Growth without headcount

By four o'clock on a Thursday afternoon, the quote desk at Meridian Freight had 212 unanswered requests in its inbox. Some came from customers the company had served for a decade. Others came from businesses it had never worked with, asking for a price to move a container from Shanghai to Felixstowe. The eighteen people on the desk would work through most of them by the end of Friday. Nearly a third would never get a reply.

Nobody on the desk was idle. Each quote meant checking carriers' rate sheets, the customer's history, the season, the space on the next sailing, and a margin that one of the senior pricers carried in their head. The desk was working as hard as it could. It was also the largest single brake on the company's growth.

Meridian is a UK freight forwarder and customs broker. It employs 400 people, handles 100,000 shipments a year, and turns £200m of revenue into £40m of gross profit and £10m of EBITDA. A private equity fund had bought it a year earlier, with a plan familiar to anyone in the sector: grow organically, buy smaller forwarders, and sell the combined business within five years at a higher multiple.

The first year exposed a flaw in that plan. Meridian could not grow without hiring, and it could not hire fast enough. What follows is how the company removed that constraint, and what changed as a result: the operating model, the P&L, the balance sheet and the equity story.

Growth was a hiring plan

Forwarding is a people business, and Meridian's economics showed it. People cost £22m a year, 55% of gross profit. Each 10% of growth needed roughly 10% more staff, hired months ahead of the volume and fully productive only after six to nine months. Of each extra pound of gross profit, about 40p reached EBITDA. The business had almost no operating leverage.

The knowledge that made it work was concentrated and mobile. Most of the pricing and customs expertise sat with about fifteen people. Turnover in operations roles ran at 20% a year. Every resignation took know-how out of the building, and every replacement took most of a year to rebuild it.

There was also a cost that never appeared in the accounts. Early in the programme, the team gave the same five shipment requests to five experienced pricers. The prices that came back were up to 22% apart. None of the pricers was wrong by their own standards; each applied their own experience. Across 100,000 shipments, that variance showed up as deals lost to a high price and margin given away on a low one.

Money leaked at the other end of the chain. Extra charges such as waiting time and storage often never reached the invoice, costing 2% to 3% of gross profit. Invoices went out days after delivery, and customers took 55 days to pay.

The fund's value creation plan assumed EBITDA would double before exit. On the existing model, that meant growing gross profit by more than 60%, and headcount with it. That was neither affordable nor, in a tight labour market, achievable. Each acquisition would make it harder, bringing more people, more systems and more habits to integrate.

The constraint was judgement

The obvious response was to automate tasks. Meridian had already tried: a better inbox tool, a new transport management system, a chatbot for tracking queries. Each made documents move faster. None changed the economics, because none touched the decisions the work was organised around.

Every shipment at Meridian carried more than twenty decisions. Whether to quote, and at what price. Which carrier, on which route. How to classify the goods for customs. Whether a delay needed action, and what action. What to invoice, and how much credit to extend. The workflows existed to gather information for those decisions, route it to someone qualified to make them, and act on the result. The queues, the hand-offs and the approvals were scaffolding around judgement.

For thirty years, a business could automate only the decisions it could reduce to a rule. Anything that rested on a document, a precedent or an experienced opinion stayed with a person. That limit has gone. Models can now read a request, weigh context, apply precedent and explain their reasoning. For Meridian, the question was no longer which tools to buy. It was which decisions to move, in what order, and how to prove the value in the accounts.

Mapping the decisions

The programme followed the five phases of Sapionic's method: Map, Readiness, Plan, Deliver and Embed (Exhibit 1).

It began with a Decision Discovery workshop for the leadership team, followed by six weeks mapping the work. The team catalogued 140 recurring decisions across six stages of the value chain, from quote to cash, and sized each one against the EBITDA drivers it moved: volume, price, cost to serve, leakage and working capital. The distribution was steep. Twelve decisions carried most of the value, and the pricing of quotes shaped more of the gross profit than any other.

In parallel, the readiness review covered technology, data, legal and organisation. The findings were typical. Shipment data sat in four systems. Carrier rates lived in some 900 spreadsheets. Three years of shipment history was usable once cleaned. There was no AI policy, and customs decisions needed an audit trail that HMRC would accept. None of this stopped the programme. It set the order: foundations first.

