Modeling client lifetime value when engagements are irregular and unpredictable

Two revenue shapes sit inside the same firm. KPMG's statutory audit work arrives on a calendar: same client, same quarter, a fee that moves a few percent a year. Lazard's advisory revenue arrives when a board decides to sell a division, so a client can pay nothing for three years and then produce one of the largest fees of the quarter. Any LTVLTVLifetime Value: the total revenue (or profit) a customer generates throughout their entire relationship with your business.View full definition → (lifetime value) number you put in front of a partnership has to survive both shapes, because the same client often supplies both.

The subscription formula writes off the quiet client at month 13. That one error pushes firms to cut nurture spend on the accounts that are appreciating fastest.

Why the SaaS LTV formula breaks here

LTV = (Average Revenue per Account × Gross MarginGross MarginGross margin is the share of revenue left after subtracting the direct cost of producing goods or services, expressed as a percentage of revenue.View full definition → %) / Churn RateChurn RateChurn rate is the percentage of customers or revenue lost over a period. It measures how fast a business loses its existing customer base.View full definition →

This assumes steady periodic revenue and a stable churn rate. Neither holds outside the annuity half of the book. A client who has not instructed you in 18 months is not churned, they are between mandates: the general counsel who hired your firm for one dispute is dormant until the next one arrives. Treating dormancy as death understates the relationship and misprices the client base underneath it.

Annuity work and episodic mandates behave differently

Recurring compliance work behaves close to a subscription: re-engagement probability near 1, low fee variance, and the real risk sitting in tenure limits rather than satisfaction. For audit, that limit is written into law. Under the EU audit reform rules in force since 2016, public interest entities must retender the statutory audit after ten years and rotate the firm out at twenty (longer only under a joint audit arrangement). A KPMG-style audit annuity therefore has a legal expiry date no matter how content the audit committee is, and your horizon assumption has to respect it. Independence rules cut the other way as well: the incumbent auditor is barred from selling a long list of non-audit services to that client, so the expansion multiplier on an audit relationship is capped by regulation rather than by trust.

Episodic mandates invert all of it. Frequency is a probability rather than a fact, fee size is a distribution with a long right tail, and one mandate can exceed a decade of compliance fees. Lazard has long reported the count of clients paying fees above a set threshold rather than anything resembling recurring revenue, which tells you how the underlying economics actually behave.

The core shift: think in engagement probability, not renewal rate

Instead of asking "will this client renew," ask two separate questions:

1. What is the probability this client re-engages us in a given period?
2. Given re-engagement, what is the expected size of that mandate?

This is the logic behind probabilistic customer lifetime models used in retail and telecom, notably the BG/NBD model (Beta-Geometric/Negative Binomial Distribution) popularized by Fader and Hardie for irregular purchase behavior. You do not need the full statistical machinery to borrow the mindset: separate *frequency* from *monetary value*, and model both as distributions rather than constants.

Building blocks for a professional services LTV

1. Engagement frequency curve

Take clients acquired more than three years ago and ask what share generated at least one new mandate in years 2, 3, 4 and 5. That gives you a curve instead of a single churn figure. A boutique M&A advisory firm might find 40% of clients return within 24 months, another 15% between 24 and 48 months, and the remainder never. Two traps here. Right-censoring: a cohort acquired 14 months ago has not had time to come back, so pooling it with a five-year-old cohort drags your frequency estimate down. Survivorship: clients purged from the CRMCRMCustomer Relationship Management: software and strategy to manage and analyse customer interactions throughout their lifecycle.View full definition → when they went quiet must stay in the denominator, or you will model only the winners.

2. Expansion multiplier

Compute it from your own billing ledger: second and third mandate value against the first. Repeat mandates are frequently larger, because scope widens once trust exists (audit becomes audit plus tax plus advisory, where independence rules allow). Read the result against the sector norms the benchmarking lesson supplies rather than borrowing a multiplier from someone else's practice mix.

3. Relationship horizon and a discount rate

Use an observed horizon: how many years does a relationship stay live before going genuinely cold (firm sold, business closed, sponsor departed)? For many B2B professional services firms that runs 7 to 10 years, estimate, far longer than the gaps between mandates. Note that the horizon lives in a person, not a legal entity. When the buying sponsor leaves, reset re-engagement probability to your cold prospect base rate until a second relationship is built. McKinsey's alumni network, tens of thousands of former consultants now sitting in buyer seats, is that same mechanism running in your favour.

