Scope note: This page covers custom GPT and knowledge system builds. For the assessment that identifies whether a knowledge system is the right solution, see C16 AI Tools Assessment. For workflow automation that complements knowledge systems, see C18 AI Automation Build.
"The most common failure mode for custom GPTs is building them around a vague brief. The system prompt changes the tone — the output sounds like it knows your business — but it does not actually know anything specific about how your business works. The result is output that requires the same amount of editing as a generic prompt, just with different vocabulary. The fix is not a better prompt. It is structured knowledge capture before the build begins."
Freelance SEO & AI Consultant — London
Generic wrapper vs real knowledge system
The distinction is not technical — it is structural. One changes the tone; the other changes the workflow.
- System prompt changes the tone but not the workflow
- No structured knowledge base — model draws from general training
- Output sounds specialised but is not reliably useful
- No maintenance plan — becomes stale as the business changes
- Abandoned within weeks because the output requires too much editing
- Narrow scope defined around a specific workflow
- Structured knowledge base with your processes, voice, and context
- Consistently useful output across the range of real inputs
- Maintenance plan tied to business changes
- Still in use 12 months later because it saves real time
What the build engagement involves
Four stages from use case definition to handover — with knowledge capture as the critical middle step.
Use case definition
We identify the specific workflow the GPT will change — not just the output it will produce. A GPT built around a vague brief produces vague outputs. The use case definition specifies the input, the process, the output format, and who uses it.
Knowledge capture and structuring
The hardest part of building a useful knowledge system is extracting and structuring the knowledge that makes it useful. This means capturing your processes, your brand voice, your decision rules, and your context — and organising it so the model can use it reliably.
Build and calibration
The GPT is built with narrow scope, structured context, and a repeatable process. Calibration involves testing against real use cases — not just checking that the output sounds right, but that it is consistently useful across the range of inputs it will actually receive.
Handover and maintenance plan
You receive a working GPT, documentation covering what it does and how to update it, and a maintenance plan. Knowledge systems become stale when the business changes — the maintenance plan defines what triggers an update and who is responsible for it.
Phase 02 (knowledge capture) is highlighted because it is the step most often skipped — and the one that determines whether the GPT is consistently useful or just sounds specialised.
What London SMEs use knowledge systems for
Four use cases that consistently produce the highest ROI — because they replace high-frequency, low-complexity work that currently consumes disproportionate time.
Internal Q&A and onboarding
A GPT that answers the questions new team members ask repeatedly — drawing from your actual processes, not generic documentation. Reduces the time senior staff spend on repetitive internal queries.
On-brand content production
A GPT trained on your brand voice, tone guidelines, and content examples — so the output requires editing, not rewriting. Particularly useful for teams producing high volumes of similar content types.
Proposal and brief generation
A GPT that takes structured inputs (client type, scope, budget range) and produces a first-draft proposal or brief in your format. Reduces the time from client enquiry to first draft.
Process decision support
A GPT that guides team members through a decision process — which product to recommend, which service tier applies, which escalation path to follow — based on your actual decision rules.
Fixed price per engagement, scoped before work begins. Price varies by the complexity of the knowledge base and the number of use cases. No hourly billing, no scope creep.
Not sure whether a custom GPT is the right solution? The £999 assessment identifies the highest-value AI opportunities across your workflows.
Knowledge systems and workflow automation work together — the GPT handles information retrieval and decision support; automation handles the trigger-action layer.
View all AI consulting services and how they connect — from assessment through to implementation and ongoing optimisation.
Frequently asked questions
What is the difference between a custom GPT and a knowledge base?
A custom GPT is the interface — the configured model that receives inputs and produces outputs. A knowledge system is the structured information architecture that the GPT draws from. The two work together: a GPT without a well-structured knowledge base produces generic outputs regardless of how the system prompt is written. The knowledge capture and structuring phase is what makes the GPT consistently useful rather than just sounding specialised.
Do I need to have completed the AI Tools Assessment first?
Not always. If you have a specific, well-defined use case — a particular workflow you want to change, a specific output you want to produce reliably — we can scope the build directly. If you are unsure which workflows would benefit most from a knowledge system, the AI Tools Assessment (C16) is the right starting point. It identifies the highest-value use cases as part of its analysis.
What does the knowledge capture process involve from my side?
Knowledge capture requires your time and access to the information that needs to be structured. Typically: one or two sessions to walk through the processes, decisions, and context the GPT needs to understand; access to existing documentation, brand guidelines, and examples; and a named internal contact who can answer questions during the structuring phase. The more complete the input, the more useful the output.
What platforms do you build on?
Most builds use the OpenAI GPT builder (for ChatGPT-native deployments) or a custom implementation using the OpenAI API (for integrations into existing tools or workflows). Platform selection depends on where the GPT will be used and who will use it — a GPT accessed via ChatGPT has different requirements from one embedded in a CRM or internal tool.
How do you keep the knowledge system up to date?
The maintenance plan delivered at handover defines what triggers an update (a process change, a new product, a brand refresh) and how to make it. Simple knowledge systems can be maintained internally without consultant support. More complex systems with structured knowledge bases are typically reviewed on a quarterly basis — either independently or as part of an ongoing retainer arrangement.
Build a GPT that changes the workflow
Book a 30-minute discovery call. We will review your use case, confirm whether a custom GPT is the right solution, and scope the build before any commitment is made.