Every step is a promise. The broken ones are worth most.

01Losses multiplyAnd the weakest step is where a fix pays most

A bag of meal on a shelf in Lusaka is the end of a chain. Grain was bought at harvest, stored, powered, milled, certified, packed, delivered and paid for. Each step lost a little. A week’s delay. A lot that spoiled. A generator run through a power cut. A refund that took a year. None of the losses is large. All of them are still in the bag.

Each step keeps a share of what reaches it, so the shares multiply. Eight steps at 90 percent deliver 43 percent. Let one work at 50 and the chain falls to 24. Fix that step and the chain nearly doubles; make a good step perfect instead and 24 becomes 27. Point for point, a gain at the weak step is worth almost twice as much, so start there, if a point costs about the same everywhere. Often it does not: a step is often weak because it was hard to fix.

What matters is that the steps depend on each other. In 1986 the shuttle Challenger was lost to one rubber seal that failed in the cold, and in 1993 Michael Kremer used that failure to explain why poor and rich countries are so far apart: a weak step lowers the value of every strong one, and the best-run firm loses most. The shuttle is the extreme case. The bag of meal is the ordinary one.

That is why so many Zambians trade. In 2013, the last time it was measured, 39 percent of adults were starting or running a business; owners in business for over three and a half years were under a tenth of that. A trader’s chain has three steps: buy, move, sell. A maker’s has ten: inputs, power, processing, certification, packing, delivery, payment, refunds, and the permits in between. At 90 percent a step, the trader’s chain delivers 73 percent, the maker’s about a third. The count depends on how steps are drawn. The gap does not.

The border should help. Zambia is landlocked, so every import carries the freight, and on paper a maker selling at home starts with that cushion. On the shelf it does not show: the supermarket chains stock what their South African buyers list, and imports from Asia land below local cost (the author’s observation from the mill). Every imported input and every export pays the freight too. Trading is not a lack of ambition. Here it is the sensible calculation.

That finds the step where a point is worth most. It does not say why that step loses so much here, or whether it can be mended at all. So look at what a step is.

02A step is a promiseAnd a promise binds in proportion to what it costs to break it

Look closely at any step and it is a promise between two parties. Deliver by Friday. Pay in thirty days. Approve within the statutory period. Refund the tax. A step works when the promise is kept on the date, and loses when it is kept late, in part, or not at all. The losses down a Zambian chain are promises that do not bind: the unpaid bill, the refund that never arrives, the permit with no deadline, the loan nobody collects.

A promise is kept when breaking it costs the promiser more than keeping it. That cost comes from four places, the same four throughout this page. Exit: a customer who leaves, a supplier who stops, a counterparty who can walk away. A court the other side can actually use. Visibility: a break seen by someone who matters to the promiser. Self-execution: money that moves on its own, a deduction at source, a key that does not turn.

The second sentence is the first under pressure, tested here against two simpler rules: cash alone, which squeezes every creditor together, and first come, first served. The rule is not new. A sovereign’s promise is believed when something can make breaking it costly to the sovereign; where courts are weak, trade runs on relationships whose future is worth more than a broken promise. What this page adds is the Zambian record.

The mechanism decides, not the intention. The United States cut its payment time to small contractors from thirty days to fifteen; payroll at those firms rose ten cents for every dollar paid sooner, partly offset by fewer jobs at other firms where labour was scarce. Europe wrote the same right into law, with interest after thirty days, and less changed than intended: a small firm will not claim interest from a payer it depends on. The promise bound where breaking it cost the payer automatically, not where the supplier had to dare to ask. The United States was not short of cash, so the lesson carries only to arrears a Treasury could pay.

Two qualifiers. A target is not a promise: it has no counterparty and no cost for missing it. Once someone pays for missing it, it has become a promise. And some promises no decision can keep. In February 2026 the National Health Insurance Management Authority (NHIMA), the state health insurer, owed clinics K500 to 600 million against a 45-day contract, and said it did not have the money. Raising the cost of breaking that promise would only move the loss to another creditor. This page is about the promises that could be kept and are not: the steps where the fix is a decision, not money. Those are the ones worth finding first.

