Examples of our work

The problem, what we found, what we changed, what runs now.

Five operations we were shown. Customers are named where they have agreed to it and kept private where they haven't. In each case we understood the work first, built the system around it, and we run it month to month.

Match-day operationsSquad dataFinance administrationAcross the businessFrozen food delivery

Match-day operations

Ninety-four analysts. Hundreds of fixtures. One coordinator.

The problem

A sports-data company sends a freelance analyst to every professional football match in Scotland. One coordinator ran the whole operation: who covers what, holidays, call-offs, postponements, television changes, cup replays, confirmations, expenses and pay. The same coordinator also recruits the analysts, across the whole of Scotland, from the Borders to the far north. It lived in spreadsheets, an inbox and a phone, and every change meant an evening of calls.

The coordinator was spending around 30 hours a week running it. Now it takes about 3.

Read the full story
What we found

The fixture list was never the hard part. The hard part was everything around it: knowing who could realistically get to each ground, finding out about a change before the analyst did, chasing confirmations, covering a call-off at short notice, and then working out at month end who was owed what. Recruitment had the same problem: knowing where cover was short before a match went unstaffed, and finding people who live near those grounds.

What we changed

We built the coordinator's whole operation into one system, around the way it was already run. The fixtures come in from the public feeds every hour. The allocation is prepared from each analyst's availability, real driving time, tier, existing games and the coordinator's own rules, and the coordinator approves it. Everything the analysts need to be told is drafted for the coordinator to send; nothing goes out until it is approved.

What runs now

94 analysts and approximately 240 matches a month, through one operating view.

  • The analysts. Who they are, where they live, which tiers they are cleared for, their holidays, and a personal page each where they see their games.
  • The allocation. Prepared from the rules, driving times and availability; re-run when the rules change, with an undo; never touching a game the analyst has already been told about.
  • Changes and call-offs. A moved kick-off, a postponement or a replay is seen on the feed and the affected analysts are listed. A call-off brings up the cover for one-tap approval, and nobody is sent back to a game they declined.
  • The communications. Monthly fixture lists, new-game and change emails, stand-downs, a Monday check-in where each analyst confirms or calls off from a link on their phone, chasers and day-before texts, all in the coordinator's own words.
  • Recruitment. A coverage map shows the grounds where cover is short. From any of them, the coordinator writes to the nearby clubs, one email each from their own mailbox, and the replies come back into the campaign. Applicants register online, are sent for training, and are tracked from first contact to their first match.
  • The checks. The month compared against the customer's own fixture sheet and payment sheet, difference by difference, and the weekly confirmation list for the customer produced from what was actually confirmed.
  • Expenses and payments. Fees by tier, mileage from the coordinator's own rate table, expenses the coordinator can override, and each analyst's monthly statement emailed from the system.
  • Invoicing. The month's invoice to the customer is built from the games actually covered, checked line by line, and raised straight into the coordinator's QuickBooks with one click.
  • The record. Every change, send and decision in one activity log.

"I came to Bob with a job that was consuming far too much of my time every month. The role involves placing around 200 to 320 matches a month across Scotland with more than 94 freelance analysts, previously all on spreadsheets, emails and phone calls. Bob took the time to learn how we actually work, rules and quirks included, and built a system that does the heavy lifting. It crews the month, messages the analysts, chases confirmations, spots matches that change at the last minute, and works out everyone's pay and mileage. It has cut my admin from around 30 hours a week to about 3, which allows me to focus on managing my team rather than drowning in admin.

What stands out is his urgency. When I ask for a change, it is done the same day, every time. He treats my problems like they are his own."

Warren Hawke, Founder, Strenua

Changes waiting for approval

Motherwell v St Mirren · 3 analysts to tellMoved to Sunday 15:00, TV pick
Arbroath v Queen's Park · cover neededPostponed, waterlogged pitch
Call-off · Kelty Hearts v Cove Rangers · cover readyNearest free analyst, 22 min drive · one tap to approve
October statements · readyFees by tier and mileage from what was confirmed
Squad data

Fifty-one football clubs kept current every week.

