What we do · 04 · AI Design
AI assistants, AI-made films and images, brand-tuned models and the training to use them safely.
Illustration: a studio photograph behind the record for one generated image. It shows the request, the brand-tuned model used, a brand score of 0.93 against a 0.85 threshold, cleared rights and no personal data. Approval from a named creative director is still pending.
What we offer
You can start with any one. All four share the same models, evaluation sets and approval rules.
01 / 04 · Assistants and agents
AI assistants that show sources, take corrections and hand over to a person.
Typical 6–12 weeks to a working prototype
Explore AI Application Design
02 / 04 · Generative media
AI-made films, images and edits, with creative direction and a rights log.
Typical 4–8 weeks to a working studio
Explore AI Content Studio
03 / 04 · AI Design
Image and copy models tuned to your brand, scored on a brand-fidelity test set.
Typical 6–10 weeks to a tuned model
Explore Brand AI Tools
04 / 04 · Measured AI adoption
AI use policy, pilots, roadmaps and training for brand and marketing teams.
Typical 6-week diagnostic · 12-week pilot
Explore AI Strategy & ConsultingHow it works
Each output goes to the model that scored best on your examples and is checked for brand, rights and safety before review.
Request
Spring launch: a hero still, product copy in four languages, a voice read and 8-second film cut-downs for India, the UAE, the UK and Germany.
Eval floor 0.85 · below it, a lane never reaches review.
Claude · brand voice prompt set
Top brand-voice score on the 120-prompt set
4 × product copy · 4 languages
Brand-tuned Flux model · v4
Only candidate above the brand floor
12 stills · 4:5, 1:1, 16:9
ElevenLabs · licensed brand voice
Consented voice, contract on file
4 × 20 s reads
Script says “clinically proven”, which is not on the approved claims list.
Veo · storyboard-locked
Best motion consistency on the brand set
8 × 8 s cut-downs
Human approval
1 lane flagged. Voice: script says “clinically proven”, which is not on the approved claims list.
Approve stays locked while any lane is flagged. The approver can send a lane back, never wave it through.
Audit log
Illustrative run. The models chosen change from brief to brief.
Assistant design
We design for four moments people meet in live assistants, so they can check an answer, fix a mistake or stop an action.
Mock: an assistant answer with two numbered citations and a medium confidence meter.
01 · Uncertain
Citations and a confidence level sit on each answer, so people can check the claims that matter.
Mock: a streaming answer showing what the system is reading, elapsed time and a stop button.
02 · Slow
Answers stream in, progress stays visible and stopping keeps the partial result, so waiting feels shorter.
Mock: an answer where the person has corrected a date in place, with an undo control and a note that the correction was logged.
03 · Wrong
People can edit in place, undo or reach a person, and each correction is logged for the next evaluation set.
Mock: an agent plan of three steps; the last step, sending an email to 1,240 people, waits for approval.
04 · Consequential
Agents show their plan, and any step that spends money, sends messages or publishes waits for a person.
How we use AI
We re-run the tests whenever a model or dataset changes, and keep four controls around every AI task.
We write test prompts and scoring rules from your brand guidelines, then score every candidate model and update against them.
Brand, rights, claims and personal-data checks run on every output, built to the OWASP Top 10 for LLM Applications.
Approval is a fixed step in the workflow: the approver sees what was flagged and why, and cannot skip it.
Model, version, prompt, inputs, scores, checks and approver are stored with each asset and exported with content credentials.
Illustrative scores. Your comparison runs on your own brand material.
Brand AI Tools compared
Brand Design writes the brand rules and the checks that enforce them. AI Design tunes models on your brand material. Teams producing content in volume can use both, linked at the scoring step.
Identity, voice, tokens and rules.
Trained on your brand, scored against its rules.
Every asset checked, from any source.
Signed, logged, published.
| Question | AI Design · Brand AI Tools | Brand Design · Brand AI Tools |
|---|---|---|
| What it is | Image, product and language models tuned on your rights-cleared material. | Rules, checks and templates that keep every output on brand. |
| What it produces | New assets that already look and sound like the brand. | A verdict on any asset from any source: on or off brand, and why. |
| Built with | Flux, Adobe Firefly, ComfyUI and open-weight model families, tuned and scored. | Your brand guidelines, written as tokens, rules and a brand-check agent. |
| You own | The dataset, the scoring set and the model weights. | The rulebook, the templates and the checking workflow. |
| Start here when | Generation is off brand, slow or dependent on a few people. | Output from many teams and tools is going off brand. |
| Brand-tuned models | Brand governance tooling |
How we work with you
A typical engagement runs eight to twelve weeks. The order does not change: we agree how success is scored before anything is generated.
We agree who uses it and what it may see and do. Its limits, failure cases and evaluation set are written before any feature.
We build on a live model from the first week, because static designs cannot show how AI answers feel in use.
We score candidate models on your material and test with your users, including prompts the model gets wrong.
We set up automatic checks, approval roles and the audit record, and write down the regulatory position.
Your team runs it: specifications, prompts, weights, evaluation sets and a playbook for the day a model changes.
Platform stack
The technologies we work with, in four layers. The prompts, evaluation sets and checks around each model stay in place when the model changes.
