It comes as a solution, not as a problem. Nobody declared a value hypothesis or a baseline, because the conversation already started at the tool.
«Let us put AI in customer service.»
Services · AI Value Discovery
A quarter of a team costs more than assessing first.
Turns a demand for AI into a value hypothesis, a solution design and a decision criterion. Two to seven weeks, depending on complexity, and you leave with a recorded decision: build, test or stop, with the math open beside it.
Forty minutes, free and with no commitment. They settle whether the assessment makes sense, how complex your case is and how long it takes. If it does not make sense, we say so.
About the h|ve Platform
The assessment does not end in a document nobody reopens.
Every h|ve service runs on the same platform. Here, the value hypothesis, the owners of the decision and the acceptance criteria are importable and become executable context: whoever builds, with us or without us, starts with the agreement already in place. The platform is not a prerequisite: the service works on its own and the artifacts are yours, in an open format.
What this service touches
- Map · context
- Flow · orchestration
- Guard · barrier
- Judge · deviation
- Ledger · record
The hypothesis, the baseline and the owners go into the Map. The decision record, with author and basis, goes into the Ledger.
What stays on the platform after this service
The value hypothesis with your baseline, the acceptance criteria with target values and the decision record. It is the base the next service starts from, whether Scan & Go or FDE Velocity, instead of starting by asking again what the company wants.
The problem
The demand arrives with the answer already built in.
Someone saw a demo, read a news story, heard it from a peer at another company, and the request that reaches leadership already comes with the solution chosen. The question nobody asked is which problem it solves and what that is worth.
The thing works. Without a baseline there is no comparison, so the assessment becomes opinion, and the opinion of whoever sponsored it is predictably positive.
«It is running fine, people liked it.»
The honest answer is that there is no way to know. Whoever answers that loses next year’s budget, and the next initiative arrives the same way.
«What was the return on what we invested in AI?»
This cycle breaks at a single point: before approval, by declaring the criterion.
The asymmetry
An assessment whose recommendation is always to build is not an assessment.
Assessing costs a fraction of what building costs: weeks of assessment against a quarter of a team. Most of the value of this service lies in the times the recommendation is not to build, and that recommendation is a legitimate result of the work, not a failure of it.
How it works
Four steps, and each one only ends when someone of yours signs.
Map, measure, design and decide are four verbs any consultancy would use. What makes them a method is three rules, and the exit gate of each step, signed by someone at your company, never by us.
The process is counted, not described
How many times a month, how many people touch it, how long each handoff takes, how many go back. It comes out of the tracker, the repository or the system where the process runs. Interviews explain the record, they never replace it.
The number comes out as a range, with the input declared
Every output is a range, never an exact value, because part of the input is declared by whoever answers. Each line of the math carries the value, where it came from and what would make it change.
The design is the minimum that tests the hypothesis
It is not the full architecture. It is the smallest thing able to confirm or knock down the value hypothesis. That limit is what separates this service from Scan & Go and from FDE Velocity.
- 00Framing40 minutes, free, before anything starts
Exit gateThe decision is open, there is someone who can say no, and there is a record to count. If it does not pass, the assessment does not start, and we say why.
What comes outThe complexity level of your case, the calendar and what needs to be ready on your side.
- 01Map2 · 3.5 · 6 days, depending on complexity
Gate · the snapshotThe mapping numbers are recognized by the people who operate the process. The area leader signs. If it does not pass, we go back to counting.
What comes outThe process in countable units, with the source of every number, and the slice where there is volume, repetition and a stable rule.
- 02Measure2 · 3 · 5 days
Gate · the baselineThe cost and volume assumptions are confirmed in writing by whoever answers for the budget. Without that, the math is not defensible and does not move forward.
What comes outCurrent exposure and possible recovery, both as ranges, with the assumption and the sensitivity beside each line.
- 03Design1.5 · 2.5 · 4 days
Gate · the acceptance criteriaWhoever will build agrees that each criterion is verifiable. The technical leader signs. If it does not pass, we rewrite the criterion, never the value.
What comes outDesign of the minimum slice, requirements, acceptance criteria with target values and the AI budget.
- 04Decidehalf a day to one day, in a meeting
Gate · the decisionThe decision was made and recorded in the meeting, with the person who can say no present. The record is written there and signed before everyone leaves. Without that, the delivery is not complete.
