JDD Agency

JDD Agency We get it, wrestling with systems in the workplace is tiring! Let’s make your business work for you, together. 🚀

JDD Agency helps fellow UK small business owners automate the boring stuff – so you spend less time wrestling and more time with family.

Ask a leadership team why they shut down a venture two years ago and you will get a story. Ask what they believed at the...
08/09/2026

Ask a leadership team why they shut down a venture two years ago and you will get a story. Ask what they believed at the time, and how confident they were, and the room goes quiet.

That is not a memory problem. It is a recording problem.

Most innovation portfolios run on two facts: what stage each idea has reached, and who is championing it. Neither tells you anything about the quality of the thinking behind it.

So the review becomes a persuasion contest. The venture with the most convincing sponsor advances. The one whose sponsor is travelling does not. Everyone knows it happens, and nobody writes down enough to prove it.

Satori starts somewhere else. It tracks ventures against a fixed set of key success factors, and for each one it captures how confident you are, and why.

The second one changes behaviour.

A pipeline stage is a fact. A confidence score with a written justification is a position, held by a named person, at a known point in time, about something that had not happened yet. It can be wrong. It can be revisited. It can be compared with how things actually turned out.

That is the difference between a portfolio you report on and a portfolio you learn from. Not because the scores are right, but because the reasoning outlives the meeting that produced it.

The structure around it is plain. Business units, venture boards, success factors, stages. Nothing clever, and that is rather the point: the discipline is the product, not the dashboard.

So the next time something gets quietly shelved, the question is not whether it was the right call. It is whether anyone wrote down their reasoning well enough that the answer is a record rather than a story.

Most AI in business software only ever suggests. It flags something, and then a human decides whether to care.That looks...
03/09/2026

Most AI in business software only ever suggests. It flags something, and then a human decides whether to care.

That looks like responsible design. Usually it means nobody trusts it enough to let it stop the work.

RiskZen is used to carry out property risk assessments. An assessor walks a building, records what they find, and files a report that becomes a compliance record. AI Validation reviews that report and raises issues before it goes out.

Until recently, the assessor could submit anyway. The AI raised an issue, and the report was filed regardless.

So the team closed it. A fire risk assessment can no longer be submitted while AI-raised issues are outstanding. Each one has to be resolved, verified, or explicitly dismissed by a person first. Not a warning banner. The submission is blocked.

Which is where it gets interesting. Once a check can stop work, someone will want past it. In the original design, an assessor who disagreed with an issue left feedback, and that feedback changed what the AI raised next time, on every future building, for every other assessor.

The people being reviewed could quietly retrain the reviewer.

So the gate did not ship alone. Feedback still applies immediately to that assessor's own report. It now waits for an approver before it changes anything for anyone else.

The block covers fire risk assessments only, the work AI Validation has been proven against, and it can be switched off centrally without a deployment.

Anyone can add AI to a product. Far fewer would let it block the work. If you wouldn't, the problem is rarely the model. It is that nobody has decided who is allowed to overrule it.

Every promise a product makes is true when it is made. That is not dishonesty. It gets written when the product is small...
26/08/2026

Every promise a product makes is true when it is made. That is not dishonesty. It gets written when the product is small, when a team can still hold every case in their heads. Then reality adds cases, and the promise develops exceptions nobody writes down.

We build for a sports tips platform whose premium feature makes an unusually specific promise: whatever you filter for, you get one pick per match. The most confident call, priced with real bookmaker odds. Not a shortlist.

A promise that precise has two ways to die.

The first is a filter combination nobody tried, quietly returning two picks for one fixture. The user does not see a bug. They see a product that cannot make up its mind. So the rule is not a tidying step that runs afterwards. It sits inside the single query that produces the results, and nothing can skip it, because there is no separate step to skip.

The second way is slower, and far more common. The business grows. A new competition gets switched on. The obvious way to build a filter is a list of names inside the feature, accurate the day someone typed it. Now the business sells something its own feature has never heard of. Nothing breaks. The option simply is not there. And nobody reports it, because nothing failed.

