Measuring engagement that predicts allocations

A wholesaler at a mid-sized manager once bragged that his best lead was an advisor who had downloaded nine whitepapers. The advisor never wrote a ticket. In the same month a different advisor logged into the client portal twice in one week to check a fund's holdings, then called and placed 4 million dollars. One behaviour looked like engagement. The other predicted an allocation.

Telling those two apart is the whole job here.

The problem: vanity engagement vs. predictive engagement

Marketing in this sector sits far from the money. The buyer is a financial advisor, a fund selector or an institutional gatekeeper, and marketing almost never sees the signature. So teams count what they can see: downloads, opens, event RSVPs. Cheap to count, and on their own close to meaningless. A 42 percent open rate on a monthly commentary moves no basis points of AUM.

Engagement scoring is the discipline of finding which observable behaviours actually correlate with money arriving, and weighting them by how strongly they do.

Step 1: Pick one conversion event

You cannot score engagement without an outcome to predict. Anchor everything to a single event drawn from the stages the funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.View full definition → lesson already names: a first ticket, net new money above a threshold, or acceptance into formal due diligence.

Retail-facing teams usually take "first ticket within 90 days". Institutional teams should not. A consultant-led search runs 9 to 18 months, so a model trained on closed institutional tickets is learning from behaviour that happened two years ago, before your website was redesigned and before half the contacts changed firms. Use entry into due diligence as the proxy outcome and accept the loss of precision.

Step 2: Inventory the signals

Two of these deserve a caveat. Sales enablement vendors such as Seismic sell this telemetry, and their published case studies are marketing for their own product, so test the lift on your own book before you believe a claimed uplift. And third-party lookup data, Morningstar page views on your strategy or Broadridge's aggregated intermediary flows, arrives at firm level, not person level. That mismatch matters in step 3.

Step 3: Test which signals actually correlate

Take your closed tickets from the last 12 to 18 months. For each behaviour, compare the conversion rate of advisors who did it against those who did not.

A worked example (illustrative numbers, not real firm data). You have 1,000 tracked advisors and 80 wrote a ticket, an 8 percent base rate.

• Of 200 advisors who logged into the portal 2+ times in a month, 40 converted: 20 percent, 2.5x the base.
• Of 300 who downloaded 3+ research pieces, 30 converted: 10 percent, 1.25x.
• Of 150 who attended a live roadshow, 36 converted: 24 percent, 3x.

lift = conversion_rate_with_signal / base_conversion_rate

portal login (2+):   0.20 / 0.08 = 2.5x
research (3+):       0.10 / 0.08 = 1.25x
roadshow attendance: 0.24 / 0.08 = 3.0x

Research downloads barely move the needle. That contradicts the wholesaler's instinct, and it is the kind of finding that reallocates budget.

Three ways this analysis goes wrong, all common:

Leakage. If you count behaviours right up to the ticket date, you will discover that people who buy your fund read your fund page. Freeze the observation window at the moment the opportunity was created, or at a fixed 30 days before the ticket. Lift numbers usually halve when you do this, and the halved numbers are the true ones.

Thin cells. The roadshow row above rests on 36 conversions. Below roughly 30 conversions in a cell, the confidence interval is wide enough that a 3x lift and a 1.5x lift are hard to separate. Report the count next to every weight so nobody builds a 35-point rule on nine events.

Correlation borrowed from sales. Roadshow attendance may score high because wholesalers invite the advisors they already expect to buy. The invitation list is the predictor, not the room. Test it by looking only at advisors who registered from an open marketing email.

The entity problem sits on top of all three. At a wirehouse or a large IFA network, the person downloading the due diligence questionnaire is a research analyst who will never write a ticket, and the tickets come from advisors who touched nothing. Roll signals up to the office or firm, score at that level, and the research download that looked worthless at person level often turns into your strongest institutional predictor.

Step 4: Build a simple lead score

You do not need machine learning to start. A weighted additive model works and is explainable to compliance and to sales.

score = (portal_logins_flag * 30)
      + (roadshow_flag * 35)
      + (research_3plus_flag * 15)
      + (email_click_flag * 5)
      + (recency_bonus)     # active in last 14 days

# threshold: hand to sales when score >= 50

Weights come from the lift analysis, not from opinion. Give the score a decay, something like a 30-day half-life, so an advisor who binged content in March does not still sit at the top of the queue in July. Allocation windows open and close with client reviews and market moves.