What comes outThe decision record, with the criterion that held it up and what would have to change for the answer to be different.
No phase produces new material for the next phase to explain. The final report does not repeat what the phases produced, it points to them. An assessment that ends in a hundred slides costs three times more to make and is read by fewer people.
The model
Any process, any industry, five questions.
The math is the same for customer service, legal, finance or engineering. What changes is where each number comes from. The fourth input is the one most assessments forget, and it is where most AI cases stop adding up.
- Volume
How many times the process happens a month, counted in the record and not estimated in the meeting.
- Unit cost
What each occurrence costs today: person time times hourly cost, plus direct cost if there is any.
- Addressable share
What part of each occurrence AI can absorb, with the mechanism named. Without a named mechanism, the share is a wish.
- Review load that remains
The minimum quality for the result to count, and what the human review that still exists costs. It is where most AI cases die.
- Cost to operate
Model, integration, maintenance and governance, added up over the year.
The math, in one line
net = (volume × unit cost) × addressable share × (1 − review load) − cost to operate
The range comes from applying the declared uncertainty to the addressable share and to the review load, which are the two inputs nobody is sure about at the start. A process where AI gets it right ninety percent of the time looks excellent, and may be worth nothing: if every output has to be checked by a person, that person still reads one hundred percent of the cases, and the gain is only the difference between reading and writing.
- Benefits that cannot be counted.Satisfaction, morale, image, being ahead. They go into the report as text, named as not quantified, and are never added to the number.
- Saved hours that do not become capacity or money.Saving two hours a week for fifty people gives a hundred hours on paper and zero in cash, unless someone says what will be done with them.
- Gains from another initiative.When two initiatives touch the same process, the gain does not add up. Adding what is not measured contaminates what is.
The three exits
Build, test or stop.
The decision meeting presents all three with the math for each one, and the decision is made right there, with the person who can say no in the room. The record is the artifact that still holds when the person who decided is no longer at the company.
The hypothesis holds and the path is clear. It moves on to Scan & Go or to FDE Velocity, depending on whether the bottleneck is governance or capacity, or to the team you choose.
The record carries the baseline and the acceptance criteria with it, so the delivery can be checked afterwards, whoever builds it.
The hypothesis holds and evidence for one piece is missing. Out comes a small experiment, with a declared criterion, a short deadline and a stop condition.
When the evidence on the addressable share or the review load does not exist, the correct recommendation is to test, not to build.
The math does not add up, the process has no volume, the rule changes too much, or the data does not exist. It is the result that saves the most money, and the one that rarely reaches you, because it does not generate the next project.
You keep the record of what was assessed and the open model, to redo the math the day the volume or the data changes.
When the exit points to a h|ve service, the decision record says so in those words.
Build and test can go ahead with h|ve or with any other team. The record names that dependency in writing, so the decision stays yours and whoever reads it months later knows exactly what was recommended and by whom.
What stays with you
Seven artifacts, short, and the editable math.
Everything in an open format, including the calculation model, so the company can redo the math on its own when the volume changes. Three pages is not a limit on effort, it is a limit on reading.
- The decision record
One page. What was decided, the criterion that held it up, and what would have to change for the answer to be different.
Who uses itWhoever needs to understand the decision months later, including when the answer was to stop.
- The business case
Three pages: the problem told in numbers and the math as a range; the five inputs with source and sensitivity; the three exits with the math for each one.
Who uses itWhoever decides, and the committee.
- The open model
An editable spreadsheet, one tab of inputs and one of math. You test your own doubt on the spot and apply the same skeleton to the next initiative on your own.
Why it mattersIt turns the service into an installed capability, not an opinion.
- The process snapshot
Up to four pages, with the source of every count.
Who uses itWhoever operates the process, to check.
- The requirements of the slice
Proportional to the minimum slice that tests the hypothesis, as a document or as cards.
Who uses itWhoever will build.
- The acceptance criteria
One page. Each criterion is an indicator with a target value and a moment of verification.
Who uses itWhoever builds and whoever approves.
- The AI budget
One tab: model, integration, maintenance and governance, over the year.
Who uses itWhoever answers for the budget.
How long it takes
The highest variable counts, not the average.