So the filters are not a list. They read the platform's own catalogue of what it currently covers, the same data every other part already treats as the truth. Turn something on, and the feature already knows.

Most companies have a private copy of what the business sells, sitting inside a feature, a report, an integration, a form. It was right once. It goes stale in silence, and a customer finds it before you do.

The promises worth making are the ones your own system cannot quietly stop keeping.

Most software loses deals to a competitor that never appears on a shortlist: the export file nobody wants to retype.Ask ...
24/08/2026

Most software loses deals to a competitor that never appears on a shortlist: the export file nobody wants to retype.

Ask why a business is still on software it complains about weekly and the answer is rarely about features. It is years of records in a shape nobody chose, and a move measured in somebody's evenings.

We build a platform that schools and activity clubs use to run enrolment. Every club that arrives brings the same obstacle with it: a spreadsheet from the old system, a PDF register, or a photograph of a printed class list. The old answer was that an admin retypes it, which is the point where switching stops being worth it.

The new answer is to drag the file in and let an AI model read it. That is not the interesting part.

The moment an AI model interprets a business's records, you have a question it cannot answer. What if it is wrong? With family data, mostly right is not a passing grade.

So the extraction never touches the database. It produces a preview, every row on screen, each child matched to a class or session pack the club actually runs. Only when an admin approves is anything written, and it is re-validated on the way in. Matched rows face the same age-band rules that apply when staff enrol a child by hand, so the importer gets no shortcut a person would not get.

Families that move keep their enrolment active and are not asked for payment details again on day one, so the change of platform stays invisible to the parent.

An AI model is remarkably good at interpreting mess, and categorically not accountable for the result. Extraction is a suggestion. Committing is a decision.

A demo tells you an AI model can do the job. It tells you nothing about whether you can walk away from it.Here is what w...
20/08/2026

A demo tells you an AI model can do the job. It tells you nothing about whether you can walk away from it.

Here is what walking away actually looks like.

We built a system called X-Tees. Anywhere on X, someone sees a post with an image in it and comments a tag underneath. Minutes later, a reply lands in that thread with a link to buy the image as a t-shirt. Five sizes, a shirt colour taken from the words in the post, a real storefront. Nobody designed it, nobody listed it, and nobody wrote the reply.

The impressive part, the design itself, is one call to Google's Gemini. That is the easy part.

The work is in the seams. The model refuses sometimes, and usually it is right to, so the refusal is detected and the person gets a message instead of silence. The bot's own posts create mentions, which the bot sees, which it replies to, so there is guard logic that recognises its own output before it starts talking to itself.

None of that is glamorous. All of it is why the thing can be left alone.

When someone shows you an AI automation and it works, you have learned one thing: the AI model can do it. Ask what happens when it says no.

Somebody on your team will delete the wrong person this year.It happens in every business that runs its staff through so...
19/08/2026

Somebody on your team will delete the wrong person this year.

It happens in every business that runs its staff through software. A leaver gets processed twice. Two colleagues share a surname. A tidy-up goes one row too far.

What happens next tells you more about your software than any feature list does. Either an administrator fixes it in seconds, or somebody raises a ticket and a colleague cannot log in until it clears.

Here is the strange part.

The account that was deleted was often never destroyed. Modern frameworks commonly mark a record as deleted rather than erasing it. The row stays. The person simply stops appearing.

That was the situation in a client's admin panel. Deleting a user revoked their access and took them off the list, while the record sat in the database the whole time.

There was just no screen that showed it, and no action to bring it back.

So every accidental deletion became a developer job, editing data by hand to undo a click that took half a second.

They could have built a recovery system. They looked at what already existed instead.

A Deleted tab beside the Active one. A Restore action leaning on recovery behaviour already in the framework, with the same confirmation step deletion had. Roughly 180 lines of code, in one change.

