The connection itself
Through supported APIs where they exist, and through whatever is reliable where they do not - including databases, files and screen-level automation as a last resort.
Integrate AI
Take one AI use caseall the way to production.
A model wired into a workflow people already use, chosen on the evidence of an evaluation harness rather than a partnership badge. Four to eight weeks, priced to the outcome after a free initial audit.
Every mark here links to the engagement it belongs to. See the work →
An AI Integration Sprint puts a language model into a workflow people already use, in four to eight weeks, priced to the outcome after a free initial audit. Claude, OpenAI, Gemini or another model is chosen on the evidence of an evaluation harness rather than a vendor partnership. The engagement suits a company with a defined use case and a live product, where accuracy, latency or cost has to be proven before rollout.
Choose thiswhen
- You have a defined use case and a live product
- Accuracy, latency or cost needs proving first
- It must reach production, not a prototype
You get
Model selection and an evaluation harness, a cost ceiling with approval gates, and production rollout with monitoring.
Pricing
Priced Outcome
One agreed price for an agreed outcome, quoted after a free initial audit and approved before delivery begins.
What the estimate depends on- The scope of the outcome you want
- The stage the application is at
- The condition of the existing codebase
- The team the work requires
- How the system is used in production
- The operational complexity around it
Work out what it has to return first: the ROI calculator for operational work, how we measure for everything else.
Deliverables
What weactually deliver
A shared data model
One agreed definition of a customer, an order and an invoice across systems that each had their own.
Failure handling that does not lose work
Retries, queues, and errors surfaced with enough context to act on.
Model choice, evidenced
An evaluation harness that scores candidates on your task and your data, plus a cost ceiling and approval gates. The model that ships is the one that won the harness, not the one behind a partnership badge.
Visibility
A panel showing what is connected, when it last synced, what the model is costing and exactly what failed when something did.
Where it starts
The four places thisusually shows up
Operations
Work moves between people by hand, and every handoff can stall.
- The same record is typed into two systems
- A spreadsheet sits between two applications
- Volume grows, headcount is the only lever
Finance operations
Invoices and approvals depend on somebody remembering to check a folder.
- Month end is a week of copying
- Approvals live in an inbox rather than a system
- Nobody knows the cost per document
Customer support
The answers exist, in documents and past tickets, and finding them is the job.
- The same answer is rewritten daily
- Response time depends who picks it up
- The demo worked, production did not
Reporting and data
The number the business runs on is assembled by hand, twice, differently.
- A report is rebuilt weekly from exports
- Two dashboards disagree
- The data cannot be queried safely
Proof
Engagements thatlooked like this
CrystalCommerce How CrystalCommerce kept scarce inventory accurate across sales channels
No baseline was captured on this engagement.
COCOLI How COCOLI automated furniture supply across three European markets
No baseline was captured on this engagement.
Tunelark How Tunelark connected lesson sales, scheduling and instructor payouts
No baseline was captured on this engagement.
Is this the right engagement for you?
Describe what is actually going wrong. We come back with the outcome, the engagement that fits and the price basis - after a free initial audit.
Recognise this?
Where a model that never reached the workflowusually shows up
The demo works and nobody will ship itIt performs on the handful of examples somebody tried. How often it is wrong, how long it takes and what it costs at real volume are all still unknown.01
The model cannot see the whole recordTwo systems each hold part of the truth and somebody reconciles them by hand, so anything built on top is working from half the picture.02
The nightly exportData moves between systems on a schedule, so every answer is drawn from a picture that is up to a day old.03
The old system has no APIIt is business critical, it is not going anywhere, and every integration proposal has died on that fact. This one has to survive it.04
Nobody knows when it breaksA sync fails, or the model quietly starts returning something wrong, and the first person to notice is a customer acting on it.05
How we decide
Integrations fail on the exceptions, not the happy path.
Any competent developer can move a record from A to B once. What matters is what happens the four hundredth time, when B is down and A has already committed.
