Before and after configuration
Before
- You re-explain the background in every conversation
- AI answers are generic and don't fit your business
- No shared guidelines, so work is slow
- Everyone figures it out alone, and nothing can be reused
After
- AI keeps understanding your business context and its role
- Output is based on your materials and internal standards
- The team uses AI by shared rules, so results are steadier
- Knowledge, prompts and configuration can be maintained and reused
Five configuration services
Turn AI from something that answers one-off questions into a capability your team can reuse
Dedicated Project workspaces
Give AI a role, boundaries and working rules
- RoleWho it is: this sets its tone, expertise and point of view
- ScopeWhat it handles, and what must go back to a person
- Output formatWhat it hands back, whether a table, a list or a finished draft, agreed up front
- Collaboration rulesWhen it just does the work, and when it asks you first
- Everyday workflowsWeekly recurring work written as Process Sheets, ready to call up with one line
We set up a dedicated AI workspace around your business goals, covering role, scope, output format, collaboration rules and everyday workflows.
Deliverables
- Role and scope
- System prompt setup
- Output format standards
- Collaboration rules
How we know it's done
Give a colleague who wasn't involved only the configuration documents, and have them rebuild the same workspace on their own and run one everyday workflow. When the same task goes to two people and comes back in the same format, it's done.
Knowledge bases
Let AI work from your company's materials and internal standards
We organize your company's materials, product descriptions, service processes, past cases, FAQs and internal standards into a knowledge structure AI can draw on.
Deliverables
- Knowledge base structure
- Structuring your materials
- Scope of knowledge
- Retrieval tuning
How we know it's done
Ask ten real questions from your own business. Every answer can point back to the specific document it came from. When something isn't in the knowledge base, it says so instead of making up something plausible.
Prompt systems
Make every interaction more precise
We turn frequent tasks, decision criteria and output requirements into reusable prompts, so no one has to describe the need from scratch each time.
Deliverables
- A library of prompt templates by scenario
- Reuse guidelines and examples
- Output quality checklist
- How to keep improving them
How we know it's done
List your frequent tasks. Each one can be called up with one line and returns output that meets the standard. When someone else uses the same template and the quality holds, you no longer depend on one person's touch.
Context management
Keep AI up to date as your business changes
We help your team build a context system it can maintain over time, so AI keeps up with changes in the business instead of being told everything again each time.
Deliverables
- Context structure
- Update rules
- Task entry points
- Versioning recommendations
How we know it's done
Change one business rule and update it once by the maintenance rules. In the next task, AI uses the new rule without you having to mention it in the conversation.
Team AI training
Give your team shared guidelines for using AI
We help your team learn to break down tasks, set up context and judge the quality of AI output, and build AI guidelines that fit your business.
Deliverables
- Task breakdown methods
- Team collaboration guidelines
- Output acceptance criteria
- Training and Q&A materials
How we know it's done
By the end of training, everyone can write a task brief with acceptance criteria on their own, and use those criteria to check someone else's output and point out what fell short. Knowing how to check comes after knowing how to use.
We use this method every day
None of the scenarios below are polished-up client case studies. Website content, customer support replies, finance collaboration, team training: our own business runs entirely on this way of configuring AI. Every item here is part of our day-to-day.
Dedicated Project workspaces
One workstream, one AI workspace
How well AI works depends entirely on whoever is operating it. When they switch devices, switch accounts or hand over their work, everything built up evaporates.
We give every business workstream its own AI workspace: website content, customer support and finance each have their own. The workspace's role, rules and workflows are all written into configuration documents. Change the person, the device or the account, and the same workspace can be rebuilt from the documents with nothing important lost.
Knowledge bases
Customer support answers: from a veteran's memory to something AI can look up
Product knowledge, service processes and approved wording are scattered across chat logs and veterans' memories. New hires learn by osmosis, and the answers drift as people come and go.
We organized our own customer support team's product knowledge, service standards, conversation guidance and compliance wording into one structured set of documents. Every reply the support AI gives takes its wording from that set: what to say, how to phrase it and what never to say all have a source.
A knowledge base has to be governed, not just built
The real trouble starts after the knowledge base is built. The same rule exists in several conflicting copies, and when AI reads the outdated one, it answers with full confidence.
We set governance rules for our own knowledge: each kind of knowledge has exactly one original, and everything else is a copy. Information only flows out from the original, copies are logged in a register, and when the original is updated, every copy on the register is updated in turn. AI always reads the latest version.
Prompt systems
A one-key reply engine for customer support
Reply quality depends on how the support agent is doing that day. When things get busy, the wording slips.
We turned frequent support scenarios into physical shortcut keys. One press, and AI reads the wording guidelines and policy documents and drafts a compliant reply; the agent checks it and sends it. A new agent works with the wording of our most experienced agent from day one.
