Free diagnostic — about two minutes
Surfaces where time may be leaking and whether a deeper assessment is likely to help.
THE PRACTICAL GUIDE
AI implementation: from a working demo to a system your team runs.
AI implementation is the work between a model that impresses in a demo and a system that runs inside your business: connecting it to your data and tools (integration), putting it into production with review, monitoring and an owner (deployment), and handing it over so your team can change it. The model is the smallest part.
SENIOR ENGINEERING · MUNICH-BASED · EU & US CLIENTS · BUILT TO HAND OFF
THE SHORT ANSWER
What AI implementation is — and what integration and deployment mean.
Three words are used for this work, often interchangeably, and they are three different amounts of it. Knowing which one a quote covers is the first thing an implementation buyer needs.
- AI integration
- Connecting the model to what it needs: your data sources, the tools it may call, the permissions it runs under, the systems it writes back to. The half most quotes stop at.
- AI deployment
- Putting the integrated system into production: a review step where a human decision changes the outcome, monitoring that says when it drifts, logs, a fallback, and a named owner.
- AI implementation
- Both, plus the handover — documentation, a runbook and the ability of your own team to change it. An implementation is finished when the people who run the system are yours.
WHAT GOES WRONG
Four ways an AI implementation fails.
Rarely because of the model. Almost always because of the plumbing around it.
01
Data the model cannot reach
The demo ran on a clean export. Production runs on the systems as they are: permissions nobody mapped, fields that mean different things in two departments, a source that updates on Fridays. Integration is where most of the time goes, and where most quotes are silent.
02
A workflow nobody mapped
AI was pointed at the work before anyone wrote down the trigger, the decisions, the exceptions and the owner. The system then automates a process that only existed in one person’s head, and breaks the first time that person is on holiday.
03
A pilot with no owner
The pilot worked. Nobody was named to run it, so it ran until the first change in the data underneath it, and then it stopped — quietly, with its budget line still green.
04
Review that is uniform, or absent
Every action sent to a person becomes a rubber stamp; no action reviewed becomes a liability. Review has to sit where a human decision changes the outcome, with enough context to make one.
Every engagement here starts by mapping the real workflow and ends with a named owner — the two failures that cost the most and show up last.
HOW LONG IT TAKES
Weeks, with an end date each.
- Two weeks — AI Workflow Audit
- Which workflow to implement first, ranked by return and by effort, and what the first build has to achieve.
- 4–6 weeks — AI Automation Sprint
- One production system: integrated with your data and tools, deployed with review and monitoring, documented and handed over.
- 3–5 weeks — AI knowledge base
- A retrieval system over your own documents, scored against real questions before it goes in front of staff.
- After handover
- Thirty days of support, then your team runs it. No retainer is required and no platform is licensed.
Durations are for scoped engagements on one workflow or one knowledge base. A quote in days is a demo; a quote in quarters is a programme.
IN-HOUSE OR HIRED, AND WHO
How to choose who implements it.
Three questions for any partner — and one for yourself first.
The question before the partner: in-house or hired? Hire in-house when AI work is continuous and you have the engineering leadership to direct it; bring someone in when you need one system shipped, or the first one, before a full-time role is justified. Many companies do both, in that order — the agency versus in-house comparison sets the two side by side.
01
Does the work end in a running system?
Or in a recommendation you still have to build. A deck is not an implementation, however good the deck.
02
Who owns it afterwards?
Code, documentation and the ability to change it — or a platform you pay for as long as the system lives. A partner who cannot answer this one is selling a dependency.
03
Can they show a measured result?
From a real deployment, with the sample size beside it, rather than a cohort claim. One honest case with its numbers beats a page of logos.
WHAT IT COSTS
A fixed fee, quoted after a call.
Each engagement is a fixed fee, quoted after a thirty-minute conversation once the workflow is known. There is no platform licence, because the system runs on your stack; the ongoing model costs are billed directly to you and are usually modest. The readiness assessment is the cheapest way to find out which engagement is the right one.
ONE MEASURED CASE
Answers in under five minutes, from about an hour.
A knowledge base built and run inside an operating company: answering a question used to mean asking colleagues, searching several systems and sometimes waiting for a meeting — about an hour of effort. Afterwards, answers came back in under five minutes, cited to the source document, across roughly 1,000 queries a month from about 100 internal users. One deployment, printed with its numbers; it is not a cohort claim.
HOW TO START
The path from the first question to a system in production.
AI Workflow Audit — two weeks
Prioritises 5–10 candidates by expected value and implementation effort, and defines the first build.
AI Automation Sprint — 4–6 weeks
Builds and deploys one scoped production system: integrations, review, monitoring, documentation, handover.
Your team owns what is delivered. Need an internal knowledge system instead? See the AI knowledge base engagement. Want the wider picture first? Read the AI workflow automation guide.
Start your free AI auditFREQUENTLY ASKED
Questions buyers ask before implementing.
What is the difference between AI integration, AI deployment and AI implementation?
Integration connects the model to your data and tools. Deployment puts the connected system into production, with review, monitoring and an owner. Implementation is the whole of it — from a working demo to a system your team runs — and integration and deployment are its two halves. A vendor quoting "integration" alone is quoting half the work.
How long does AI implementation take?
A two-week AI Workflow Audit decides which workflow is worth implementing first. A scoped AI Automation Sprint then ships one production system in 4–6 weeks; an AI knowledge base goes live in 3–5 weeks. Anything quoted in days is a demo; anything quoted in quarters is a programme, not an implementation.
What does AI implementation cost?
A fixed fee per engagement, quoted after a thirty-minute call once the workflow is known. There is no platform licence — the system runs on your stack — and the ongoing model costs are billed directly to you, and are usually modest. Prices are not published because a two-week assessment and a six-week build are different amounts of work.
Do you need an AI agent to implement AI?
Rarely. Most valuable systems combine ordinary rules, retrieval, a model call and a human review step. The least complex architecture that reliably completes the job is the right one; an agent belongs where the work genuinely needs a sequence of decisions, not where a demo of one looked impressive.
What should be tested before an AI system goes into production?
Real cases, including the ugly ones: a question set drawn from actual work, scored on whether the answer is right and whether the source supports it, plus the exceptions the demo never saw. The decision to go live is made against a number agreed in advance, not against a demo that went well.
Who owns the system after deployment?
Your team. Everything Unlocked Consulting builds is delivered on your stack with documentation, a runbook and a handover, without a proprietary platform or a required retainer. Thirty days of support follow; after that the people who change it are yours.
NEXT STEP
Find the workflow worth implementing first.
Two minutes for the diagnostic. If the opportunity is real, a senior-led assessment defines the system before anyone writes production code.
Start your free AI audit Prefer to talk first? Book a 30-minute call ↗