Operational decisions that learn
Operations-heavy firms already hold the data they need to improve every quote, schedule, purchase, and operational process, to spot opportunities early where an adjustment will pay. Firms that build this into how they decide will gain an advantage that compounds.
Business as usual, only tighter. Mid-market, operations-heavy firms handle volatility well. Buffers, quick reactions and experienced people have carried them through unpredictable times, on margins that have always been tight. Conditions in the UK and elsewhere have now tightened further. Payroll taxes and energy prices have risen and interest rates have been kept high, and the pressures feed off one another: rising fuel costs push up inflation, which raises other costs and keeps rates high. Regulation and geopolitics add compliance work and less predictable supply. No board sets energy prices, Bank Rate or tax, and beyond the certainty that things will stay uncertain, little can be forecast. Management can respond only through what it does control: its operating structures, its processes and how it makes decisions.
Why now. The quality and timeliness of everyday operational decisions now sit at the centre of competitive performance. The data to improve them already sits in a firm's own systems, and the tools to bring it together have become practical and affordable. Few firms are yet using them this way. Firms that move first will build an advantage that is hard to catch.
The blind spots of being good at it
The problem of being good. Operational teams are used to rough patches and surprises, and they handle them well. Dealing with daily uncertainty is business as usual. But working from one problem to the next creates blind spots. Attention goes to whatever is loudest today, doing what worked yesterday, and the things that signal a change rarely make a noise.
Why blind spots? These firms aren't short of data. Finance, ERP, payroll, production and quality systems, energy meters, CRM and spreadsheets each handle their own area well. What they rarely have is a way to bring that information together to support more reliable business operations and improved decisions. Experience and judgement fill the gap. That works in a crisis, but it leaves three blind spots.
Change. Trends show up as margins thinning gradually, not as a crisis, so nothing prompts a response. UK energy prices rose almost 18% in the year to September, and the prices businesses pay for materials and fuel rose 8.3% (ONS). Yet it arrived one bill at a time, and no single invoice prompted a review. Problems also often appear only when two factors combine. A business can absorb a higher tariff, and it can absorb smaller customer orders. Each manager reports their own number as within tolerance, and the problem sits across their systems. The real implications can only be seen when everything is understood together.
Opportunities. The same pressures don't hit every firm equally. A rival with older lines may be hit harder by the same energy rise, leaving your cost per unit below theirs on one product family. Sales sees more enquiries, production sees a full line and finance sees a slightly better margin. Nobody puts the three together, and by the time the opening is obvious, a competitor may have taken it.
Repetition without learning. This is the costliest. Most operational decisions recur: quoting, scheduling, ordering stock, planning changeovers, handling late deliveries and timing maintenance. Each one leaves a record of what was decided and what happened, yet every occurrence is treated as a one-off. Quotes are never compared with actual job costs, lessons stay in one person's head, and the process runs the same way next time. The business has the experience but doesn't keep it.
Decisions that improve each time
Decision systems. A decision system deals with all three. It is a way of working first and software second, using AI, conventional software or nothing new at all. Today's AI can work across data from a firm's existing systems without replacing them, once definitions and data quality are good enough. It can apply the logic of statistical process control, familiar from quality management, to many decisions and signals at once rather than one process at a time. The serious gain comes when you adapt the process, not when you deal with the incident.
The system works as a four-step loop:
Detect: measures such as energy per unit, labour hours, input costs, lead times, enquiry rates and margin by product show what is persistently shifting, up or down, as opposed to noise.
Understand: each shift is linked to likely causes across data sources and sorted into what you control and what you don't.
Adapt: someone with authority decides whether to fix the problem, compensate for it or seize the opportunity.
Learn: every decision and its outcome are recorded and used to adjust the next one.
Lasting value growth. The last step is where the value builds. Take a firm that quotes 40 jobs a month. Its timesheets and ERP already hold the actual hours, but who compares the two systematically? A system that compares each quote with the actual hours learns which job types are consistently under-quoted, and by how much, and adjusts the next estimate. As an illustration: if a quarter of the work, worth £50,000 a month, is under-quoted by 8%, the firm loses £48,000 a year. If the gap halves each quarter, by the fourth quarter the loss is running at about £6,000 a year. The same applies to a hotel rostering staff against occupancy, a facilities firm pricing reactive jobs, and to schedules, purchasing and maintenance.
The hardest decisions gain most. The approach matters most for the hardest decisions: those that cut across teams and balance demand, stock, suppliers, capacity, cash and margin. When the evidence sits together, leaders see the trade-offs before they make them.
This is why the gains compound. A process that improves slightly every time it runs, hundreds of times a year, pulls ahead of a competitor that hasn’t improved. A business that learns from repetition is also best placed to notice when the pattern itself starts to change.
Where most firms are
Most firms aren't doing this yet. About one UK business in three now uses AI (DSIT, 2026), mostly for writing and other language tasks, and only one organisation in ten has built it into core operations (Accenture). That gap, between using AI for words and using it for operational decisions, is the opportunity.
Pressure to "do something with AI" is also pushing firms towards tools before the question is clear. Software renewals and new projects can lock in cost and reduce flexibility. But this is a change in how the business makes decisions, not a technology purchase. BCG advises putting 10% of the investment into models, 20% into system integration and data, and 70% into people and processes. What is usually missing is joined-up direction across the business, not analysts or software.
Mid-market firms have an advantage here: short decision chains, leaders close to operations, and change that is simpler to carry through. That matters because the best responses cut across departments. The constraints are real too: thin data skills, uneven data quality and no spare management time. These are reasons to start small, not reasons to wait.
Where to start
Six questions for the leadership team:
Which operational decisions do you make every week or month, and which matter most to margins?
For each one, do you compare what you decided with what actually happened, and does that change the next decision?
What has been persistently disappointing over the last two years, and what has quietly improved?
Which measures have drifted steadily in one direction, and which problems appear only when two causes combine?
Which causes are outside your control? What internal lever compensates for each one, and who owns it?
Where does the evidence sit, and does anyone see it together?
Then pick one recurring decision where the data already sits in two or three systems and delay already has a price. Start closing the loop between what was decided and what happened.
You can't control energy prices, interest rates or tax. You can make every quote, schedule and order a little better than the last. The information is already there. What's missing is a decision system that learns from it.
About Planned Data. Peter Osborn built Redwood International from a sole trader to 14 offices before its sale, and now advises owner-led UK firms independently. These six questions open our Objectives and Baseline Review, a fixed-scope first stage that ends with a recommended first priority. planneddata.com
Comments