L1
Chosen per task on the evaluation set.
L2
Image, film and voice, and the models we tune.
L3
Agents, workflows and the product around the model.
L4
Where the work is designed, stored and measured.
Frameworks we build to
Our work is built to or aligned with these frameworks. They shape the evaluation plan, the automatic checks and the records we hand over. We do not claim certification against them.
Transparency duties for generated content and chatbots, and a risk classification written down for each system.
Our AI management follows its structure: roles, impact assessment and a record of every model in use.
Govern, map, measure, manage: the structure of our risk register and evaluation plan.
Automatic checks tested against the LLM Top 10, starting with LLM01 prompt injection.
AA for every AI interface: streamed output announced, agents keyboard-operable, captions on generated film.
Lawful basis, minimisation and no personal data in training sets without it.
India’s Digital Personal Data Protection Act applied to prompts, logs and datasets.
We apply each standard only where the work involves it.
What you get
Prompts, evaluation sets, model weights, check settings and approval records sit in your repositories and accounts from the first week.
01 · AI Apps
02 · Content Studio
03 · Brand Models
04 · AI Strategy
What changes
Each measure is baselined in the first weeks and reported from the audit log. The figures below are typical ranges; your targets are set per engagement.
01
Share of tasks users complete on a fixed prompt set, including cases where the model is wrong.
Measured by Moderated sessions, before and after
02
Share of generated assets that meet the minimum brand score with no manual edit.
Measured by Scorer + raters on every output
03
Median time from brief to signed approval.
Measured by Timestamps in the audit log
04
Model spend plus review time for each asset released.
Measured by Model invoices + logged review time
05
People using the new workflow every week, three months after the pilot ends.
Measured by Weekly active users ÷ licensed seats
06
Edits a person makes before accepting an AI answer or asset.
Measured by Edit events per accepted output
Services & packages
Services from all four AI Design capabilities, each bought as a short sprint, a fixed-scope project or ongoing capacity. Choose one and your enquiry reaches the right team with it attached.
How to buy
How we work with you
Ask a quick question, send a project brief or issue a formal RFQ. The lead for the work reads each one in full, and any services already in your brief go with it.
Or book a thirty-minute call01
For a first conversation, a press request or anything that does not need a scope yet.
You get A reply from a named lead
02Recommended
Goals, audiences, a budget band and timing, so our first reply can outline the work.
You get Options and a first scope after one call
03
Your documents, deadlines and the procurement and security rules the work must meet.
You get Receipt confirmed and a named bid lead
How it is priced
Each package shows its pricing model. Work starts once a written scope and quote are agreed.
Opens a project brief with this package chosen.
In your brief
In your brief
In your brief
In your brief
In your brief
In your brief
| Package | Every engagement includes | Best for |
|---|---|---|
| SprintA short, fixed-scope engagement that answers one defined question. |
|
Discovery, a diagnostic, a prototype or a decision you need to make soon |
| ProjectA defined scope, delivered for a fixed price. |
|
Work you can describe up front: an identity, a platform or a set of tools |
| MilestoneA larger build in phases you approve and pay for one at a time. |
|
Programmes too large for one contract, where you want control at each step |
| RetainerReserved monthly capacity to run, improve and extend what we built. |
|
Live brands and products that need a steady team without hiring one |
| EnterpriseA multi-workstream programme with a dedicated team, governance and agreed service levels. |
|
Large organisations running change across markets, portfolios or business units |
| SquadA dedicated team that works inside your stack and sprint schedule. |
|
Teams with a clear roadmap that need more senior people quickly |
Questions
Answers on models, ownership, rights, regulation and where your data is kept. For anything else, ask the team.
Book a call
The ones that score best on your own examples. For language we work with Gemini, OpenAI, Anthropic, Mistral and open-weight models. For image, film and voice we use Flux, Adobe Firefly, Runway, Veo and ElevenLabs. Each task also has a fallback model.
You do. What we make for you is yours once it is paid for. Tools we already had stay ours, and you get a free, permanent licence to use them. A tuned brand model is yours to run, retrain or retire.
AI Design shapes what people see and use; Technology & Intelligence builds the software underneath. We design the conversation, the controls, generated content and brand-tuned models. Technology & Intelligence covers agents, AI that answers from your documents, infrastructure and testing in production. One project can use both.
Ours makes on-brand assets; Brand Design’s checks them. We build image and language models tuned on your brand material. Brand Design builds the rules and checks that test any asset, from any source, against your guidelines. The two meet at the scoring step.
We clear every input before use. That means licensed or owned imagery and consented voices, never scraped material. Each output carries content credentials and a record of the model, inputs and approver. If a platform’s terms limit commercial use, we tell you before choosing it.
Yes, its transparency duties can. People must be told when they are talking to a chatbot, and AI-generated images, audio and video must be marked as such. We classify each system at the start and design disclosure into the interface.
Yes. We use enterprise plans with training on your data switched off. Where material may not leave your network, we run open-weight models in your own cloud account, including India regions.
You will speak to a lead who would run the work, and get a straight answer on fit.
Three ways to start
Every engagement starts with a written scope and a quote agreed before work begins.
Choose one of the three ways above
