How Much Discount Should You Authorize? The Math Behind Your ICP

Every vendor we onboard hits the same wall in week one. They’ve accepted the framing that their ICP isn’t a targeting tool but a discount authorization framework. They’ve walked through the four LTV dimensions — retention, expansion, cost efficiency, network value. Then they sit down in the portal, see the field labeled max_margin_percent, and freeze.

How much should they authorize? 10%? 25%? 40%? Their CFO is going to ask. The answer can’t be “whatever feels right.” It needs to be a number a finance team can defend.

This post is the math. It’s not exotic — CAC payback and LTV are decades old. But applying them to agent-mediated discount authorization changes a few things, and most vendors haven’t done it from this angle. We’ll work an Australian residential electricity example end to end so the numbers are concrete.

The one question your ICP is actually answering

Strip everything else away and the question your ICP needs to answer is:

For this kind of customer, what’s the most we can give up at acquisition and still come out ahead?

That’s it. The four dimensions are scaffolding to estimate it. The max_margin_percent field is where the answer lives. Everything else in the portal is in service of computing that ceiling per ICP segment.

Note the word ceiling. You’re not setting a discount — you’re setting a budget the vendor agent is allowed to spend. The agent will often close below the ceiling. We’ll come back to that.

The base equation

The acquisition you can afford is bounded by the gross margin the customer will produce:

max_acquisition_spend  =  expected_LTV_margin  ×  payback_appetite

where:
  expected_LTV_margin  =  Σ (period_revenue × gross_margin × retention_prob)
                                  for each period in the horizon

  payback_appetite     =  fraction of LTV you’re willing to spend
                          on acquisition (typically 0.20–0.40
                          for capital-efficient growth,
                          0.50–0.80 for land-grab phases)

Your finance team has a number for payback_appetite already, whether they call it that or not. It’s implicit in how they think about CAC. The work in setting agent-channel discount ceilings is mostly about estimating expected_LTV_margin well, and decomposing it across the four ICP dimensions so different customer profiles get different ceilings.

The four dimensions, translated into dollars

The reason haggl’s ICP is structured around four dimensions instead of one big “customer score” is that each dimension contributes to LTV through a different mechanism. They translate to dollars differently. Mixing them into a single number throws away the information that makes the ceiling defensible.

Sum the four lines and you have expected_LTV_margin. Multiply by your payback appetite. That’s the ceiling for this segment, in absolute dollars. Divide by the deal’s annual revenue and you have max_margin_percent.

A worked example: residential electricity in Australia

Imagine you’re an Australian electricity retailer listing a residential ICP segment. The agent shows up with six DKIM-verified monthly bills averaging A$170, no payment gaps, plus a self-reported claim of one household in Melbourne and an open willingness to bundle gas onto the account.

Working the dimensions:

Customer profile: residential, 6mo verified spend, no gaps,
                  Melbourne VIC 3056, open to dual-fuel bundle,
                  single household.

——— Retention ———
  Industry baseline for this segment:   18 months
  Verified history (6mo, no gaps)    →  +6 months
  Open to dual-fuel (lock-in)        →  +6 months
  Expected tenure                       30 months

  Monthly retail margin per customer    A$22
  Retention contribution                30 × A$22 = A$660

——— Expansion ———
  P(gas attach within 6mo) given
    stated willingness                  0.40
  Gas annual margin                     A$140 × 24mo
  Expansion contribution                0.40 × A$140 × 24/12 = A$112

——— Cost efficiency ———
  Baseline cost-to-serve / yr           A$90
  E-billed, direct-debit, no calls
    (inferred from DKIM bill pattern)   A$35
  Annual savings × expected tenure      (A$90 − A$35) × 2.5 = A$138

——— Network value ———
  Residential single household,
  not a high-referral profile           0.05 × A$660 = A$33

EXPECTED LTV MARGIN                     A$943

× Payback appetite (0.30, capital-
  efficient growth posture)              0.30

MAX ACQUISITION SPEND                   A$283

÷ Year-1 revenue (A$170 × 12)           A$2,040

MAX_MARGIN_PERCENT                      13.9%

So for this customer profile, the ceiling is roughly 14% off year-one billing. The vendor agent receives this authorization, runs the negotiation, and probably closes somewhere between 8–12% — it doesn’t need to use the whole budget. The ceiling exists so the agent has latitude to handle hard cases, not so it spends to the max every time.

Now run the same math for a different profile in the same segment — say, a customer with three prior retailer switches in 24 months, paper billing, and a history of late payments:

Expected tenure:        9 months
  Retention contribution:  9 × A$22 = A$198
  Expansion contribution:  ~0 (won’t be around)
  Cost efficiency:         negative (paper billing, call-centre load)
  Network value:           ~0
EXPECTED LTV MARGIN        ≈ A$170

MAX ACQUISITION SPEND      A$170 × 0.30 = A$51
MAX_MARGIN_PERCENT         2.8%

The agent gets almost no room to discount this customer, and that’s correct. The churn risk and cost-to-serve mean any deeper discount is destroying value. The buyer agent, seeing the rejection, can either bring stronger evidence (which moves the profile back toward higher LTV) or walk. Both outcomes are fine for the retailer.