- Idempotency, so a retry does not create a duplicate order.
- Queues, so an outage delays work instead of destroying it.
- Reconciliation, so drift between systems is detected rather than discovered.
- Alerts to us, because we operate what we build.
In practice
What fixing a model that never reached the workflowlooks like
Accounts, orders, credit status
ERP ↔ CRM
Sales sees what finance sees, without anyone exporting a spreadsheet to prove it.
Email → structured record → ERP
Mailbox → system of record
The inbox stops being a queue that only one person can see.
Invoices, payments, reconciliation
Accounting ↔ operations
Money and operations agree on what happened, continuously rather than at month end.
Files → extraction → structured data
Document store → workflow
The contracts, specs and POs sitting in a shared drive become data the business can act on.
Go deeper
The decisions behindthis kind of work
- ArticleWorkflow automation vs. AI agents: which should run your process?A practical comparison of deterministic workflow automation and AI agents, including where each fits and why most production systems need both.
- ArticleWe mapped customer-service enquiries. Here are 8 automation opportunities.A mapped service workflow for order-status, history and delivery-change enquiries, with AI limited to classification and drafting.
- ArticleWe mapped supplier product-data updates. Here are 8 automation opportunities.A mapped supplier-data workflow for inconsistent spreadsheets, identifiers, units, price changes, validation, approval and ERP loading.
Questions
About a model that never reached the workflow
Usually not. Between the database, scheduled file exchanges, whatever endpoints do exist and - as a last resort - controlled screen-level automation, there is normally a reliable path. We will tell you honestly how fragile each option is before you commit to one.
Sometimes, and that is precisely why we operate what we build. Vendor changes are a maintenance category we plan for rather than an incident that surprises everyone. Monitoring tells us before it tells your customers.
Yes, and we prefer it. They know your environment. We are usually the ones doing the workflow and integration work while they keep owning infrastructure and access - with clear boundaries written down so nobody is guessing who does what.
The design decides that before launch. Where a wrong call would be expensive, the workflow stops and routes to a person instead of guessing. Failures are logged and alerted, failed items are queued rather than dropped, and there is a rollback path. We operate what we build, so the alert comes to us.
No. Custom code and configuration built for you belongs to you, systems run on infrastructure you control wherever possible, and documentation is a deliverable rather than an upsell. If you stop working with us, your business should not stop working.
Not this one?
The otherengagements
Our roadmap needs a senior team behind it.AI-Native Engineering Pod3 months minimumPriced OutcomeManual work is eating our capacity.Agentic Workflow Automation8-12 weeksPriced OutcomeThe prototype cannot survive real users.Vibe Code Rescue Sprint4-6 weeksPriced OutcomeThe old system is too risky to change.Legacy Modernization2+ quartersPriced OutcomeNobody owns the product after launch.Product CareOngoingPriced Outcome
Before you commit
What you wouldbe signing up for
One bounded outcome
A named result, a fixed window and a scope agreed in writing before delivery starts.
- Commitment
- 4 to 12 weeks
- Price basis
- Priced Outcome
- The problem has an edge you can describe
- You want the cost known first
- Someone inside owns the result
A team against a roadmap
Senior capacity inside your codebase, moving a roadmap week after week.
- Commitment
- 3 months minimum
- Price basis
- Priced Outcome
- The work is continuous, not bounded
- Recruiting would cost a quarter
- You want the same people throughout
Ownership after launch
Monitoring, patching, upgrades and a named engineer for live software.
- Commitment
- Ongoing, cancellable
- Price basis
- Priced Outcome
- It is live and business-critical
- Nobody owns it out of hours
- The risk needs an owner
One agreed price for an agreed outcome, quoted after a free initial audit and approved before delivery begins.
Named quotes go here as clients approve them. See what the engagements produced →
Next step
Which two systems is someone keeping in sync by hand?
Name them and we will come back with what it would take to connect them properly - including the cases where the honest answer is that the export and a person is still cheaper.