Four variables set the complexity of your case, and with it the effort and the calendar. Company size, industry and the value of the investment do not come in. A five-thousand-person company with a well-recorded process in a single area is level one. An eighty-person company with four areas and no record is level three.
| Variable | Level 1 | Level 2 | Level 3 |
|---|---|---|---|
| Areas involved | One | Two or three | Four or more |
| Process record | Exists and is reliable | Exists in part | Almost nonexistent |
| Scope assessed | One process | One process with variants | Several processes, or a queue to rank |
| Who decides | One person | A committee | Committee plus risk or compliance |
- Effort
- 6 days
- Window
- 2 to 3 weeks
One area wants to automate a task it already measures.
- Effort
- 10 days
- Window
- 3 to 4 weeks
A process that crosses two areas, with a partial record.
- Effort
- 16 days
- Window
- 5 to 7 weeks
A queue of initiatives competing for the same budget.
What moves the calendar lives on your side: the schedule of the people who operate the process, access to the record (it goes through security, and that takes from two days to three weeks) and the schedule of the decision meeting. Booking that meeting two weeks ahead, in the very first week, saves more time than any gain in the earlier phases.
Who it is for
Who it fits, and who it does not.
People reach this service by different paths and ask for the same thing: permission to decide without depending on personal conviction. The framing exists to say in the first conversation when the assessment does not serve, and where to go instead.
FitsWhen it makes sense
- There is more demand for AI than capacityThe queue needs order, and you want a comparable criterion across the initiatives competing for the same budget.
- An initiative was already tried and stalledAnd nobody can explain why. Without a baseline, the explanation becomes opinion.
- The year’s budget is being put togetherAnd the AI line needs a justification: a value hypothesis that survives the committee’s first hard question.
- Someone outside technology asked about the returnOn what has already been spent. The committee wants a baseline and an acceptance criterion declared before the next execution.
Does not fitAnd where we point you
- The decision has already been madeYou want a document that backs what you are going to do anyway. The path is Scan & Go or FDE Velocity directly.
- There is nobody who can say noWithout someone with the authority to stop it, the assessment becomes retroactive justification, which is a different and worse product. In that case the assessment does not start, and we say so in the framing.
- There is no record of the process at allWithout volume or time anywhere, mapping becomes collection first. Collection is a separate piece of work, under that name, and not hidden inside level three.
- The pilot is already runningAfter that the service loses its main function. Governed Method Packs, to measure and adjust what already operates.
Who delivers
The two partners, from the framing to the decision meeting.
Whoever runs the framing is whoever runs the assessment, and the part that cannot be delegated is the judgment of two things: which share of the process is really addressable, and how much human review remains afterwards. You talk to the same two people from the first day to the last, and nobody has to retell the story halfway through.
André JB RodriguesCo-founder and Business Lead
Runs the forty-minute framing, the process snapshot with the people who operate it, the business case and the decision meeting.
Gustavo HartzCo-founder and Technology Lead
The technical judgment of inputs three and four, the design of the minimum slice and the acceptance criteria, reviewed with whoever will build.
Questions
The questions of whoever signs.
The ones the committee asks before approving, and the ones you would ask before starting.
What if the recommendation is not to build?
It is a legitimate result of the service, and the one that saves the most. The decision record explains what was assessed and what would have to change for the answer to be different, and the open model stays with you to redo the math when the volume changes. When the exit points to a h|ve service, the record says so in those words.
What if we already know what we want to build?
The service still serves to declare the acceptance criterion and the baseline before starting, which is what makes the delivery checkable afterwards. If the decision is already made and closed, the path is Scan & Go or FDE Velocity directly, and we say so in the framing.
How long does it take?
Two to three weeks at level one, three to four at level two, five to seven at level three. The complexity level is set in a forty-minute conversation, before anything starts. Before that, any deadline would be a guess.
Where did that number come from?
It is the question the committee will ask, and every line of the math already answers it: the value, the source of the input and what would make it change. The cost and volume assumptions are confirmed in writing by the company itself before the math moves forward, so the first line of defense is that the number is yours.
Do we need the h|ve Platform under contract?
No. The service works on its own, and the artifacts are yours in an open format. When the platform is there, the hypothesis, the owners and the acceptance criteria come in as executable context, and whoever builds starts with the agreement in place. It is an advantage, not a prerequisite.
The next step
Start with the framing conversation.
Forty minutes settle whether the assessment makes sense, how complex your case is and how long it takes. The four questions beside this are the ones we will ask; answering them now already moves the conversation forward.
We reply within one business day.Rather write? hello@hybridvelocity.com