The problem

Fifty-one semi-professional football clubs change squads throughout the season, often without a single reliable source of truth.

Every change checked by the club itself, in leagues where there is no reliable public record to fall back on.

Read the full story
What we found

Squad changes are announced in different places, at different times, and often not at all, and clubs themselves need an easy way to correct the record.

What we changed

The system finds the changes every day. Every movement goes into a season ledger, the clubs are asked to check their squad, and the full history is kept. Every entry carries its evidence, so a change that was never announced and a player who has genuinely left can never look the same afterwards.

What runs now

One maintained record for 51 clubs, with club confirmation and a clean weekly file in the shape the customer's systems expect. Each club signs in with a code emailed to it, sees its own squad and can correct a player, mark one as left or add a signing, applied at once; it can also look up any other club in its league for the match programme. Every change says who made it, when and what it was before, and can be undone. Quiet clubs are chased, and the coordinator can see who has signed in and who has not.

This week's movements

Huntly · newSigned a goalkeeper from Formartine · Tuesday
Civil Service Strollers · leftMidfielder left for Spartans · Wednesday
Fraserburgh · checked by clubManager confirmed the squad · signed in Thursday
44 of 51 clubs have checked their squad 7 chased this morning
Finance administration

Stop typing paperwork that already contains the answer.

The problem

Supplier invoices, receipts, credit notes and new supplier forms arrive by email as PDFs. Somebody opens each one, reads it, types it into the accounts system, picks the account and VAT rate, attaches the file, and then does it again for the next one. The information was already in the document.

Read the full story
What we found

The expensive part was not the typing. It was the reading and checking: the wrong account code, a VAT rate that does not match, a supplier set up twice, an invoice filed twice, and the one mistake that cannot be undone, paying a supplier whose bank details were changed by a fraudster.

What we changed

We put the work where the paperwork already lands. An invoice arrives in Gmail; one tap files it as a draft bill in Xero with the PDF attached. Two separate AI models read every figure and must agree before anything is trusted; if they disagree it flags the document rather than guessing. The account is matched to the business's own chart of accounts and learned per supplier, the VAT rate matched to its real Xero rates, every line of a multi-line invoice captured, foreign currency handled. Nothing is posted until a person approves it.

What runs now
  • No duplicates. An invoice is never filed twice; an invoice and its receipt become one record; an existing supplier is recognised, not recreated.
  • Paid and approvals. An already-paid receipt is filed and marked paid in one tap. Any bill at or over the sign-off limit is held as needing approval.
  • Suppliers and fraud. A new supplier is set up from their own form. Sort codes, account numbers and IBANs are checked by rule and cross-checked against each other; anything that does not add up carries a "verify before paying" warning.
  • Credit notes recognised and filed as credit notes, not bills.
  • Answers in Slack. Five reports over the live Xero figures: daily control, aged debtors, supplier bill risk, cash pressure at 7, 14 and 30 days, and a month-end summary. The totals are worked out in code, never by the AI, and only people whose Xero role allows it get an answer.

Waiting for approval · Xero

Invoice INV-20441 · Northgate Print · draft bill£1,240.00 · due 28 Oct · two models agree on every figure
Receipt · Fellside Catering · filed and paid£86.00 · already paid by card
New supplier · Harbour Signs Ltd · verify before payingSort code and account number checked and cross-checked
Across the business

Catch what no single system can see.

The problem

A subscription business delivering frozen food to customers' doors ran on seven systems: subscriptions, payments, the courier, the help desk, reviews and Xero, each fine on its own. What cost it money fell between them. A late delivery became a complaint, then a one-star review, then a cancelled subscription, and nobody saw the chain until it had happened. Supplier invoices arriving by email were typed into Xero by hand.

Read the full story
What we found

Nobody needed another system to look at. They needed what the existing systems already held to be read together, and the next step prepared for them, in their own words, ready to approve.