03Who owns the stepThe state’s order of payment

This page’s conjecture is that many of the steps that hold a Zambian chain back are owned by a public body and end in a payment. So who owns the step becomes: in what order does the state pay? The law writes part of the order down. Public debt is a charge on the Consolidated Fund, pensions are to be “paid promptly and regularly”, statutory payments come first, and the rest is released at the Treasury’s discretion. The Zambian record shows the order with unusual clarity: one administration paying the bills of two.

The Treasury’s 2024 arrears strategy states four principles for which old bills it clears first, in its own words: “arrears related to critical services with social impact”, which it says entails paying “contractors offering discounts to the Government on their claims”; arrears “where refinancing at a cheaper cost is possible”; “first in first out”, older bills before newer; and “local suppliers of goods and services”, whose bills are smaller than external contractors’. They are not presented as a ranking. The same strategy then sorts everything still owed in March 2024 into five numbered categories, each with a year by which it is to be cleared (its Table 2.0). It also shows what each kind of creditor was owed at the end of 2021, and what was paid in the next 27 months:

Treasury arrears strategy 2024, Table 1.0, all fourteen creditor classes, with its three utility rows (phone, power, water) merged into one: payments January 2022 to March 2024 beside the end-2021 stock. The Treasury’s own subtotal covers the twelve domestic classes, K59.5 billion owed and K17.7 billion paid; it carries the power producers separately, as a contingent liability paid by ZESCO, the state power utility, and the contractors on cancelled loans as a new class. Shares and coding are this page’s. Where payment equals the stock it went to the old bills; elsewhere some may have gone to newer ones. The last column is the Treasury’s own sorting, in August 2024, of what was still owed into five categories with a clearance horizon each (Table 2.0); farmers and the utilities were already cleared and have none. The rank correlation of each column with the share paid is in section 11.

The coded column tests five things: a duty with priority written in law (law); a creditor who can stop something the state needs within the season (supply); a large group that votes or protests together (crowd); a court order (court); a claim that grows by contract while it waits (penalty). The tests were written after the percentages were known, so the coding is not blind. The contractors on cancelled loans have no end-2021 stock, so no share to test, and are left out of the correlations.

Read honestly, the table half fits. Every class paid 80 percent or more scores on a test; every class that scores on none was paid about a third or less. The big misses are fuel and roads. Fuel suppliers could stop supply and held penalty clauses, and were paid 14 percent; their bill grew until interest and penalties exceeded the principal, and they were cleared in 2024 and 2025 with bank loans. Road contractors held interest clauses and were paid one percent. A penalty the creditor cannot execute against the State (section 05) adds to the bill, not to the pressure. The VAT row, coded none, was paid 34 percent though the mines held most of the refund stock. Two simpler rules compete. First come, first served does not fit: every row is the same end-2021 stock, and the shares run from 1 to 100. Cash alone fits better than this page’s coding: seven of the thirteen classes received between K1.6 billion and K3.5 billion whatever they were owed, so the share paid falls with the size of the bill. The nearest thing to a blind test is the Treasury’s own 2024 sorting of what was still owed into five categories, and it fits the shares best: category 1 for its “considerable social economic impact”, category 5, fuel, “the most costly category to the Government”, numbered last and refinanced first. Two cautions. The categories were written in August 2024, after the payments in the table; and a Treasury describing good triage and a Treasury paying in the order of what hurts it write the same list.

As a creditor, the state already shows which promises bind here: from the mill, the public systems that improved most are the ones that collect by deduction or before the service, promises that execute themselves. Where it must chase, it does not: taxpayers owed the Zambia Revenue Authority (ZRA) K71.3 billion at the end of 2025, and the Citizens Economic Empowerment Commission (CEEC), the state’s enterprise fund, had no repayments by the end of 2024 on K394 million lent.

One correction to the question itself. “Who owns the step” is rarely answered by “the state”. The state is not one organisation with one will. It is the Treasury, line ministries, agencies, the Bank of Zambia and a governing party, each bound by a different cost. The Food Reserve Agency (FRA), the state grain buyer, shows it across its 2025 and 2026 seasons: the Treasury wrote to the IMF, the state’s lender under its 2022 to 2026 programme, that the FRA would buy at market-based prices and be funded only within the budget; the agriculture ministry and the FRA bought four times the plan. Each kept the promise that would have cost it most to break. An outside sanctioner binds the part of the state it measures, the Treasury, and not the parts it does not. So the question for any chain is not whether the state will keep a promise, but which part of the state owns the step, and what breaking it costs that part.