Before anything was built, each priority decision was tested against history. Given last year's quote requests, could a system have priced them as well as the best pricer? On the standard routes it could, and more consistently. The plan ranked the twelve decisions into three delivery waves, gave each a named owner, and set out how autonomy would be earned: the system would first recommend, then decide within limits with people approving above them, then decide alone where the evidence supported it. The investment case was £7m over two years, with payback inside the second year. Every benefit was tied to a line in the management accounts before the board approved a pound.

Engineering one decision

Most AI programmes go wrong in delivery because they build before they understand. Every decision at Meridian went through the same four engineering steps: Codify, Rewire, Build and Compound (Exhibit 2). The quote shows how.

Codify. The team spent three weeks with the six best pricers, making explicit what they did implicitly: the factors they weighed, the rules they never broke, the precedents they reached for, and where they disagreed with each other. Three years of quotes, won and lost, supplied the evidence. The result was a decision architecture and a knowledge base that Meridian owned: route histories, carrier performance, customer terms, pricing rules and the reasoning behind them. This is the step almost nobody does, and the one that decides whether the rest works.

Rewire. The workflow was then redesigned around the codified decision. The system would price standard requests end to end, from the customer's email to the quote. People would handle key accounts, new routes, unusual cargo and anything above a margin or value threshold. The quote desk became a pricing team. Its job changed from producing prices to improving the decision that produced them.

Build. Agents were built into the systems Meridian already ran: the transport management system, the customer portal and the shared inbox. For six weeks the system ran in shadow, proposing a price for every request while people made the call. When its prices matched or beat the best pricers on win rate and margin, it took over the standard routes. Every quote carried its reasoning and went on record.

Compound. The knowledge built for pricing, above all carrier performance and customer history, fed straight into the next decisions: carrier allocation and billing. The second decision took about half the time of the first. By the third wave, most of what a new decision needed already existed.

Quotes that had taken 24 to 48 hours went out in under five minutes. Every request received a reply. The win rate on quotes rose from 18% to 24%, and each price was set on the probability of winning and the margin, not on habit.

Scaling decision by decision

The remaining decisions followed in three waves, each decision taking six to ten weeks from Codify to production. Quoting and billing went first because they paid fastest: more revenue and fewer unbilled charges. The second wave moved bookings, documents and shipment exceptions. The third took on customs classification, carrier allocation and credit limits, the decisions that needed the strongest controls.

By the end of year three, the system made about 80% of the decisions in Meridian's core workflows. People handled the rest: key accounts, carrier negotiations, new routes and the exceptions that need judgement.

The hardest part was not technical. Automating 80% of decisions does not remove 80% of the people. Roles are bundles of tasks, the work that remains is the hardest, and new roles appear. Meridian finished year three with 250 people, not the 100 that a spreadsheet would suggest. It got there without a redundancy programme. A hiring freeze on operations roles, combined with 20% annual turnover, did most of the work. Thirty new roles, most filled from inside, owned and improved the decision systems. The best pricers and customs specialists became decision owners: their judgement now ran on every case, and they were accountable for making it better.

Five disciplines ran through every phase, from the first workshop to the last release (Exhibit 3).

A different operating model

Three years on, Meridian operates differently (Exhibit 4). Before, people sat in the middle of every workflow, moving information between systems and making the decisions from experience. Now a decision system sits in the middle, and people work at the edges: with customers, with carriers, on the exceptions, and on the system itself.

Sapionic describes the target operating model through five properties, the 5S: Speed, Scale, Skills, Science and Scrutiny (Exhibit 5). Each one changed a line in the accounts.

What it did to the P&L

Gross profit rose from £40m to £55m, from three sources. Volume grew by a quarter, because every request was answered within minutes and more of them were won. Gross profit per shipment rose 8%, because prices and carriers were chosen by models built on the company's own history. And 2% of gross profit that used to leak through unbilled charges was recovered.

Costs moved the other way. People costs fell from £22m to £16m while volume rose by a quarter. The decision systems cost £4m a year to run. EBITDA rose from £10m to £27m (Exhibit 6), and the share of gross profit reaching EBITDA, the ratio forwarders watch most closely, rose from 25% to 49% (Exhibit 7).