What we changed

We built SMART Hub to sit above the systems the business already used. Each morning it reads them together, works out what needs attention first and prepares the fix. Every action waits for one tap.

What runs now
  • Supplier bills into Xero in one tap. An invoice arriving by email is read, coded to the right account and VAT rate, and filed as a draft bill with the PDF attached. Two independent readings must agree before a figure is trusted, and nothing is ever filed twice.
  • A morning operating brief across every system: revenue, cash, subscriptions, deliveries, support and reviews, with the one priority that matters most today and the evidence behind it.
  • Deliveries at risk. Parcels with no recent scan, failed delivery attempts and returns to sender, listed before the customer has to ask.
  • Customers at risk of leaving. Subscribers who have had a late delivery, an open complaint or a one-star review, with the revenue at stake and the reply drafted.

Operating brief · Friday 07:02

31 deliveries at risk · traceNo recent scan from the courier
312 subscribers flagged · after a late deliveryReplies drafted, waiting for one tap
Supplier bill · filed to Xero · both readings agreeCoded, VAT matched, PDF attached
3 new one-star reviews · all about delivery
Frozen food delivery

Keep the box frozen without paying for early delivery every time.

The problem

A business shipping frozen boxes to customers by courier had to decide, every day, which boxes needed an earlier delivery slot or extra ice packs to arrive frozen. Guessing meant either paying for early delivery and ice that were not needed, or boxes thawing on a warm doorstep.

Read the full story
What we found

The decision depends on things that are known in advance: the forecast where the customer lives, how long each box size stays frozen with one, two or three ice packs, when it leaves cold storage and the latest it could arrive. None of that needs judgement. It needs the rules written down and applied to every order. Deciding is half of it; the other half is knowing whether it worked.

What we changed

We built a planner that pulls the Met Office forecast for each delivery area, follows each box from cold store to doorstep and picks the cheapest option that keeps it frozen: early delivery, more ice packs, both, or neither. Where early delivery is not available it adds packs instead; where nothing is enough it marks the box as at risk. There is no AI in it. Every decision comes from rules the team can change, with a sentence saying why. It also reads the courier's tracking and measures the result, which is what tells the business whether paying for early delivery and extra ice is earning its keep, and where the courier is letting it down.

What runs now

A plan for each delivery day: a map of every delivery showing which need changing, the hourly temperatures behind each decision, how many ice packs production needs frozen before packing starts, an order list, a pick list and a courier file ready to upload. For every customer whose delivery is changing, a personal email written from their own forecast: the town, the day, how warm it will get and when, what has been done and what to do when the box arrives, ready for the team to send in the way they choose. A Results page shows what actually happened: early deliveries on time, late on hot days, arrived warm, failed, by delivery service and by area, over any period, with the individual deliveries behind each number. The team can upload their own order file and run it through the same rules, with duplicate orders, invalid postcodes and unknown box sizes flagged. It is set up to run alongside the existing process first, so the team can compare its decisions with their own before relying on it.

Deliveries on Monday · packed today

Bristol area · 2 boxes · deliver earlyHottest point 24°C at 15:00 · early by 12:00 keeps it frozen
Torquay area · 1 box · second ice packNo early slot here, so more ice
Customer emailHi Leo, it is going to be warm in your area on Tuesday, reaching 24°C around 15:00. To keep your box frozen we have asked the courier to deliver it by 12:00.
Results · from the courier's trackingOn time, late on a hot day, arrived warm, failed · by service and by area
Where AI fits

Use AI for the part it is good at.

AI is useful for reading, interpreting and drafting. It is much less useful when a business needs certainty. Where a value, date, price or status can be taken from a real system, we use the real value. Where AI is useful, it prepares the work and a person approves it.

Read

Let AI deal with unstructured information such as emails, documents and written requests.

Check

Use rules, source systems and deterministic data wherever the answer must be exact.

Approve

Keep a person in control of anything that affects a customer, supplier, account or business record.

Show us the process that is costing you time or money.

We look at how the work is done before talking about software. You get the findings in plain English and decide what happens next.

How the review works