A worked example

A mid-sized accounting firm client:

• Initial engagement value: $20,000
• Annual probability of re-engagement (from cohort data): 25%
• Expansion multiplier: 1.5x
• Relationship horizon: 8 years
• Gross margin: 35% (typical professional services range, estimate)
• Discount rate: 10%

Expected value per re-engagement is $30,000, so the expected repeat stream is $7,500 a year. Undiscounted across eight years that is $60,000, plus the initial $20,000, giving $80,000 revenue and $28,000 gross profit LTV.

Discounting matters when the money is years out. At 10%, the eight-year annuity factor is 5.33, so the repeat stream is worth about $40,000 today. Total $60,000, gross profit LTV about $21,000. Skipping the discount rate inflates the number by a third and is the first thing a finance partner will attack.

A naive model that declares the client lost after 18 months of silence counts only $7,000. The defensible figure is still three times that, and the gap is the argument for nurture spend during quiet periods.

One caution on the arithmetic. With a 25% annual probability, most clients return once or never; the mean is carried by a minority who return three or four times. Budget nurture programs against the segment mean, and never commit per-client spend as though every account will deliver it.

What this changes about acquisition spend

Set the $21,000 against the loaded acquisition costacquisition costCustomer Acquisition Cost (CAC) is the total sales and marketing spend divided by the number of new customers gained in a period. It measures how efficiently you grow.View full definition → the sibling lesson teaches you to build, not against first-mandate profit. A $20,000 first engagement at 35% margin returns $7,000, which looks like a thin payback on an $8,000 pursuit. Against lifetime value it is comfortable, and it justifies harder acquisition spend on the right client profile.

Two second-order effects. Firms that judge business development on first-year ROIROIReturn on Investment: the ratio of net profit to the cost of an investment. A 300% ROI means each dollar invested returns $3.View full definition → chronically underfund the work that pays on a 3 to 7 year horizon: thought leadership, alumni contact, referral cultivation. And discounted LTV is not cash. A partnership distributing profits annually cannot fund this year's draw out of a mandate expected in year six, so thin working capitalworking capitalWorking capital is the difference between a company's current assets and current liabilities, measuring short-term liquidity and the funds available to run daily operations.View full definition → can rationally cap episodic-client acquisition even when the model says buy.

Segment before you model

Don't build one LTV number for the whole firm. Segment by:

• Practice area (litigation and deal mandates are lumpier than recurring compliance)
• Client size (larger clients carry longer horizons and higher multipliers)
• Referral source (referred clients usually show higher re-engagement in a firm's own dataown dataData collected directly from your own customers and prospects through your own channels: your most reliable and privacy-compliant source.View full definition →; verify it in yours)

Average these together and you will move budget toward whichever segment the blended number flatters.

Data you actually need to track

To run this with real numbers, your CRM (customer relationship management system) needs at minimum:

client_id, engagement_start_date, engagement_end_date,
engagement_value, practice_area, referral_source,
is_repeat_engagement (boolean), prior_engagement_value

Most firms have this scattered across billing systems, partner spreadsheets and memory. The first practical step is consolidating engagement history into one queryable table. Practice management and billing platforms usually hold the dates and values already; the fields almost always missing are referral source and the identity of the sponsoring contact, and without those two you cannot segment, nor spot the sponsor departure that just reset a relationship to zero.

🎬 [VIDEO: "Customer Lifetime Value Explained (BG/NBD Model Intuition)" - youtube.com - a walkthrough of probabilistic LTV modeling logic adaptable from retail to relationship-based B2B services]

Key takeaways

• Dormancy is not death. Model re-engagement probability separately from mandate size, and never apply a subscription churn rate to episodic work.
• Annuity work and deal mandates need different assumptions: audit-style annuities are capped by rotation and independence rules, episodic mandates by a long-tailed fee distribution.
• Build from four inputs: frequency curve, expansion multiplier, relationship horizon (often 7 to 10 years, estimate) and a discount rate. Skipping the discount rate overstates LTV by roughly a third on an eight-year horizon.
• Guard the frequency estimate against censoring and survivorship, and remember the mean is carried by a returning minority.
• The bottleneck is data hygiene, not statistics: consolidate engagement history, referral source and sponsor identity before modeling anything.