A baseline you can date.
Without one, progress is a feeling, and feelings do not survive a budget review.
Method
AZIMUTH. First a bearing, then proof.
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. That is the method: decide where AI is worth pointing, know where you stand today, agree where you are going, then measure the difference on a schedule nobody can argue with.
It fits on one page, which is deliberate. A method your team cannot remember is a method your team will not use.
Bearing
Deciding where AI is worth pointing, then committing it to paper.
We start at the top, with the people who own the P&L, and find the use case, sometimes two, where AI would move a number that matters. Then we write it down: the baseline, the success threshold, the timebox, and the kill criterion, the result that means we stop.
That last part is the one most methods skip. It is the one we will not.
Data
Finding the data, cleaning it, and moving it to where the work happens.
Most AI work stalls here, quietly, so we do it in the open. We find which data the use case needs and where it lives, judge its state, then clean it and build the pipes that carry the right data, in the right shape, to where the work runs.
If the data is not there, we say so, plainly, before a feature is built.
Build
Whatever the use case calls for gets built, on your data.
Now we build the actual system: an AI product, an agent or set of agents, a dashboard, a retrieval layer over your own documents, an automation. Every AI and data cost sits on screen as it runs, so economics get tested next to quality.
The hard part is scope. The build tests the bearing, not a wish list. New ideas wait.
Run
A small, timeboxed pilot on real data with real users.
The pilot runs on your real data with the people who would use it every day, because synthetic data and volunteer testers give you synthetic results. No black boxes and no “trust the process”: anyone on your side with a login can watch the run happen live.
What you learn here is whether it holds up in the real workflow, not the demo.
Fix
The same measurement as the baseline, same method, new date.
In navigation, a fix is a confirmed position: not where you feel you are, where you actually are. We repeat the baseline measurement, same method, and set the two numbers side by side. The gap is the result. The fix also reconciles the money: the pilot’s cost against what scaling costs at real volume.
A pilot that works but cannot pay for itself has failed.
Scale
Positive return proven, the pilot graduates to production.
Scale is the phase every other phase is protecting. The bearing was met and the economics held: a positive return at real volume, not a working demo. It moves to production engineering, hardened and handed over, with your people trained to run it.
Short of that, two honest outcomes remain: adjust the bearing and run again, or stop, in writing.
// the pilot
Real data, real users, every cost on screen as it runs.
The principles behind the phases
Four rules the phases enforce.
Six phases are the shape of the work. Underneath them sit four rules that do not bend, whatever the engagement.
A metric agreed before the work, not after.
Success criteria invented once the results are in will always be met. That is why they are worthless.
Costs you can watch as they run.
The economics belong in the evidence, not in a surprise invoice in month eight.
An ending where you no longer need us.
Every engagement is built to hand your team the keys. That is the deliverable.
These four rules have not changed since we wrote them down, and we do not expect them to. The pages around them are reviewed every quarter.
Where the method came from
We did not design this in a workshop.
AZIMUTH grew engagement by engagement, out of failures we watched from inside marketing and technology leadership.
- The pilot with no baseline.
- The success metric invented after the numbers came in.
- An AI budget nobody could reconcile.
- The build that ran for months before anyone thought to check whether the data existed.
Each phase is here because we watched its absence sink a project. Then we ran our own platform on the same rules, and that is where the method had to survive a real P&L, which was ours.
FAQ
Questions we get.
Is AZIMUTH an acronym?
No. It is the navigation term, picked because it means something exact: a bearing you commit to and check. We left it as one word on purpose.
Does every engagement really follow all six phases?
The phases scale with the work, and so does the cost. A two-week audit runs them lightly and is priced that way. A production build runs them formally. What never gets skipped, at any size, is the dated baseline and the written bearing with its kill criterion.
What if the kill criterion fires and we want to keep going?
It is your budget and your call. What you get from us is the evidence and a written recommendation, so the choice to continue is made with open eyes instead of momentum.
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