For businesses held together by spreadsheets, WhatsApp and workarounds
Modernize how your
business operates.
Fewer disconnected tools. Less work typed in twice. One place to see what's actually happening. We start by understanding how your business really runs, then decide what to fix, connect, automate or build.
Systems running live today in travel, B2B trading and marketplace businesses.
What we solve
If this sounds familiar,
you're not alone.
The same data lives in five places
And the version that's actually right lives in someone's head.
Approvals happen in WhatsApp
So nobody can tell you what's waiting, on whom, or for how long.
The business runs on one spreadsheet
That breaks the moment the person who understands it goes on leave.
You find out what happened, not what's happening
Numbers arrive days late, assembled by hand, already out of date.
You pay for software nobody uses
Bought for the business you were three years ago, worked around ever since.
Your team retypes the same information
Out of email, into a sheet, into the system. By hand. Every single day.
Our position
The right answer isn't always custom software.
And it isn't always AI.
Sometimes it's configuring what you already have. Sometimes it's an integration. Sometimes it's a deterministic workflow. Sometimes it genuinely needs AI. And sometimes the business deserves a custom platform.
We figure out which, before we touch a single tool.
Easy to claim. Here's the project where the answer was no AI at all, and why that was the right call.
How we work
Understand, then earn
the right to build.
Understand
Map how the business actually operates, not how the org chart says it does.
Simplify
Remove unnecessary process and system complexity before adding anything new.
Integrate
Connect the systems that already work instead of replacing them on instinct.
Automate
Automate the parts of the process that are genuinely predictable and deterministic.
Apply AI
Use AI only where interpretation, reasoning or generation adds real value.
Build
Engineer custom software for the parts no existing product solves properly.
How we think about AI
Rules where certainty exists.
AI where it doesn't.
Pricing, tax, thresholds, approvals, accounting. If the answer should always be the same, it's code, not a prompt.
Extracting the unstructured, understanding unusual requests, drafting communication, where interpretation is genuinely required.
Exceptions, commercial judgment, final approval. Accountability doesn't get automated.
A worked example
One message,
traced end to end.
Every enquiry gets a real answer within a minute, at 2am, on a Sunday, without anyone on staff. And no customer is ever quoted a price the business didn't set. Here is the system behind that, stage by stage, and exactly where the AI is and isn't allowed to operate.
-
Inbound
A message arrives
Guest messages the business on WhatsApp. No form, no portal, no app to install.
-
Deterministic
Gate
Is the assistant allowed in this conversation at all? If a salesperson has taken it over, it never runs.
-
Deterministic
Screen
The message is capped in length and checked for hijack attempts before the AI is ever called.
-
Deterministic
Meter
Daily limits per person, and a cap on how long any conversation runs, counted in the database, not left to the AI.
-
Deterministic
Settle
Four messages fired off in ten seconds shouldn't get four replies. They collapse into one answer, against live inventory.
-
AI
Understand and draft
Only now does the model run, on the one job it beats code at: working out what someone actually wants, and answering like a person would.
-
Deterministic
Validate and route
The reply is checked before any customer sees it. Then a rule, not the model, decides: send, escalate, or close.
-
Human
A person closes
Qualified leads go to a salesperson with the full context attached. The assistant never confirms a price, availability or a booking. One click pulls any thread back to a human, mid-conversation.
Six of the eight stages are ordinary code. That is what keeps the running cost predictable and stops the system promising a customer something the business can't deliver. Most of the cost, and nearly all of the risk, in an "AI system" lives in the parts that aren't AI.
Selected work
Three systems,
three different arguments.
Deliberately not equivalent. Each one solved a different kind of business problem.
The operating platform behind a villa marketplace
Four lead channels resolving into one CRM record, a catalog that redeploys the public site on edit, and an AI assistant structurally prevented from confirming a booking.
A system with no second chance
Registration, QR passes, door check-in and a raffle desk, built around one question: what happens when a piece of this fails while a hundred guests are waiting?
The one where we shipped no AI at all
A booking platform for island owners who run on spreadsheets and paper diaries. We kept the spreadsheet, keyed agents by company name, and shipped zero AI.
Three more in flight across travel, B2B trading and legacy ERP consolidation. See all work.
How an engagement starts
No fixed-menu pricing
on day one.
We scope after we understand the problem, not before.
FAQ
Straight answers,
before you ask.
What does KoreFoundry do?
KoreFoundry modernizes how businesses operate by untangling fragmented systems, automating workflows, applying AI where it makes sense, and building the software that's missing.
Is KoreFoundry a software development agency or an AI consultancy?
Neither, by design. We don't start with a technology decision. We first map how a business actually operates, then decide whether the right fix is configuring existing software, an integration, a deterministic automation, AI, or custom-built software, in that order of preference.
Do we need to commit to a large project to start?
No. Engagements start with a free discovery call, followed by a scoped Systems Review. There's no fixed-menu pricing on day one. Scope and cost are defined after the problem is understood, not before.
How do you decide what should be AI and what should be ordinary code?
Anything with a correct, repeatable answer (gating rules, rate and turn limits, input screening, output validation, routing, pricing, tax, approval thresholds) gets written as deterministic code. AI is used for the narrow middle: interpreting an unstructured request, matching it against real data, drafting a human-sounding reply. Commercial commitments (confirming a price, availability, a booking) are never made by the AI. In the live assistant traced in the worked example, six of eight stages are ordinary code and exactly one is the model.
What kind of businesses does KoreFoundry work with?
Businesses that have grown into a mix of spreadsheets, WhatsApp, ERP or CRM tools, manual approvals, and people holding processes together, commonly in travel and hospitality, B2B trading and distribution, and operational ERP modernization.
Bring us the messy business problem.
We'll figure out what the right system
should be. And then build it.
Book a discovery call
or write to info@korefoundry.com