What the table cannot show, pairs can: the same creditor, with only the cost of breaking changed.

04Same creditor, different costHold the creditor fixed and move the cost. The promise moves with it.

If cash alone decides, a promise moves with the cash. If the rule is right, it moves with the cost of breaking it, even when the creditor stays the same.

The farmers, two seasons

Same buyer, same farmers. In 2025, July deliveries were still unpaid in December. The minister cited a bumper harvest and too few private buyers; the FRA owed farmers more than K7 billion and settled with a K5 billion bank facility secured on its own maize. In 2026, an election year, the FRA was told in July to pay on the spot. By 28 August, two weeks after the vote on 13 August, it had bought over a million tonnes and paid over K6 billion, most of what it then owed: a million tonnes at K347 a bag is about K6.9 billion (this page’s arithmetic). One season pair is not a pattern, and the planting season binds too: the 2021 arrears to farmers were paid in full in 2022 to 2024, with no vote near. The Fund’s published reading, in January 2026: the FRA’s purchase prices “are typically set just below prevailing market price levels, but have been set above market prices in the last three elections cycles”, and “when budget allocations are insufficient, the cost is shifted to farmers through arrears accumulation”. For 2026 the FRA’s board promised to process payments “within three days of grain delivery”. One unpublished observation, the author’s own from the maize trade in Southern Province and to be weighed as such: within weeks of the vote, farmers were complaining of non-payment again.

The same season shows what a target is worth. In its letter to the Fund in late 2025 the government wrote that it would fund the FRA only in line with budget commitments, and that it would buy “at market-based prices”; its agricultural transformation programme proposes reviewing the FRA’s mandate “with a view to reverting” to reserves and last-resort buying. In May 2026 the FRA planned at least 500,000 tonnes; by the close on 30 September it had bought over 2 million. The budget for maintaining strategic food reserves was K2.1 billion; 2 million tonnes at K347 a bag is about K14 billion (this page’s arithmetic). A firm that planned on the announced step back planned on a target. The letter was the Treasury’s promise and the buying was the ministry’s and the FRA’s; each kept the promise that would have cost it most to break.

Salaries and allowances

The state’s own staff are one creditor with two promises. In September 2025 the finance minister said central government salaries had been paid “consistently and on time” since 2021, with one exception, the August 2025 payday, which he put down to a technical glitch; that is his claim, and no counter-claim was found. Allowances owed to the same staff were 20 percent paid. Every household sees the payday; an allowance is a claim one person makes alone. Salaries paid through grants bind less: through 2024, staff at the University of Zambia (UNZA) were paid late because the monthly grant came late.

Pressure from outside

The IMF programme of 2022 to January 2026 was the largest outside cost placed on the state, and it bound what it measured. New foreign arrears sat under a performance criterion that could stop a disbursement; domestic arrears under an indicative target, which cannot. The target was missed at June 2025 and met at September, because the fuel operation counted. In 2023 the cash balance improved “including on the back of accumulation of expenditure arrears”; when foreign debt payments resumed in 2025, goods and services were cut by about one percent of GDP. The programme ended in January 2026; talks on a successor opened on 30 September. Between programmes, even the target has no anchor.

The Zambia Medicines and Medical Supplies Agency (ZAMMSA), the state’s medicines buyer and distributor, shows what outside pressure can do, and what it cannot. Its promise is plain: medicines bought for clinics reach clinics. In May 2025 the United States cut US$50 million a year of medicines aid, citing systematic theft since 2021 and a forensic audit that, more than six months in, had not been published. The audit’s findings reached the press two months later. In October 2026, after audits found procurement at inflated prices, unaccounted-for stock, missing system records and medicines diverted from public health facilities, the government separated all 150 staff of ZAMMSA and said they would receive their dues in accordance with the law, the Employment Code that binds it as it binds any private employer. For years before any of this, the patients bore the cost of the broken promise. Seventeen months passed between the outside cut and the separation, and the timing alone does not show what caused it. Changing the people resets the step. It does not make the next team’s promise bind.

One grid, two kinds of bond

In the 2024 drought, one grid, ZESCO’s, kept two promises differently. The mines were largely protected and imported power; households and small businesses absorbed load-shedding of up to twenty hours a day, for months. A mine’s lost output shows in next month’s export figures. A small firm’s shows nowhere.

