Model 3 — What To Add Next
The Problem With Model 2
Model 2 optimises two variables: the advance rate and the charged borrow rate. Those are the two a term sheet puts on its front page. They are not the two that move your return the most.
This page is a written plan, not a calculator. It ranks what to add, and it puts the items that change the answer most at the top. Every number below is worked at a net asset return of 60%, with the bank at a 25% advance rate and a 17% charged borrow rate, all in shillings. Model 2 now opens at 55%, so re-run anything you want to quote.
Part 1 — The Variables Worth Negotiating, In Order
Ranked by how much each one moves return on equity, not by how prominent it is in the document. The first three are worth more than the charged borrow rate, and none of them is on model 2.
1. Eligibility criteria and reserves. The biggest lever on the page, and the least discussed. A lender does not advance against your whole book. It advances against eligible receivables, after deducting reserves for dilution and yield. So the number that matters is not the headline advance rate, it is the effective advance rate:
a_eff = a × eligible × (1 − reserves)
An 80% advance rate, against a book that is 75% eligible, with a 5% reserve, is really
0.80 × 0.75 × 0.95 = 57%.
At 60% net return and a 17% charged borrow rate, that gap is worth 115 percentage points of return on equity: 232% at a true 80% advance rate against 117% at an effective 57%. Winning ten points of headline advance rate is worth less than winning back one exclusion in the eligibility schedule. Negotiate the schedule before you negotiate the rate.
2. The revolving period, and whether collections may be re-lent. Model 2 is a single period. Real receivables pay down and the cash comes back. If you may re-lend that cash, the same equity earns the same return several times over the life of the facility. If the facility amortises instead, every shilling collected repays debt and your book shrinks. This is the difference between a facility and a loan, and no rate concession compensates for getting it wrong.
3. Facility size and the price of undrawn capital. A commitment fee is charged on what you have not drawn. Take a facility twice the size you draw, with undrawn priced at 70% of the live rate, and your effective cost is not 17%:
r_eff = r + 0.70 × r × (undrawn / drawn)
At a facility twice your drawdown: 17% + 11.9% = 28.9%. An oversized facility nearly doubles your
borrowing cost.
At an 80% advance rate that costs 48 points of return on equity, from 232% down to 184%. Headroom you do not use is not free optionality. It is a rate rise you agreed to in advance.
4. Tenor, and the tenor of the hedge. A three-year commitment is worth materially more than a one-year commitment that has to be renegotiated at the worst possible moment. Model 2 has no time in it at all. Worse, the hedging box assumes a hedge for the whole life of the facility. If the hedge market only reaches one year and the facility runs three, you are unhedged for two of them, and nothing on model 2 shows that.
5. Covenants, triggers and the cash sweep. A portfolio-quality trigger that diverts all collections to the lender can end your business while the facility is still technically performing. Model 2 shows the return in the good case only. The right question is not "what does this earn", it is "at what level of arrears does this facility take my company away from me".
6. The advance rate. In model 2. Still the strongest of the two headline terms, by roughly 2.3 times at the bank's quote.
7. Recourse and the first-loss position. Who takes the first loss, and how far up does it go? A first-loss piece you fund yourself is economically an advance-rate reduction wearing different words.
8. The charged borrow rate. In model 2. Ranked eighth deliberately. At the bank's quote the lender earns about one point over Kenyan government paper. There is very little there to win, and every hour spent arguing it is an hour not spent on items 1 to 3.
9. Fees. Arrangement, structuring, agency, legal, valuation. Removed from model 2 by decision D4 to keep it teachable. They are real, they are mostly one-off, and they should be amortised into an all-in rate before any comparison.
10. Currency of the facility. Covered by the hedging box on model 2, but only as a price. The availability question in item 4 matters more than the price.
11. Prepayment rights and make-whole. The right to refinance when terms improve is an option you own. A make-whole clause sells it back to the lender for nothing.
12. Concentration limits. Caps per obligor, region or vintage. These bind quietly, and they reduce the effective advance rate exactly like item 1.
Part 2 — What Model 3 Should Compute That Model 2 Cannot
1. Replace the advance rate with the effective advance rate. Three inputs, one line of arithmetic, and it is the largest correction on this page. Model 2 currently reports returns that no real facility will deliver, because it takes the headline advance rate at face value.
2. Show the downside, not only the return. This is the honest gap in model 2. Leverage multiplies your estimate error at exactly the same rate it multiplies your return:
∂ROE/∂y = 1 / (1 − a)
At a 25% advance rate, one point of error in your asset return moves return on equity by
1.33 points. At 80%, it moves it by 5 points.
y_breakeven = a × r
The asset return at which equity earns nothing. At the bank's terms, 4.3%. At an 80% advance rate
and 17%, 13.6%.
So the cap is not simply better. It is better and it moves your ruin point from 4.3% up to 13.6%, and it makes you five times as sensitive to being wrong about the one number nobody has verified. Model 2 says "take the cap" without ever showing that price. Model 3 should show both curves on the same axes.
3. Make the asset return a range, not a point. The 60% is a single unverified figure carrying the whole argument. Take a low, central and high case, and report return on equity for each. A lender who sees the low case handled trusts the central case more, not less.
4. Add periods, and let collections re-lend. Single-period is the assumption that most understates the business. It should be the first thing to go once the effective advance rate is in.
5. Charge for undrawn capital, and make utilisation an input. Then facility size becomes a decision rather than a free parameter, and item 3 in part 1 becomes visible.
6. Model the lender's return as well as your own. Model 2 sees one side. With both sides you can find the deals that are better for both parties rather than merely better for you, which is what actually gets signed. That set is the efficient frontier, and the argument "this costs you nothing and is worth 15 points to me" closes deals that "your rate is too high" does not.
7. Timing. Drawdown lag, collection lag, and the gap between deploying capital and earning on it. All of these reduce the realised return below the modelled one.
8. Tax. Interest is usually deductible and the return is usually taxed. Both change the answer and neither is hard to add.
Part 3 — What I Would Build First
Three changes, in this order. Each is small. Together they change the conclusion.
First, the effective advance rate. Three sliders, one multiplication. It is the difference between a 232% headline and a 117% reality, and it redirects the negotiation to the schedule where the money actually is.
Second, the downside curve. Plot the break-even asset return beside the return on equity. It is the only thing that makes "take the maximum advance rate" a responsible piece of advice rather than a reckless one.
Third, re-lending over several periods. The largest understatement in the current model, and the one that turns the advance-rate argument from strong into overwhelming.
What I would not build. A perceived-risk ladder. Model 2 had one and it was removed, because the only quote we hold prices no credit risk and therefore reveals nothing about how a ladder rises. Rebuilding it would mean inventing a shape and then reading conclusions off the invention. Get a second real quote at a high advance rate first. One real quote is worth more than any amount of modelling.