The path was not a straight line (Exhibit 8). The £7m investment pushed EBITDA down to £7.8m in year one, and the programme had paid for itself during year two. Meanwhile volume and gross profit rose as headcount fell. The case does not depend on everything going right: at half the modelled benefits, EBITDA would still have risen by about 60%.

The cost structure changed shape as well as size. Before, costs rose with volume. After, most costs were fixed and each extra shipment cost very little. A fixed cost base is harder to cut in a downturn, but Meridian's is now small relative to its gross profit. Even treating every cost as fixed, gross profit could fall by 49% before the business stopped making money, against 25% before.

What it did to the balance sheet and the equity story

For a forwarder, the largest balance sheet item is the money owed by customers, because it pays carriers and duties before its customers pay it. Invoicing at each shipment milestone cut the unbilled work from ten days of revenue to three. More accurate invoices meant fewer disputes, and customers paid in 45 days instead of 55. Working capital fell from £20.3m to £14.2m while revenue grew by a quarter: from 10.1% of revenue to 5.7%. Growing on the old model would have tied up £11.6m more.

The same £40m of debt that had been 4.0 times EBITDA at the start was 1.5 times EBITDA by year three, without a pound repaid. That leaves room to fund acquisitions. At an illustrative multiple of 8 times EBITDA, enterprise value rose from £80m to £217m, and the equity from £40m to £177m: 4.4 times, before cash generated, fees or any change in the multiple. The business needs less capital and is worth far more (Exhibit 9).

Put the P&L and the balance sheet together and the change shows up in the measure investors watch most closely: return on invested capital (Exhibit 10). ROIC is the product of two things: the profit after tax left by each pound of revenue, and the revenue supported by each pound of capital. The programme moved both. The after-tax margin rose from 3.0% of revenue to 7.3%, because gross margin rose and operating costs fell from 75% of gross profit to 51%. Capital turnover rose from 7.1 to 11.3 times, because working capital fell while revenue grew. Return on operating capital rose from 21% to 82%.

For the fund, the more telling figure includes goodwill: the £52m it paid above the operating assets. On that basis, Meridian earned 7.5% before the programme, below an illustrative 10% cost of capital. At the price the fund paid, the business did not cover its cost of capital. By year three it earned 25%, two and a half times that cost, and its economic profit moved from a loss of £2.0m a year to a gain of £10.9m.

No higher multiple is assumed, although there is a case for one: the business now earns well above its cost of capital, scales better and carries less risk. Part of that case rests on an asset the balance sheet does not show. The most valuable thing Meridian built is not on it. Accounting rules do not allow a company to record its own know-how, data or decision logic as an asset. Before the programme, Meridian's key asset was not recorded and not owned: it could resign. Now it is still not recorded, but Meridian owns it, controls it and can prove it. The due diligence question changes from "What happens if your top fifteen people leave?" to "Show us the decision logs."

That asset keeps compounding. In year three, Meridian bought a smaller forwarder in the Midlands. Its quoting, billing and customs moved onto the decision system in four months, and its data made the system better for both businesses. (The acquisition is not included in the figures above.) Each new acquisition becomes cheaper to integrate than the last, and more valuable once it is.

Lessons for leaders

Meridian is fictional. The pattern is not. Four lessons apply to any business whose growth depends on expert judgement.

Start with the decisions, not the tools. Automating tasks makes documents move faster. Moving decisions changes the economics. Map the decisions inside your workflows, size them against EBITDA, and let the workflow follow. Get the decisions right and the workflow redesigns itself.

Measure the variance before you build. The same case decided differently by different people is the defect nobody measures, and often the most expensive. Baseline it first, or you will never prove the benefit.

Make the value reach the accounts. Efficiency that never reaches the P&L is not value. If you cannot trace a saving to a line in the accounts, assume it is not there.

Let the experts own the system. The people whose judgement you codify are the people who make it work. Give them ownership of the decision, and their expertise stops walking out of the door.

One warning. As more companies in a sector make this move, the cost advantage will be competed away to customers as lower prices. The first movers keep the margin; the late movers pay for the price war. What lasts is the Science: better decisions, and a growing record of decisions and outcomes that no competitor can buy.

Before, Meridian was a labour business with a software cost line. Three years later, it is a software business with a labour cost line, and its people own the customers, the carriers and the judgement inside the system. The question for any leadership team is where the judgement sits in their own business, and how much growth it is holding back.

More insights