When the cash ran out in 2020, the state defaulted on its dollar bonds and kept every kwacha bond and bill paid. The kwacha paper is held mostly by the country’s own banks and pension funds; the Treasury’s stated reason for keeping it out of the restructuring was “to preserve financial stability and ensure a well-functioning domestic debt market”. Two different things happened at once. The default itself is the plainest case where cash beat cost: four years outside the market was a known price, and it was paid. Within the shortage, the order held: of the two promises, the one whose breaking would cost most at home was the one kept. The dollar bonds have been paid on schedule since the 2024 exchange.

The state keeps the promises that can hurt it. Change what breaking one costs, and the same creditor is paid.

05Who paysThe creditor carries the wait, and the Treasury saves on it

When a promise breaks, somebody carries it. Verified domestic arrears stood at K84.1 billion at the end of August 2025 on the 2026 budget speech, up K3.3 billion from December 2024; how that figure sits beside the Treasury’s own K88.7 billion of March 2024 is in the numbers table. At the Treasury’s own borrowing costs in October 2025, 13.9 percent on bills and 17.7 on bonds, carrying K84 billion instead of borrowing it saves about K12 billion to K15 billion a year in interest (this page’s arithmetic: 84.1 times 13.9 percent is K11.7 billion). That is more than twice the K5.5 billion the 2026 budget gives to clear arrears (K4.7 billion) and pay court awards (K0.81 billion).

The saving is on the classes with no interest clause: ordinary suppliers, refunds, staff claims. On the classes with penalty clauses the Treasury’s own account shows the opposite. By the end of 2023 the fuel suppliers’ interest and penalties exceeded the principal, and the bank loans that cleared them, at under 13.25 percent, were “notably lower than the penalty rates previously incurred”: on those bills, waiting cost more than borrowing. The saving on the rest holds while suppliers do not price the wait into their bids. Where they do, the state pays for its arrears in higher contract prices, spread thin and hidden. Either way the firm carries it first, at the 29 percent banks charge, if a bank will lend. In Europe, once late-payment rules were tightened, firm exits fell most in the sectors that sell most to government.

The arrears strategy had promised at least K6.8 billion a year across all its legacy classes, roads and fuel refinancing aside. The budget’s named arrears line is K4.7 billion; whether other lines carry the rest is not shown in the speech. It was the Treasury’s own target, in this page’s sense of the word: no counterparty, and no stated cost for giving less. Over the 27 months of the Treasury’s table, K17.7 billion of old bills were paid on the twelve domestic classes and K13.9 billion of new ones created on the same classes. None of this needs bad intent. Nobody ranks suppliers last. Nobody has to.

Winning in court does not move a creditor up. The law bars execution or attachment against the State; judgments are paid from the Compensation Fund as money allows. In a 2022 constitutional case over costs agreed in 2009 and still unpaid, the State told the court it had lacked the funds, and that judgment debts going back to 1999 were unpaid too. The court dismissed the petition, and urged that the Fund be given enough to clear judgment debts promptly.

Sometimes the bill is moved rather than carried. In 2025 the FRA borrowed against its maize to pay farmers, and the fuel arrears were cleared with a bank loan. Both creditors cost the state something to keep waiting. The ordinary supplier costs it nothing, so the ordinary supplier waits.

A sale that is paid in a year was not a sale. It was a loan to the Treasury at zero.

06The radiusWhere no court is used, a promise reaches as far as its owner can walk

Where a small firm cannot use a court, a promise binds only as far as the firm can reach. Enforcing a contract in court took 611 days and cost 38.7 percent of the claim, on data to 2019. So a bank lends to a small firm against a title deed, or against a salary it can deduct at source: a promise it can execute itself. Studies of firms where courts are weak find the same. Trade credit goes to customers known through networks, dealt with for years, or with nowhere else to buy; firms that trust the courts extend more of it.

From the mill: it supplies on credit only to established companies, and only in Southern Province and Lusaka, where it has people. One neighbour buys on a handshake. That is the radius. Credit goes where the owner can walk over, to a counterparty who will be there next season. Beyond the radius, cash. Making things means crossing the radius, because a long chain runs through strangers. Trading stays inside it.

