This is what "more efficient, less manual work, more revenue" looks like in practice. Eight things we build for a business with invoices, stock and suppliers to deal with, then eight faults we have found and fixed, with the real figures attached.
payments taken across our products
Not one chargeback. Not one dispute.
records ingested and reconciled
Into a live product, not a spreadsheet exercise.
pages on one product
Built, maintained and monitored in-house.
Real figures from products we built, own and run. Anonymised, and shown as percentage growth rather than in pounds.
From nothing to a live channel without spending on ads.
Search Console clicks for one of our own products, in its first four months. No advertising spend.
In the order most owners care about it. Money first, marketing last.
Invoices raised from the job, sent when they are due and chased on a schedule that runs whether or not anyone remembers.
Bank, sales and suppliers lined up automatically, so a person only looks at the handful of lines that disagree.
A record of what you actually used, checked against what you were actually charged. Example B is what its absence costs.
One quote, priced from your own cost data, that becomes an order and then an invoice without being typed twice.
What you own, where it is, who has it, when it is due back and what it has earned while out.
Who is where tomorrow, what they need with them, and changes that reach the people affected.
Numbers that trace back to the records behind them. Most reporting problems are counting problems, not chart problems.
A site that does a job, plus ranking in search and being cited by ChatGPT, Perplexity, Copilot and Gemini.
Every fault below was found and fixed in systems we build, own and run, which is why the exact figures can be printed at all. Each one is a counting or control failure that gets more expensive with scale, because a bigger operation runs longer before anyone notices.
Of those reaching the payment step, 87% paid, out of a sample of 15. The rate was never the problem: the form in front demanded ten fields before it would show a price. Cut to five, price first.
Where this sits in a larger business
A tender win rate counted only on enquiries that survived a four-page form.
624 units charged against 407 real jobs, because two parts of the same process each ordered the work. Nothing compared what was bought against what was delivered, so no control aimed at the supplier could have caught it. Residual after the fix: one duplicate in 1,006.
Where this sits in a larger business
Double-ordered stock, correctly supplied and invoiced. Plant left on hire after it came back to the yard.
One payment account billed three businesses, so the same money was counted twice; income landed under two transaction types, and filtering on one understated two of them by 43% and 30%; and 112 test records counted as real sales. All three corrected.
Where this sits in a larger business
One merchant facility across several trading names; intercompany transfers counted at both ends.
A number nothing was checking, or a check that could not tell failure from a quiet day.
A connection quietly expired, and every layer beneath returned an empty list rather than an error, so an outage and a quiet day produced an identical report every morning. The repair exposed three further faults of the same shape, including a currency conversion falling back to a rate of 1.
Where this sits in a larger business
Any feed trusted because it has always arrived.
The lesson, and it applies to any business
A check that reports nothing found when it fails is indistinguishable from success. Monitor that it ran, not only what it returned.
A cost was estimated from an approximation while the authoritative figure was already being fetched and discarded, so two parts of the same document disagreed. Because documents were links rather than attachments, one fix corrected every past one.
Where this sits in a larger business
Any document mixing live data with a stored assumption: quotes, schedules, valuations.
Both were written as: if the API key is missing, this must be a developer machine. Production had no keys configured, so both agreed, and the paid product went to anyone holding the right link.
Where this sits in a larger business
Approval limits that pass when the approver field is empty. Access that defaults to allow when the directory is down.
The lesson
Never treat a missing credential or limit as evidence of anything except a missing credential or limit. Absent must fail closed.
A promotion across thousands of pages appeared to be sending traffic; two independent measurements agreed almost all was automated. What separated people from bots was time on the page, not the bounce rate everyone looks at. It stopped a project about to be built on the number.
Where this sits in a larger business
Any metric about to justify a spend that nobody has tested for realness.
Two pages on one site: the better-ranked draws 4 visits from AI assistants on around 10,000 Bing impressions; the other has a sixth of that visibility and draws 268, up from 85. Ranking and domain strength cannot explain it, because both share them.
Where this sits in a larger business
Buyers asking an assistant to shortlist suppliers in your sector before running a search.
Status: shipped and being measured, not yet proven
The 34-page rollout is live and tracked. It has not yet produced a proven result, and we will not claim it worked until it has.
The first conversation costs nothing. We use it to work out whether there is anything here worth doing, and if there is not, we will tell you.
Or email us directly: dave@daveknowsai.com