An administrator can fix their own mistake in seconds. Nothing is re-entered. Nobody waits on engineering.

Which is worth a thought about your own systems.

The gap you are budgeting to close may not be a capability gap at all. It may be a capability nobody can reach.

Think about your main phone number. The one on your website, on your contracts, in every calendar invite you send.You ha...
19/08/2026

Think about your main phone number. The one on your website, on your contracts, in every calendar invite you send.

You have probably never worried about it once.

Here is the part nobody mentions. You do not own that number. It is a pointer. Behind it sits physical voice infrastructure rented from telecoms carriers, and carriers have bad days like everybody else.

When one does, the fix sounds simple. Point the number at the backup instead.

Easy enough for one number. We build the software behind a leading global bank's conferencing platform, where the dial-in numbers run into the thousands, from more than one wholesale carrier. Each has to be moved individually. And any one of them can refuse.

This is where most continuity plans quietly fall apart. Failover gets imagined as a switch you flip. It is not.

So it was built expecting that. Every number carries its own state: waiting, moving, moved, skipped, failed. Whoever is running the switch watches a real completion percentage climb, number by number.

When a run ends with a few stubborn numbers left behind, one action re-runs only those, and never disturbs the ones that already moved.

So "did it work?" becomes "all but a handful, and here is that handful."

None of this was built after something went wrong. It has been developed since 2022 and extended again in 2025, for a day that may never come.

Most companies sum up their resilience in one sentence. We have a backup. The sentence that matters is the one you need halfway through using it, with failures on the screen and customers already calling.

Meanwhile the businesses on that platform keep dialling the same number they always dial, and never think about it.

Which is exactly what they are paying for.

Ask a compliance team where their week actually goes. Almost nobody says the checks.The identity check itself takes seco...
16/08/2026

Ask a compliance team where their week actually goes. Almost nobody says the checks.

The identity check itself takes seconds. What takes the time is everything around it. Working out which clients need one. Remembering to start it. Chasing the ones sitting outstanding. Clearing the entries that turned out not to apply to that client at all, which somebody still has to open, understand and dismiss.

Which is why so many firms can say their AML checks are automated and still have people spending real hours on AML. The verification step was automated. The process was left exactly as it was.

When we built anti-money-laundering checking into an accountancy platform, we went after the process instead.

First, the thing being checked had to be unambiguous. Every business client now resolves to exactly one named individual, its primary contact, linked to that person's own identity record. Every check and every identity document attaches there, and the system enforces that link rather than trusting staff to maintain it, so it cannot drift or duplicate.

Automating an ambiguous process does not remove the ambiguity. It produces it faster, and with more confidence attached.

Then the process itself.

Checks start on their own for clients who are eligible. Nobody has to notice a new client, decide whether a check is due, or remember to trigger it.

A check that turns out not to be needed deletes itself. It does not sit in an error state waiting for someone to interpret it.

And the status of every client is on their record in plain sight, verified, outstanding or flagged, with the reason attached. A person gets involved at the point where a person's judgement is genuinely required, and not before.

It has been live since early August, with its own automated tests covering the rules that matter, and every existing client was migrated onto the new structure by a repeatable, reversible process rather than a manual clean-up.

If your firm carries AML obligations, whether you lend, broker or advise, the second-order benefit is the one worth noticing. Your duty is not only to run checks. It is to show, client by client, which named individual was verified, when, and against what. When the process runs itself, that evidence assembles as a by-product of ordinary work instead of being reconstructed from memory the week a regulator asks.

So the useful question is not whether your AML checks are automated. Most firms will say they are.

It is how many people it still takes to run them.

Most businesses keep records to remember what happened. A smaller number keep records to prove it.The difference costs n...
15/08/2026

Most businesses keep records to remember what happened. A smaller number keep records to prove it.

The difference costs nothing until the day something you already certified turns out to be wrong.

That day arrived for a property manager responsible for compliance across a residential estate. New fire-safety guidance reversed an earlier finding. Every apartment already held a fire door certificate stating one thing, and the correct answer was now something else.

