Monoculture

Pete Harvey

Meet Tom

Tom runs a haulage business. Forty trucks, 80 drivers, a yard in the midlands, and £17m of revenue. He cares about his numbers and he runs a tight operation.

He has no idea which of his services is profitable.

Not because the data doesn’t exist. It does. Jobs are logged in one system, invoices go out through Xero — except those for his biggest customer, who insists on Tom using their billing system. Trucks off road for maintenance live in a maintenance schedule in Excel, and sales are booked through a CRM system. Five systems, five partial views into his business. Nobody has been able to build a single integrated view.

Tom had his worst ever January on record this year, but he only found out in March. The investigation into the causes took two weeks and never really led anywhere.

Fifteen of the jobs booked for next month will run at a loss. Fuel prices have risen 8% month-on-month for three months and his prices haven’t kept up. On the days of those jobs, his team won’t have capacity to complete far more valuable work for his largest customer. Tom instructed his management team to reprice. They haven’t had time to update his systems. He doesn’t know any of this yet.

Tom is every small business owner

Tom isn’t one person. He’s every small business owner that I’ve spent time with in the last 12 months. The details change but the shape of the problem doesn’t.

Tom’s business has become too operationally complex to run from spreadsheets, disconnected systems, and human memory alone.

A pattern across every sector

This isn’t a haulage problem.

In 2024, the UK lost 30 independent shops every single day.¹ The hospitality sector is 14% smaller than it was in 2020, with 62 net closures every month.² In 2013, 90% of vet practices in the UK were independently owned. Today that figure is 40%.³

Meanwhile Tesco is trialling personalised pricing based on customer behaviour data.⁴ Gail’s uses an algorithm to identify optimal sites to open — where there are weak independent cafes to compete against.⁵ The largest veterinary group in the country has partnered with Microsoft to embed AI across its operations.⁶

The pattern is the same in every sector. The corporation had access to the right data and knew how to use it. The independent small business did not.

The engine of the economy

This matters because small businesses are the engine of the economy. In the UK alone there are 5.7 million of them. 99% of all businesses, employing 60% of the workforce, generating half of all private sector revenue.⁷ The numbers are similar across every developed economy.

And yet small business labour productivity is half that of corporations. Every hour worked in a large company generates twice the revenue of an hour worked in a small one. McKinsey estimates that closing that gap could add 5% to GDP in advanced economies. In the UK that is £115 billion in additional output.⁸

The policy environment makes things worse. Successive governments see small business as an easy target, most recently using the 2024 budget to raise national insurance contributions and cut the thresholds. Corporations were able to absorb this double blow in a way that small businesses couldn’t. Policies come and go, but advancements in technology are inevitable.

The AI adoption gap is compounding

The Microsoft Copilot adverts shown during Premier League football might make you think that small businesses are using AI every day.⁹ They are not. The OECD found that the adoption gap between large and small firms is already wider for AI than for any technology that came before it — including the internet and cloud computing.¹⁰ And the gap compounds. The more a company uses AI, the better its data gets, which makes the AI more effective, which drives further adoption. The corporations already winning on productivity are the ones racing ahead with AI. The small businesses already behind are falling further back every day. The SME productivity gap is massive, and growing.

The obvious objection is the decline of small businesses has nothing to do with data. Sector mix does explain some of the productivity gap; a haulage firm generates less revenue per hour than a law firm. But McKinsey finds that there are some sectors like administrative services (office cleaning, facilities management, security etc.) where small businesses are actually more productive than corporations. Sector mix alone is a poor explanation of the productivity gap.

Economies of scale explain the gap much better. Large corporations spread fixed costs across thousands of employees, buy at prices small businesses can’t negotiate, and can afford the legal, finance and compliance functions that smaller businesses can’t. These advantages are real and mostly permanent, but they don’t explain Tom’s problems.

Private Equity isn’t the cause either. Blaming PE for the SME productivity gap is like blaming the chainsaw for deforestation. PE targets fragmented industries with few dominant corporations, it doesn’t create the fragmentation. Something else does.

The mundane root cause: broken data

The real reasons are obvious to anyone who has run a small business. Small businesses can’t access capital on fair terms. They can’t automate their operations using technology. And the instincts that built these businesses — sharp, experienced, hard-won — are being outgunned by competitors who have turned those same judgements into algorithms. Three separate problems. One mundane root cause.

Every system in Tom’s business speaks a slightly different language. None of them recognise each other. The people who once translated between them have left, retired, forgotten, or are too busy keeping the place running to explain it all again. So the business drifts slowly out of understanding, one spreadsheet at a time. This stops Tom from automating processes. It stops him from building compelling cases for raising capital to expand.