Turning "what we never say" into a list AI follows automatically
Compliance relied on verbal reminders. People forget, and holding someone accountable after something goes wrong is already too late.
We compiled sensitive industry wording, banned phrases and recommended alternatives into a word list, and AI checks every piece of outward-facing content against it before generating. Non-compliant wording is avoided at the draft stage, so the control happens at generation, not afterwards.
Write every job for AI as a complete task brief
Tasks given to AI in a few words get ground into shape through rounds of follow-up questions. Next time, you start from scratch again.
Every formal task we give AI is written as a complete task brief: all the source material included, the scope of changes marked out, and acceptance criteria that can be checked one by one. AI gets it right the first time, and people check against the criteria rather than by feel.
Context management
One project handbook, and AI knows the business from the start
Every new conversation means explaining the business background all over again. Leave something out and AI guesses, and wrong guesses are the hardest to spot.
We maintain a continuously updated project handbook for our own business systems: business structure, rule boundaries, common operations and strict no-go areas are all in it. In any new session, AI reads the handbook and gets to work, so its judgment rests on facts, not guesses.
AI's memory shouldn't live in the conversation
Close the conversation and AI forgets everything. Work left halfway yesterday has to be explained from the start today.
We set handover rules: at the end of each round of work, conclusions, decisions and next steps go into a memo document (an end-of-day log). At the start of the next round, AI reads the memo before doing anything (a start-of-day read). Across days, devices and accounts, the work doesn't break off, because memory lives in documents, not in conversations.
How several AIs work at once without stepping on each other
With multiple devices and multiple AIs working in parallel, the biggest worry is both sides changing the same thing. Who overwrites whom comes down to luck.
We drew up a boundary table for parallel work: which area belongs to whom, which places require a heads-up before changing, and in any conflict, stop and report rather than force a merge. The rules are in the document AI reads at the start of every session, so parallel AIs each stay in their lane.
Team AI training
A shared team course, plus guided reading
The tools are bought and the accounts are open, but everyone figures things out alone. Understanding is uneven, good practices don't spread, and bad habits spread fast.
Our own team aligns its understanding through a structured shared course: why this has to be learned, how models think, how to choose tools, how to take apart your own systems, how to write work as workflows, and finally how it all applies to each person's role. It comes with guided reading: questions people get stuck on go back in as FAQs, and the material keeps being revised as it's used. The course doesn't end when it's taught. It gets richer the more it's used.
When AI does the work, permissions need redesigning
Letting AI work directly in business systems worries people on both sides. Either you don't dare open it up and AI is stuck doing odd jobs, or you open it up and can't rein it in, and when something goes wrong no one can find the boundary.
We separate judgment from execution. AI proposes, a person decides. When it's time to act, the work goes through tiered channels: things that take a few clicks in an admin panel, things that only touch files in one's own area, and things that touch shared areas, which must stop and become a request. Each channel has its own review. Most everyday work becomes something anyone who can use the system can do, so colleagues who don't code can let AI work with confidence.
Want to apply this to your own business? Book a diagnostic and we'll help you decide what to configure first.
Our process
From the first conversation to handover training, every step is visible
Requirements call
30 minutes
Confirmed requirements
Solution design
1–3 business days
Configuration proposal and scope
Configuration
3–7 business days
Configured workspace and operating docs
Validation
1–2 business days
Test report
Handover training
1 hour
Full deliverables and a recorded training session
How we scope a project
Single service
Pick one of the five services above. The clearest delivery scope.
Combined services
Several services together, often a knowledge base plus prompt systems.
Full configuration
All five services together, including context management and team AI training.
Every project differs in the volume of materials, the business scope and how the team works together. We suggest booking a diagnostic first, then settling on the right plan and scope.
Take the 4-question self-checkLaunch isn't the finish line
Many clients ask at this point: what happens after delivery? Three answers.
We check in, unprompted
After a configuration goes live, drift tends to happen quietly. By the time you notice the output is off, the problem has been building for a while. So we don't wait for you to report it. For combined configurations, we check in on day 30 after delivery to recalibrate. Full configurations include two monthly recalibrations during the support period. Each recalibration looks at three things: is the configuration still running smoothly, does the output still match your standards, and is there new material that should go into the knowledge base.
We deliver a method, not just a configuration
Every delivery comes with operating documents and training: how to maintain the knowledge base, how to refine the prompts, how to make changes when the rules change. The method goes to your team, so it survives staff changes. You're never locked in to us.
Will it still hold up a year from now?
AI tools change fast, and configurations drift along with them. There are two ways to handle that. If you need someone keeping watch, ongoing support is an optional service: regular recalibration and keeping up with tool versions, with the scope set during the diagnosis. If you'd rather keep up yourself, the method we hand over is enough to do it. Either way works. The choice is yours.
FAQ
Write one role clearly, and it's a page.Write the whole company clearly, and it's a brain.
Build the brain first: see the Intelligence Hub