And the radius shrinks from the top. A firm given no credit extends little; its buyers, given little, extend less. In the mill’s words: “we prefer cash because we do not have access to credit, and that limits our growth.” The share of registered firms with a bank loan fell from 10.1 percent in 2019 to 3.1 percent in 2025; banks paid for 0.8 percent of what firms invested. Lusaka and the Copperbelt took 80 percent of the loans made in 2025.

Lending to the state competes with lending to firms. A fifth of all bank assets sits in government paper, and the sector earned a return on equity of 30 percent in 2024. At one of the largest banks, government securities out-earned all its loans in 2024, a year after loans had earned more. NAPSA, the national pension scheme, holds nearly half of its K87 billion in government paper. The worker’s saving lends to the Treasury at 14 to 18 percent; the firm that employs the worker borrows, if at all, at 29. The Bank of Zambia’s own research found that the more a bank lends to the government, the less it lends to everyone else. The arrow may run both ways, since small firms a bank cannot read are hard to price. Either way, the bank that lends to the Treasury instead of the mill has shortened every chain below the mill.

Credit goes where the owner can walk over. A record that travels lengthens the walk.

07What makers doFewer promises from others, and the ones that remain cost more

The makers that survive need fewer promises from others, or make the ones they rely on execute themselves. This describes survivors. It is not advice, and the mill is one specimen.

They own more steps. The mill makes, packs, delivers and sells through its own people, so each of those promises is to itself. They make their own steps cheap. The roles a forty-person firm never had, the accountant, the analyst, the compliance clerk, now cost electricity and an owner who can steer them. Those roles were weak steps, because the firm did without them, so fixing them gains a lot. And by the arithmetic of section 01, the stronger a firm’s own steps, the more each remaining broken promise costs it. Software widens the gap this page describes, and raises the return on closing it: a firm whose own steps are strong can see which promise from outside is costing it most.

They sell for cash, or on short fixed terms, and walk away from buyers whose promise does not bind, even when the buyer is the state and the volume is tempting. They pay their own suppliers on a fixed day, because a promise that binds is worth money to the other side: Kenyan dairy farmers accept a lower price from a buyer who can be trusted to pay monthly in a lump sum. What a firm cannot move on its own is the subject of the next section.

08What this makes possibleThey exist. They can be found. Some can be fixed.

The weak links are there. This section says what they are worth, what has become cheap, and whose hands each one needs.

The country adds about 700,000 people a year on the projections of ZamStats, the statistics agency. The current plan targets a million jobs over five years; the World Bank puts the need at more than ten million by 2050. They will probably not come mainly from factories in zones: across Africa, large modern firms have raised output without adding many jobs. The jobs are probably in the middle of the chain, where the most promises have to bind.

The value is off the farm. The people are on it. The distance between the two is the chain. Across districts in ten African countries, places with more salaried jobs that sell beyond the district have more local jobs too: for Zambia, about 3.7 for each. How many one promise made to bind would bring is not known.

The return on a point at a weak step is the value of everything behind it, so the map of weak links is also a map of returns. A decision is cheap to change; a dam is not.

What software moves

Software does not by itself create a cost for breaking a promise. But three of the four sources of that cost, visibility, self-execution and exit, are now things a small firm or a single office can build, and the first of them needs no new law. Visibility: the record of who kept which promise now costs little to keep, and a promise broken in public costs something without a court. Self-execution: interest that accrues without a claim, escrow that releases on delivery, a warehouse receipt that is the collateral. Exit: a firm whose record travels with it can leave a bad buyer and be trusted by the next; in one Caribbean lender’s data, phone use predicted repayment about as well as a credit bureau did. Software does not move a court. It does not move cash.

Software that anyone can copy is being repriced towards zero. What keeps its value is software tied to a step that did not get cheaper: power, roads, borders, a licence, a buyer who pays on a date, a record. From the mill, e-GP, the public purchasing system firms must use to sell to the state, is the hardest public system to use: a step that did not get cheaper. Across 187 countries, stricter procurement rules go with better outcomes only where public offices are weak, which is where Zambia sits, so a written rule may bind here where a better interface would not.

The opportunity is in records. AI can make records cheap to keep, readable in the languages people use, and useful to a forty-person firm.