Small businesses are disappearing because managing their broken data has become something that is too complex for a small business leader.

If this is true, why haven’t small business leaders fixed it?

Corporations have spent a lot of money trying to fix it. Data engineers have been solving these problems for decades. The tools exist. The methods are well understood. Tom’s data is not broken because nobody knows how to fix it. It is broken because every solution requires expensive software and expensive experts to fix it.

Let’s take the common occurrence where Xero has one customer name (ACME Corporation) and a CRM has another (A.C.M.E Corp). They are both the same client company, but which system is correct and how do you fix the underlying data without having to do it manually for every customer? There is already a free open-source tool that could identify these duplicate records automatically.¹¹ But using it requires technical expertise: connecting systems, defining matching rules, validating uncertainty, and safely merging the corrected records back into production software. The software is free. The engineer who can operate it costs £70,000 a year.

The 30-year project SMEs never completed

Corporations do have that engineer. In fact they have a team of them. So corporations have been working on fixing their broken operational data for more than thirty years. They gave the technology they use different names — databases, data warehouses, semantic layers, business intelligence infrastructure. But the underlying project was always the same: take the raw record of what the business does, and make it legible. To managers. To analysts. And now, to AI agents.

SMEs never completed that project. Not because they didn’t need a solution, but because building it required a data engineer, a six-figure budget, and eighteen months of design work. Those are enterprise-only resources. At the SME level, they have never been accessible.

AI alone is not the answer

AI can help. Tom could ask an AI to look at his customer list and suggest which records might be duplicates. It would probably find the two accounts for the same customer that have been wrong for three years. For a one-off investigation, that is genuinely useful.

But useful is not the same as fixed. The AI has not resolved the duplicates. It has not merged the records into a single customer list, recalculated the historical margin figures, or set up anything to stop the same problem happening again next month when someone adds a new customer with a slightly different spelling. Tom would still need the engineer. The AI just helped him find the work.

More fundamentally, AI starts from scratch on every query. It has no persistent memory of Tom’s business. Ask it the same question tomorrow and it rebuilds the answer from nothing, reading the same broken source data, potentially reaching a different conclusion. It does not accumulate knowledge about Tom’s business the way an operational data layer does. It reads. It does not build.

AI can give Tom an individual answer. What Tom needs is a system he can trust.

Homegrown fixes fall short

Some small businesses try to bridge the gap themselves. A fractional CTO spends two days a week building and maintaining something in Airtable. A clever ops manager stitches together Zapier automations that mostly work. A spreadsheet that started as a quick fix becomes the closest thing the business has to a data warehouse, maintained by whoever has time, understood by whoever built it. These solutions are not nothing. They are evidence that the problem is real and that SME owners are resourceful enough to attempt it. But they are also brittle, undocumented and dependent on individuals. The Airtable base can only be updated by the fractional CTO. The Zapier automation is unauditable. The spreadsheet is unclean. Tom has seen three of these built and abandoned in the last decade. Each one solved part of the problem and created new ones.

Enterprise software is not fit for use by small businesses. Homegrown solutions don’t cleanly solve the problem. And AI agents have their own set of new needs.

To make AI work, it’s not just good enough to store the data on its own. AI can’t rely on information that is stuck inside the heads of Tom’s team. To work autonomously, AI needs to understand what each datapoint means as it is using them. A traditional data warehouse isn’t enough.¹²

What a real solution looks like

A new type of solution is needed to fix broken SME data today.

The solution needs to sit on top of the software systems a small business already uses. It needs to connect to each of them, pulling data into a single place. The solution needs to do the hard work of unifying the data, without the need for technical expertise. The solution has to make it easy to clean the existing data, and to keep it clean automatically. Crucially, it doesn’t just store numbers — it stores what those numbers mean to Tom’s specific business. One agreed definition of every key metric and term, so that everyone in the company gets the same answer to the same question.

From there, the data is ready to use. Team members can get any number they need for their job without asking someone else first. AI agents can grab the data and context they need to complete work autonomously. Other systems can pull from it when they need to. There are fewer mistakes, faster decisions and trust in numbers.

Introducing Opbox

That’s why we’re building Opbox. To make enterprise-grade data infrastructure accessible to small businesses for the first time.

If we don’t help small businesses to close the productivity gap, we will end up with a broken economy.

