Matchmaking.
Buyer and seller matching. Fit scoring, hosted buyer programmes, scheduling under constraints, no-shows and cancellations, connection density, and meeting quality afterwards.
29 articles · Matchmaking
- What a buyer seller fit score should actually containbuyer seller fit scoreA buyer seller fit score needs five bounded sub-scores covering product, industry, intent, authority and behaviour, scored in both directions before weighting.
- How to decide on weighting a match score you can defendweighting a match scoreWeighting a match score is a policy choice, so test candidate weights against held out accepted meetings, report precision at five, and publish the numbers.
- Match score calibration and why raw scores mislead your concierge teammatch score calibrationMatch score calibration checks whether a 0.72 proposal really gets accepted 72 per cent of the time, using decile buckets and an isotonic correction.
- Explaining a match recommendation so the buyer trusts the meeting proposalexplaining a match recommendationExplaining a match recommendation works when the sentence comes from the same weighted contributions that produced the ranking, so it cannot drift from it.
- Match reason codes give your team something to argue withmatch reason codesMatch reason codes replace free text with a short controlled list you can count, so acceptance by lead reason becomes a number your team can argue with.
- Showing sub scores to exhibitors without handing them a cheat sheetshowing sub scores to exhibitorsShowing sub scores to exhibitors invites gaming, so publish coarse bands on a fixed refresh cadence and hold the behaviour term back from the exhibitor view.
- Building an exhibitor product taxonomy that a matching engine can useexhibitor product taxonomyAn exhibitor product taxonomy decides what a matching engine can see, so size the tree from your own floor and count the exhibitors landing in a catch all node.
- Product category mapping from free text exhibitor profilesproduct category mappingProduct category mapping turns free text exhibitor profiles into taxonomy leaves using tf-idf cosine similarity, an auto assign threshold and a review queue.
- Taxonomy granularity for matching and the cost of going too deeptaxonomy granularity for matchingTaxonomy granularity for matching trades precision against pool size, so score the same buyer at three depths and count how many candidates each one leaves.
- Buyer interest capture at registration without adding twelve more questionsbuyer interest captureBuyer interest capture costs registration completions, so test a short picker against a long one and count the completions lost and the categories gained.
- Intent signals for matchmaking and which ones are worth scoringintent signals for matchmakingRank candidate intent signals for matchmaking by acceptance lift against the base rate, then check coverage and the error band before either one earns a weight.
- First party intent at events and where the signal comes fromfirst party intent at eventsFirst party intent at events already sits in registration, agenda and prior edition data. Here is which table each signal comes from and how to turn it into a score.
- Treating meeting requests as intent and what that does to scoresmeeting requests as intentMeeting requests as intent are the loudest signal in the building and the easiest to double count. Cap the contribution, then measure the feedback loop they create.
- Buyer authority scoring when nobody fills in the budget fieldbuyer authority scoringBuyer authority scoring has to run on purchasing role, company size band and hosted buyer status, because the budget question on your registration form is mostly empty.
- Job title seniority parsing for a global exhibitor and visitor filejob title seniority parsingJob title seniority parsing is a data job before it is a modelling job. Build a rules first mapper onto five levels, then measure it against 500 hand labelled titles.
- The buying committee at trade shows rarely arrives on one badgebuying committee at trade showsThe buying committee at trade shows sends one person and keeps four at the office. Score the account beside the badge, and count each account once however many badges it has.
- Cold start matchmaking at a first edition with no behavioural historycold start matchmaking first editionCold start matchmaking at a first edition has no accepted meetings to learn from. Hold intent and behaviour neutral, renormalise the weights, and instrument the show.
- Onboarding questions for matchmaking that people will actually answeronboarding questions for matchmakingOnboarding questions for matchmaking trade signal against drop off. Run the trade as an experiment and judge it on matched buyers per arrival.
- Borrowing priors from a sister event to seed a new onepriors from a sister eventPriors from a sister event let a new show start with category level acceptance rates, blended by a shrinkage factor so local data takes over as it arrives.
- Meeting scheduling as an assignment problem your team can inspectmeeting scheduling assignment problemMeeting scheduling is an assignment problem with side constraints. Write the objective and the constraints down, then pick a solver and make every rejection explainable.
- Why a greedy meeting scheduler is often the right first buildgreedy meeting schedulerA greedy meeting scheduler places the best proposals first and takes what it can get. The arithmetic on 500 proposals, and what the half guarantee really promises.
- Idle time in meeting schedules is the complaint you will hearidle time in meeting schedulesIdle time in meeting schedules drives more buyer complaints than match quality. How to measure internal gaps per participant and the spread between best and worst served.
- Table allocation for meetings when the hall has fewer tables than proposalstable allocation for meetingsTable allocation for meetings decides who gets a meeting once proposals exceed capacity. The arithmetic on 300 proposals, 40 tables and 8 slots, and the quota rule.
- Double booking prevention in a meeting programme with three sources of truthdouble booking preventionDouble booking prevention needs one availability table that every system writes through, because a buyer with three calendars has three systems that each think they own 2pm.
- Availability windows for meetings and the travel gaps nobody declaresavailability windows for meetingsAvailability windows for meetings shrink once flights and transfers are subtracted. Collect arrival and departure at confirmation, then recount the slots a buyer really holds.
- Hard and soft scheduling constraints belong in different parts of the solverhard and soft scheduling constraintsHard and soft scheduling constraints do different jobs, and folding both into one penalty function hides which rule your meeting schedule actually broke.
- Measuring the meeting no show rate without flattering yourselfmeeting no show rateA meeting no show rate needs a denominator everyone agrees on, because confirmed, scheduled and attended are three different counts that give three different answers.
- A meeting cancellation policy that changes behaviour before the show opensmeeting cancellation policyA meeting cancellation policy has to price notice, because a cancellation 48 hours out can be rebooked and one at nine in the morning cannot.
- Waitlist backfill for meetings turns a cancellation into a second chancewaitlist backfill for meetingsWaitlist backfill for meetings recovers the slots cancellations free up. Hold a ranked standby list per slot, and work 40 cancellations at a 55 per cent fill rate.