Set the handoff threshold by wholesaler capacity, not by model recall. A wholesaler covering a territory manages perhaps six to ten substantive conversations a day. If your threshold sends eight of them 400 names a week, the score is not a filter, and within a month the field will be working its own list again. Tune the cutoff until the weekly queue is roughly what the team can call, then argue about coverage separately.

For a clear primer on the underlying method, HubSpot's guide on lead scoring covers the mechanics well (it is generic, so translate "MQLMQLA Marketing Qualified Lead (MQL) is a prospect whose engagement and fit signals indicate they are more likely to become a customer, justifying handoff toward sales.View full definition →" to "advisor ready for a wholesaler call").

Step 5: Watch the flow, not just the score

A score only means something against the stage sequence the funnel lesson sets out. Compute conversion between each of those stages and look for where scored advisors stall. Industry surveys commonly cite sales acceptance of marketing-qualified leads in the rough range of 20 to 40 percent across B2B (estimate, varies by source and year), and asset management sits near the bottom because wholesalers have been burned by download tourists.

🎬 [VIDEO: "How to Build a Lead Scoring Model" - youtube.com - a practical walkthrough of weighting behaviors and setting handoff thresholds]

The failure mode nobody catches: the model becomes self-fulfilling. Wholesalers call high scorers, high scorers convert, the next refresh confirms the weights, and low-scoring advisors who would have bought are never contacted. Hold back a random 5 to 10 percent of scored advisors from the priority queue each quarter and call them anyway. If their conversion rateconversion rateThe percentage of visitors or prospects who complete a desired action (purchase, sign-up, contact form), calculated as conversions divided by total opportunities.View full definition → is close to your high scorers', your score is measuring wholesaler attention rather than advisor intent.

Step 6: Score for what the money is worth, not just for the first ticket

A first ticket is a milestone, not the prize, because revenue is basis points on an asset base held over years, as the lifetime valuelifetime valueLifetime Value: the total revenue (or profit) a customer generates throughout their entire relationship with your business.View full definition → lesson works through. What that changes for engagement measurement is which behaviours you reward.

Optimising purely for first tickets selects for the fastest buyers, and the fastest buyers are often model-portfolio and tactical allocators who move a block in and out on a single rebalance decision. They flatter your conversion rate and can reverse the whole position in one instruction. Signals tied to ongoing use of your material (repeat portal logins across quarters, attendance at the same technical webinar series, requests for holdings-level data) tend to track advisors who build a position and stay. Track lift against tenure at 24 months as a second outcome, not only against the ticket, and be ready for the two rankings to disagree.

Step 7: Retention engagement is its own funnel

Once an advisor allocates, the same telemetry becomes churn detection. Signals that an allocator is cooling:

• Portal logins fall to zero for 60+ days.
• They stop opening performance updates during a drawdown, which often runs ahead of a redemption call.
• Comparison lookups on peer funds in your category spike, where syndicated data lets you see it.

One signal reverses meaning depending on tenure. Two portal logins from a prospect is buying intent. Two portal logins from a three-year holder in week three of a drawdown, checking position sizes rather than the fact sheet, is usually the opposite. Score prospects and holders with separate models, or you will send a wholesaler a congratulation call the week before a redemption.

Flagging a disengaging allocator before the redemption lands is where this measurement pays for itself a second time. The benchmarks for judging that outcome belong to the retention and net flow lesson.

A note on data and regulation

You are tracking advisor behaviour, so respect the rules. In Europe, GDPRGDPREU regulation governing how organizations collect, store and use personal data, with fines tied to global revenue for breaches.View full definition → governs consent and data use. In the US, marketing communications from registered firms fall under SEC and FINRA advertising and recordkeeping rules, including the SEC Marketing Rule. Practical implication: keep scoring behavioural and consented, retain records, and never let a "warm intent" flag turn into a performance claim you cannot substantiate. This is not legal advice; loop in compliance early.

Key Takeaways

• Score by measured lift, not instinct. A download habit may predict nothing while two portal logins predict a ticket.
• Freeze the observation window before the opportunity opens, and refuse to weight any behaviour with fewer than about 30 conversions behind it.
• Roll signals up to the firm or office where analysts research and advisors buy; person-level scoring hides institutional intent.
• Set the handoff threshold by wholesaler capacity, and keep a random holdout so the model measures intent rather than who sales happened to call.
• Recency beats volume, and prospects and existing holders need separate models: the same portal login means buying in one and redeeming in the other.
• Keep scoring compliant under GDPR and SEC/FINRA rules, with records that survive an examination.