A lesson from a banana

The banana you buy at a supermarket in the UK is the Cavendish banana. If you were to buy a banana in the USA, or in Paris, or in Melbourne, or in Timbuktu, it will almost certainly also be a Cavendish banana.¹³ Over time, it has become almost the only banana that is grown across the globe. Every banana you’ve ever eaten is genetically identical.

The Cavendish banana is facing extinction. A strain of fungus that causes Panama disease threatens to wipe out our one strain of bananas. It’s not the first time this has happened. The Cavendish banana was cultivated to replace the Gros Michel banana, which was wiped out by an earlier strain of the same disease.¹⁴

Ecologists call a single species that dominates an ecosystem a monoculture. At first glance monocultures look optimal. They are efficient to plant and grow. They enable huge scale.

But monocultures fail in ways that diverse ecosystems do not. They fail fast, and completely. It turns out that diversity is the immune system of an ecosystem.

Every high street is becoming a carbon copy: Gails, Costa, Tesco, Pure Gym. No independent coffee shops. Everything factory-produced and margin-optimised.

Three local vets replaced with one big Pets at Home. Dentists rebranded and no longer accepting NHS patients. Madri on tap in every pub.

The UK economy is perilously close to becoming a commercial monoculture. A small number of corporations, with advanced data infrastructure, are displacing thousands of independent businesses without it.

The independent cafe, the family vet, the regional haulier: these are not inefficiencies to be optimised away. They are the genetic diversity of a functioning economy.

When they’re gone, they don’t come back.

Notes

Kollewe, J. (2025, January 2). UK lost 37 shops a day in 2024, data suggests. The Guardian. https://www.theguardian.com/business/2025/jan/02/uk-lost-37-shops-a-day-in-2024-data-suggests ↩
UKHospitality. (2024). Later sector closure data — our response. UKHospitality. https://www.ukhospitality.org.uk/later-sector-closure-data-our-response/ ↩
Competition and Markets Authority. (2024). Veterinary services for household pets: Final decision report — Summary. GOV.UK. https://www.gov.uk/government/publications/veterinary-services-for-household-pets-final-decision-report/summary-of-the-final-report-html#our-findings ↩
Roderick, L. (2024). Tesco launches new hyper-personalised ‘Your Clubcard Prices’ trial. The Grocer. https://www.thegrocer.co.uk/news/tesco-launches-new-hyper-personalised-your-clubcard-prices-trial/700701.article ↩
Gail’s founder: The 17 hardest business lessons no one tells you. Acquired Podcast. https://podcasts.apple.com/gb/podcast/gails-founder-the-17-hardest-business-lessons-no-one/id1537239469?i=1000692965809 ↩
Microsoft. (2024). With Copilot agents, Pets at Home unleashes an AI revolution. Microsoft News. https://news.microsoft.com/source/emea/features/with-copilot-agents-pets-at-home-unleashes-an-ai-revolution ↩
Department for Business and Trade. (2025). Business population estimates for the UK and regions 2025: Statistical release. GOV.UK. https://www.gov.uk/government/statistics/business-population-estimates-2025/business-population-estimates-for-the-uk-and-regions-2025-statistical-release ↩
McKinsey Global Institute. (2024). A microscope on small businesses: Spotting opportunities to boost productivity. McKinsey & Company. https://www.mckinsey.com/mgi/our-research/a-microscope-on-small-businesses-spotting-opportunities-to-boost-productivity ↩
Microsoft. (2024). Microsoft Copilot — Do More [Advertisement]. YouTube. https://www.youtube.com/watch?v=oBqKKy_MTJ0 ↩
OECD. (2025). Productivity in SMEs and large firms. In OECD Compendium of Productivity Indicators 2025. https://www.oecd.org/en/publications/oecd-compendium-of-productivity-indicators-2025_b024d9e1-en/full-report/productivity-in-smes-and-large-firms_968cffa9.html ↩
Ministry of Justice Analytical Services. (2024). Splink: Probabilistic record linkage at scale. GitHub. https://moj-analytical-services.github.io/splink/index.html ↩
Databricks. (2024). Redefining the semantics of the data layer: The future of BI and AI. Databricks Blog. https://www.databricks.com/blog/redefining-semantics-data-layer-future-bi-and-ai ↩
The Future Market. (2023). Biodiversity and the banana. The Future Market. https://www.thefuturemarket.com/biodiversity-banana ↩
European Commission. (2023). Calling on natural defences to turn back the banana pandemic. Horizon Magazine. https://projects.research-and-innovation.ec.europa.eu/en/horizon-magazine/calling-natural-defences-turn-back-banana-pandemic ↩