Why ceilings beat fixed coupons

Most vendors’ discount infrastructure today is built around coupons: a fixed percentage off, applied to anyone who possesses the code. Authorizing a margin ceiling looks similar at first but behaves completely differently.

• A ceiling is private; the negotiated offer is what the buyer sees. No competitor can scrape your max_margin and use it as their floor. A coupon, by contrast, leaks the instant it’s used.
• The vendor agent can choose a number below the ceiling. If the buyer submitted strong evidence but their target price is already close to your list price, the agent gives a small discount. If they’re asking for a bigger move and the LTV supports it, the agent uses more of the budget. Coupons are all-or-nothing.
• Ceilings are profile-conditioned, coupons are not. A coupon doesn’t care if you’re a high-LTV customer or a churn risk — same discount either way. Ceiling-based authorization sends the deepest budget to the profiles that earn it back.
• Ceilings degrade gracefully under fraud attempts. If a buyer agent submits fake evidence and somehow gets past verification, the worst case is they unlock a slightly higher ceiling and the vendor agent closes a few points below it. The downside is bounded. A leaked coupon has unbounded downside — it propagates.

The cross-check finance will ask for

Whoever signs off on your ICP ceilings is going to compare them to your existing CAC. That’s a fair instinct and the wrong number to compare against.

Your blended Google Ads CAC is the cost to acquire a click that might convert. Your inside-sales CAC is the fully-loaded cost to acquire a verified, qualified, closed deal. The agent channel is structurally closer to the second one: the buyer arrives with verified evidence (DKIM-signed bills, structured ICP fit), willingness-to-pay is already declared, and the “close” is a single agent-mediated handoff.

For most B2C-utility and SMB-SaaS vendors we’ve worked with, inside-sales fully-loaded CAC for an equivalently qualified lead is 2–4× their paid-search CAC. Authorizing agent-channel ceilings against the paid-search number under-spends and leaves deals on the table. Use the inside-sales number as the cross-check; you’ll usually find your ceiling can be meaningfully higher and still pencil.

Day-1 ceilings when you don’t have the data yet

The math above assumes you know your retention curves, your expansion rates, your cost-to-serve deltas, and your payback appetite. Most vendors only know some of those. Three pragmatic shortcuts for the first 90 days:

1. Start at 60–70% of what the math suggests. You can always raise it later. Raising a ceiling once you have evidence is easy; lowering one after you’ve told the market “we authorize up to X” is harder.
2. Use industry retention benchmarks for the missing inputs. Your trade association, your competitive intel, or even the published S-1s of public comps in your space. The numbers don’t need to be precise — they need to be defensible.
3. Watch the negotiation logs. Within 30 days you’ll have a distribution of where the vendor agent is actually closing — not just the ceilings. That distribution is your evidence to tighten or widen segments. If the agent is closing at the ceiling on 90% of deals in a segment, you’ve set the ceiling too low and you’re leaving money in. If it’s closing at 30% of the ceiling, the ceiling isn’t binding and the actual constraint is the buyer agent’s target price.

The control loop

Ceilings aren’t a one-time number; they’re a quarterly recalibration. Three things change them:

• Cohort retention data. The 30-month tenure assumption in the example above is a guess. After two quarters you’ll have early retention signal for the customers acquired via the agent channel specifically — and they often behave differently from your control channels (typically: longer tenure, because the qualifying evidence selected for it).
• Channel-level payback. The right test isn’t “is each deal profitable on day one” (they won’t be at meaningful discount), it’s “is the agent-channel cohort paying back within the window we set.” If yes, loosen ceilings. If no, tighten the dimensions most responsible for the miss.
• Competitive pressure inside the agent channel. As more vendors in your category turn on haggl, agents will compare offers. If your win rate falls and the rejected fits look like profiles you wanted, your ceiling on those profiles is too low relative to peers. The negotiation logs will tell you which dimensions drove the loss.

None of this requires watching individual deals. The whole point of authorizing budgets and letting the vendor agent negotiate within them is that the finance team operates one level up: setting ceilings, reading cohort outcomes, adjusting weights. The unit-economics decisions don’t happen at the deal level. They happen at the segment level, every quarter, with real data.

The mental model in one sentence

Your ICP segment ceilings are the answer to “how much margin am I willing to forgive at the start of the relationship to acquire this kind of customer?” It’s underwriting, not marketing. The four dimensions exist so the underwriting can be conditional on what the buyer actually proves about themselves. The agent does the deal. Finance sets the budget. The math holds up at the board meeting.

Related Reading:

• Your ICP Is Not About Who’s Ready to Buy
• Anatomy of a <meta name="haggl-negotiate"> Tag
• Privacy Without Trust: How haggl.ai Solves the Accountability Dilemma
• haggl.ai for Energy Providers