under different names. Margaret Okonkwo at the box office. M. Okonkwo on the giving portal. OKONKWO, MARGARET from the bank. A human reading those three knows it is one person. A system matching text does not, until it is taught.

How a gift becomes a clean recordThree channels send gifts naming the same donor three different ways: a ticketing platform sends Margaret Okonkwo, a giving portal sends M. Okonkwo, and a bank transfer sends OKONKWO, MARGARET. All three enter matching. Gifts the engine can resolve go straight to the donor record. The ones it cannot resolve go to the review queue first, where a person decides, and then on to the same record. The record holds all three gifts as one giving history.Three channelsMargaret OkonkwoTicketing · $250 · 4 MarM. OkonkwoGiving portal · $500 · 18 JunOKONKWO, MARGARETBank transfer · $1,000 · 2 NovMatchingresolvedunsureReview queuea person decidesMargaret Okonkwo3 gifts · $1,750 lifetimefirst 4 Mar · last 2 NovOne recordOne donor. Three spellings. The review queue sees only the third.Every decision you make in review is remembered and applied on the next import.

Every incoming gift lands in one of three states.

Confident. Relae is sure who this is, usually because the email address matches exactly or the name is an unambiguous match. The gift attaches to the record and you never see it.

Worth confirming. A likely match with something not quite right. A shortened first name, a changed surname, a typo. These reach the review queue as a single yes or no.

Unsure. No good candidate, or several equally plausible ones. These reach the queue leading with “create a new record”, because a new record is the safer default and a wrong merge is more expensive to unpick than a duplicate.

The queue holds only what needs a human. If your import produced two hundred gifts and eight questions, you answer eight questions.

Each row shows the incoming gift, what Relae thinks, and why. You confirm the match, choose a different record, or create a new one.

This is the part that compounds. Every decision you commit is stored, and applied automatically the next time.

Confirm once that “M. Okonkwo” from the giving portal is Margaret Okonkwo, and every future gift from that name attaches on its own. Say once that “Kettle Valley Arts Foundation” is an organization and not a person, and it is never asked again.

The queue gets shorter over time, which is the intended behaviour. A shop six months in is usually clearing a handful of items a week.

You can see everything Relae has learned, search it, and remove a decision that turns out to be wrong.

Matching stops the same person being created twice from incoming gifts. Duplicates that already exist, usually inherited from an old spreadsheet or a CRM, are a separate sweep that proposes pairs it believes are the same person.

You decide on each pair, and a merge can be undone for thirty days.

Relae only proposes what the evidence supports. A shared email address is not treated as proof of a shared identity, because couples, families and small offices routinely share one.