Practice / Practice Heatmap
Practice Heatmap
Where AI effort has gone across the practice, and why it moved: from a secure place to use LLMs in 2025 to knowledge, finance and marketing agents in 2026.
follow the heatIn 2025 the visuals were already running on ComfyUI, so the focus went to LLMs. We looked at building our own system and judged it too heavy to build and maintain, so we chose OmniChat to give everyone a secure place to use LLMs.
Through the year both grew: prompt and system-prompt libraries on the LLM side, and visual workflows that became the most prominent use in the practice.
A secure place for LLMs
Building our own system proved too heavy, so we chose OmniChat: one secure, managed place for everyone to use LLMs.
In use
Prompt & system-prompt library
Reusable prompts for writing, editing, UK spelling checks and converting exports into clean Markdown.
In use
Visual workflows
Already running on ComfyUI, then extended into templates on newer platforms. Visuals grew into the most prominent use.
In use
First writing prompts
Prompts that know the practice’s voice, for project texts and social posts.
In use
With the visual workflows largely set, 2026 shifted to knowledge and finance, then marketing: the website, journals and socials, with visuals feeding in. The systems now connect rather than sit side by side.
The bigger change is how we work. From Q2 the practice moved from a prompt and system-prompt culture to agentic workflows and agents, starting with knowledge agents.
Building Regulations agent
Agents for the Approved Documents, with a custom agentic workflow that translates the regulations’ diagrams into usable guidance.
In development
Project agents
Each live project gets its own agents: one for meeting minutes, one for emails, one for Design & Access Statements, planning applications and replies.
In development
Fees & practice data
The Fee Calculator benchmarks proposals against past projects; data agents for fees, timesheets and finance are next.
In use
Website & writers
A website revamp, plus writer agents for journals, news and socials, working with the visual workflows as one connected system.
In development
Shading shows relative focus, judged rather than measured. Dashed cells are planned.
Scoring every idea
33 ideas, three scores each
Ideas from across the practice were logged and scored with the partners for simplicity (10 = very simple), risk (1 = errors are not a concern) and impact. One score ranks them: ROI = simplicity × impact ÷ (25 × risk). Individual ideas stay internal; the pattern is below.
Design & visualisationResearch & technical deliveryDocuments & knowledgeMarketing & business developmentFinance & operations
| Category | Ideas | Simplicity | Impact | Risk | Avg ROI |
|---|---|---|---|---|---|
| Design & visualisation | 2 | 8.0 | 6.0 | 2.5 | 0.78 |
| Research & technical delivery | 4 | 7.2 | 8.8 | 5.4 | 0.49 |
| Documents & knowledge | 13 | 8.5 | 5.5 | 2.4 | 0.82 |
| Marketing & business development | 9 | 5.1 | 7.4 | 3.1 | 0.48 |
| Finance & operations | 5 | 4.2 | 6.6 | 5.0 | 0.21 |
- Quick wins sat in documents. Prompt tools scored simple and low-risk, which is why the first year leaned on them.
- The big prizes are hard. Fee, finance and planning-data agents score high impact but low simplicity and higher risk: they need integration, not just a prompt.
- That gap is the 2026 plan. Moving to agentic workflows is how the high-impact, harder ideas become buildable.
How it is used
Heatmaps and scores are reviewed with the partners and the AI, marketing and finance groups, and re-scored as tools change. The earlier opportunity map on the Strategy page shows where this started.