Each of the four costs that make a promise bind is placed by someone other than the promiser: a court by the state, exit by firms that can leave, visibility by whoever measures and publishes, self-execution by whoever builds the system, which can now be one person with a laptop. No one of them can place all four, which is why the weak steps on this page sit where two owners meet. Each case below says which hands it needs.

None of this needs a committee. Each step where two owners meet needs one promiser, one creditor and one person counting, because a step that everyone owns is the step nobody pays for when it breaks. And the promiser is an office, not a sector.

The mill is the specimen of section 07. It pays its suppliers on a fixed day and keeps its own records (the box there), and the step it could not move, when a buyer pays, is the first case above.

The next section reads the World Bank’s firm survey as partly a record of late payment. If firms that sell to the state name finance no more often than firms that do not, this page withdraws that reading.

The weak links are there. Some can be found with public numbers and a few hundred scored chains, and this page is a first, partial map of where effort would probably pay. Some can be strengthened without anyone’s permission. Within a year we will know whether that holds for one payer, one receipt and one language, and the next version of this page will say which.

09How this page could be wrongThe survey evidence against it, and five tests

The rule is a claim, not a finding, and the best survey evidence cuts against its emphasis. Between October 2025 and May 2026 the World Bank asked 608 registered Zambian firms to name their single biggest obstacle. They named finance and power, not late payment by the state.

Percent of all firms, as the published profile rounds them; licensing and permits are not among the ten obstacles it charts. Small firms name finance more often, 33 percent. The 2019 survey gave the same first two. Nine firms in ten had power cuts, 21.6 a month on average. Informal firms, the smallest firms and farms, most of the people this page is about, are outside the survey.

This page reads the finance answer as partly where other broken promises land: a firm waiting on a public payment reports a finance problem, because the survey allows one answer. That is a reinterpretation, and it can be checked. The first test: the survey asks each firm whether it secured or tried to secure a government contract, so cross that with the obstacle named. Four more tests could settle it within a year:

1. Recode the cost column of section 03 blind. Someone who has not seen the payments scores each of the thirteen creditor classes on the five tests. If the blind coding does no better than the size of the bill, the order is cash, not cost. The Treasury’s own 2024 categories are a first check (section 03).
2. How many days does one public payer take, from approved invoice to money? No office publishes it. A monthly series beside the Treasury’s cash balance would show whether the weak creditors wait even when cash improves: the rule predicts they do, cash alone that they do not.
3. What kills firms between their first year and their fourth: the market, the promises public bodies break, or a competitor who skips the obligations the formal firm pays for? In Choma, a small mill and a sunflower-oil press closed in the last two years. Nobody has yet asked them why.
4. How many broken public promises are the clinics’ kind, which no decision could keep? If most are, the argument narrows to money.

The tool at the end of the page (section 12) collects chains from the people who run them. It gathers hypotheses, not proof, and the people who run chains are the only ones who can name the steps. Where they blame a step they do not own, that is itself a claim worth checking.

10The numbersThe country, in figures with sources

Everything on this page stands on these numbers. Each is public, dated, and linked to its source.

One figure is older than it looks: the entrepreneurship survey in section 01 dates from 2013, and the text says so.

11The theoryThe papers and the models behind the argument

For readers who want the history and the maths. Each paper is one card: what the original work did, its model in the simplest form that keeps the point, and what it means here. Where a calculation is this page’s own rather than the paper’s, it is marked own illustration.

12Your turnScore a chain you know, or take one of three projects

This is where the finding starts. Chains named and scored by the people who run them, by the hundred, would turn this page’s conjecture into a map.

Pick a chain you run or know well. Name its steps in order, and score how well each one works. Then, for the weakest, answer the three questions: how much it holds back, who owns it, and whether the owner pays when the promise breaks; and one more, whether a decision could fix it. Start from an example or from blank. Nothing is sent until you press send; your draft stays in this browser.

13Sources and changesEverything this page stands on

A verified evidence base of about 170 entries sits behind this page; what is cited is listed here. This page as a PDF. Corrections to chris@zamai.pro.

Cite as: Lindinger, C. (2026). Weak links: the Zambian economy, step by step. Working draft v6.1, 4 October 2026. zamai.pro/weak-links

The framework, the coding and the judgements are the author’s. Drafted with the help of an AI system for research and editing.