The obvious response is to go and correct them.

Here is what makes that harder than it sounds. Those certificates are not notes, they are evidence. They exist to show what was assessed, when, and against which standard. Edit them to show today's answer and you have not only updated the record, you have removed any proof of what was believed before.

So when a regulator, an insurer or a buyer's solicitor asks what you knew last spring, the file cannot answer. It can only state what you know now, and imply you always knew it.

That is the trap. Fixing a record and destroying your evidence look identical from the inside. Both leave you holding a tidy set of documents that say the right thing.

So the team ruled out editing before anything was designed. Originals were never to be altered or deleted, and a correction had to be its own operation rather than a shortcut through an existing edit screen.

What that produced is a single form. Complete it once, apply it to every selected apartment, and a fresh corrected certificate is generated for each, while every original stays exactly where it was.

One correction now reaches an entire estate in a single action, instead of an administrator reopening and reissuing every apartment's report by hand.

The part that matters more: the estate owns a history rather than a snapshot. It can show the original assessment, the change in guidance and the corrected certificate, in sequence. In front of a regulator or an insurer, that is a considerably stronger position than a clean set of documents with no visible past.

Raised, scoped, built and demonstrated on a live call the same day.

Worth ten minutes in your own business this week. Pick a record you might one day need to defend: a price you quoted, an approval you signed, an inspection you passed. Ask whether your system can show what that record said eighteen months ago, or only what it says today.

If it is only today, then somewhere in your business, correcting a record is quietly overwriting your evidence, and nobody will notice until the moment it matters.

Any system worth using can be corrected. The ones worth defending can be corrected without forgetting.

The University of Oxford is older than the Aztec Empire.An entire empire rose, ruled and vanished while a university in ...
14/08/2026

The University of Oxford is older than the Aztec Empire.

An entire empire rose, ruled and vanished while a university in England carried on teaching.

What carried it through was never the buildings. Those were replaced. The subjects changed, the funding changed, the language of instruction changed. What continued was the practice, handed deliberately from one generation to the next.

Your business has one of those too. It is the part of you that has survived every system you have ever bought. And in most companies, nobody has written it down.

Here is where it actually lives. It lives in the workarounds.

The spreadsheet beside the system, because the system cannot express how you really price. The check somebody always runs before approving an order, because they know which customers cause trouble. The group chat that exists because the official notification arrives too late to be useful.

Most leaders read that as untidiness, something to stamp out with better training. It is the opposite. Every workaround is a real rule your business has that your software does not know. Someone was paid to learn it. It survived contact with real customers. It is the most accurate documentation of your operation in existence, and it is kept in the least durable place available: individual heads and private files.

You can turn all of that into something useful in a week. Worth saving, this part.

The five-day workaround log.

1. For five working days, ask every team one question at the end of the day: what did you do today that the system should have done for you? One line each. No fixing, no blame, no meeting.

2. Against each line, write three numbers. How often it happens in a week. How many people do it. What it costs when it goes wrong: irritating, expensive, or dangerous.

3. Sort by all three. Do not trim the list. Twenty lines is a healthy result.

4. Read the top five aloud in your next leadership meeting.

Three things fall out of that page.

Some of it needs no software at all, just a decision nobody has made. Those are free, and you can act on them the day you read the list.

Some of it is one person carrying a rule nobody else knows. That is a risk sitting on your payroll, and you have found it before it resigns.

The rest is the most valuable software you could commission this year, ranked by your own operation instead of by a vendor's roadmap.

Oxford is still here because its practice was carried on purpose rather than left to chance.

Yours is sitting in your workarounds. Five days and one page, and you will know exactly what to build.

Address

Southampton

Telephone

02380 970 212

Website

Alerts

Be the first to know and let us send you an email when JDD Agency posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Contact The Business

Send a message to JDD Agency:

Shortcuts

Share

Category