We measure first.
No recommendation without a baseline. Opinions are welcome in the room. They just do not close a phase.
About
We were doing this before
it was called AI.
what the market calls it
Deep learningMachine learningAI
// same field, mostly the same math. The vocabulary moved faster than the technology did.
01 Who we are
Senior consultants and data engineers who have watched this field change its name three times.
input
nul42???okdup
output
nul42???okdup
Every model, in every era, is a function that turns input into output. Wrong input, wrong output, delivered with total confidence. Garbage in, garbage out survived every rebrand because it was never a slogan. It is arithmetic.
It was called deep learning. Then machine learning. Then, once LLMs got good enough that a board member could use one, it became AI. What did not change is the job underneath. We make sure AI is applied where it will hold.
Most engagements start in the plumbing: reconciling sources, fixing joins, tracing where a number is really born, finding the field three teams populate differently. It is unglamorous, it is where the value is, and it is the part nobody demos.
We do not sell dreams. We sell the work that makes the dream survive contact with your data.
02 Where we come from
Different industries, one recurring shape.
A business sitting on data it does not trust, being sold intelligence it cannot verify.
- Media & advertising dataattribution, reach, spend
- Supply chain optimizationforecasting, routing, stock
- Financial analysisreconciliation, risk, reporting
- Market analysisdemand, pricing, share
0 m
Intelligence, on top
Now cheap and excellent. Everyone sells this. Not the hard part.
⇕
−∞
Your data, underneath
Still yours, still messy, still the thing that decides whether any of it works.
We have spent our careers on the foundation, which is why we can be honest about the layer.
03 Founder & partner
He sells the work and does the work.
Twenty-five years building digital and commerce operations across Asia and Europe. Owning P&L and international growth targets rather than advising on them. The through-line is not a title. It is the same problem in three costumes: commerce data, media data, and now model data. Each wave arrived with a new name and the same unfinished foundation.
- 2003
Founded one of China's first eCommerce agencies
Recommendation engines, when they were called recommendation engines.
- 2010s
CEO, Publicis Commerce & Performance — China & North Asia
Attribution, when the industry still believed last-click.
- Now
Founder & Partner, BearingBridge
Model data. Same unfinished foundation, new name.
The rest of the bench publishes here as each bio earns a concrete, checkable fact. No stock photos, no "passionate about." If a bio cannot end in a fact, it stays off the page until it can.
04 We build, not just advise
Consulting is the practice. It is not the whole company.
bearingbridge intelligence takes data, content, and documents from wherever they live, across sources that were never designed to talk to each other, and adds intelligence on top. The first application we shipped on it measures how visible your brand is on Google and inside AI assistants. One use of the platform, not the definition of it.
It answers the question every executive is actually asking underneath the AI question:what would this change about something I already have?Not a greenfield fantasy. An existing dashboard, made measurably better. "Should we do AI" is not a question with an answer. This is.
the shape is the argument
CRMdocssheetslogsCMS
bearingbridge intelligence
intelligence, on top
A platform that skipped the first half would be a demo. It runs on the methods we sell: dated baselines, written metrics, costs on screen.
05 On models
Certified on some. Trained on others. Loyal to none of them.
Certification means we passed the vendor's exam. It does not mean we owe the vendor a recommendation, and we hold no reseller agreements with any of them.
We benchmark on your tasks, with your data, at the API prices you would pay yourself. Anyone who tells you one model is best has stopped measuring, or was never measuring.
The winner changes by task. It changes by quarter. Try it. →
model benchmark
1Claude0
2GPT0
3Qwen0
4DeepSeek0
// illustrative of how rankings move, not a published benchmark. Cost is a real variable in the trade-off, on screen while it runs.
How we work
Values, stated as behaviors.
We say no in writing.
About a third of the use cases that reach us should not be built. The written no is the deliverable clients quote back to us years later.
We start in the plumbing.
The first phase is usually data, not models. If that disappoints you, we are the wrong firm, and you should know that before the invoice, not after.
We end in your autonomy.
Every engagement and the platform itself are designed so your team runs without us. Our retention strategy is being worth retaining.
We ship what we sell.
The methods on this site run our own product. When we say something survives a real P&L, we mean ours.
Independence
We hold no reseller agreements with model vendors, agencies, or data providers. Benchmarks are paid for at the same API prices you would pay, and recommendations carry no commission. Your team sees everything, runs everything, and owns the data. When an engagement ends, nothing about your operation should end with it.
Related work
The work, with the stops left in
Three cases that show how we run: constraints written before the design, a monitoring stream dropped after a six-week trial, and a document type abandoned because checking it cost more than keying it.
Group finance
A shorter close, with the ledger untouched
Coding, matching, and commentary drafting are pattern work. The ledger stayed off limits by design. Constraints were written before the build, and auditors were briefed early.
Nothing touches the ledgerRead the case→
Strategy
Market monitoring in two languages
Chinese-language sources were invisible to a team of four. Ingestion across two languages, weekly briefings with dated primary sources, and one monitoring stream we dropped.
Social listening dropped after six weeksRead the case→
Procurement and supply chain
Supplier documents matched before receipt
Mismatches surfaced at goods receipt, far too late to act. Extraction, three-way matching, and an exception queue moved the discovery forward to the confirmation stage.
Handwritten extraction droppedRead the case→
July 202612 min read
The job apocalypse hasn’t started. That’s what worries me.
AI hasn’t destroyed jobs yet, but after building agents across six business functions, I see companies quietly freezing hires instead of firing anyone at all.
Read the piece→
July 202610 min read
Your buyers decide before the click
Organic clicks fell while AI referrals started converting better. What AEO and GEO each buy, who should own them, and how an executive measures both.
Read the piece→
July 20268 min read
You can buy the platform, not the context
AI work is moving from bespoke builds to assembled platforms. What that changes for buyers, what it does not, and what to ask before signing.
Read the piece→Talk to us
Talk to us about AI.
A conversation with the senior team about your markets, your data, and where AI would actually pay back for you. No slides, no obligation, and if the honest answer is that AI is not your next move, you will hear that too.
Contact us→
Prefer email? hello@bearingbridge.com