A third of our answers are no
Roughly one in three use cases we assess gets a no, in writing, before anyone signs anything. A no in week one is worth far more than a yes that quietly falls apart in month four.
AI · Data · Tech · ConsultingWe put intelligence inside your data, then build the product around it.
One operating rule: evidence over hype. We plug into the data you already produce and turn it into analysis, content, and decisions your teams act on the same day.
Twenty-five years building digital and commerce across Asia and Europe. Three industry rebrands, same problem underneath.
Cyril Drouin opened one of China's first eCommerce agencies back in 2003, recommendation engines and all, then ran Publicis Commerce and Performance Marketing across China and North Asia through the attribution years. AI is the third name for the same job: the data has to be right before the model gets to be smart. Data first, AI second.
Most AI projects do not fail loudly. They fail politely.
RAND counted it in 2024: eighty percent of AI projects get abandoned or never scale. The pilot demos well, the steering committee applauds. Then the numbers never move, nobody wrote down what success meant, and the budget quietly migrates to next year's pilot. We built the firm to be the opposite of that.
We sell no one’s software
No reseller agreements, no vendor incentives. Models get benchmarked at their actual API cost, on your task, and the recommendation rides on the numbers rather than on a partnership.
Your team runs it without us
Engagements are designed for independence. Training and handover sit inside the method, not on a separate invoice. When we walk out, the work keeps running and what we learned stays in the building.
AZIMUTH
An azimuth is the angle between north and where you are heading. Navigators take one before they move and check it as they go, because feeling on course and being on course are different facts. Six legs, a gate between each. You move on when the evidence has earned it.
- BearingDecide where AI is worth pointing, then commit it to paper.
- DataFind it, clean it, move it to where the work happens.
- BuildWhatever the use case calls for, built on your data.
- RunA small, timeboxed pilot on real data, with real users.
- FixSame measurement as the baseline. Same method, new date.
- ScalePositive return proven, the pilot graduates to production.
Baseline, metric, kill criterion. All three agreed in writing before anyone builds a thing, so that six months later nobody has to argue about what success was supposed to mean.
Data, AI, tech, consulting. Start where it hurts.
The raw material, the intelligence layer, the product layer, and the practice that ties them together. Take the whole stack or a single piece.
An honest audit of what you actually have, before anyone talks about models.
- Foundations sized to your use cases, not to a reference architecture.
- Fragmented sources plugged together: CRMs, documents, sheets, logs.
- Governance that holds up across jurisdictions, not just at home.
Use cases that pay back, prototyped on your own data with a cost per run.
- Every candidate carries its real API cost before it ships anywhere.
- Benchmarked across Western and Chinese models, on the same task.
- What misses the metric gets stopped, and we put that in writing too.
Custom builds that land inside the systems your team already opens every morning.
- Integrated into existing tools, not another tab nobody opens.
- Built by the same people who ran the pilot, so nothing gets lost in handoff.
- Documented and handed over, so your team owns what runs.
Strategy, pilots, and adoption, on dated baselines and metrics agreed in writing.
- A baseline with a date on it before any build starts.
- Leadership and team training, so the judgment stays in-house.
- The recommendation to stop, when the evidence says stop.
We benchmark both sides. On your data.
We have worked on both stacks, so we recommend a model on the merits, not on which side we happen to know. Same task, side by side, at actual API cost, and the results pick the winner.
One provider is a default. Two is a decision you can defend.
East
- Qwen
- DeepSeek
- ERNIE
- GLM
- Kimi
- Doubao
West
- GPT
- Claude
- Gemini
- Llama
- Mistral
- Grok
bearingbridge intelligence, our working platform. Modules already running on real tasks, every cost visible on screen.
We measure before we recommend.
We agree the metric before we build.
When the evidence says stop, we stop.
Data was called the new oil. Intelligence is the refined product, and AI is the refinery. Our job is to run the refinery honestly, at a cost you can see, against a target you signed before we started.
Evidence over hype. That is the whole rule.