Most AI budgets are committed before anyone has confirmed which workflow to target, whether the data supports it, or what success would even look like. Discovery answers those questions first — in weeks, against evidence, for a fraction of what a misdirected build costs.
Adoption pillars covered
Discover
Primary focus
Build
Light touch
Enable
Supporting
Govern
Supporting
Delivery models supported
Project-based
Best fit
Workshop-based
Available
Embedded engineering
Not offered
Step 01 — map it
We interview the people doing the work, quantify where time and cost actually go, then score each candidate on the value it would create and how feasible it is with the systems and data you have today. Opinions become coordinates.
Value against feasibility
Each candidate scored on the same two axes, so the shortlist is defensible rather than political.
Support ticket triage & routing
High volume, clean labels, clear baseline
Contract clause extraction
Strong value, but source documents are unstructured
Internal knowledge search
Solid payoff once permissions are modelled
Meeting note summarization
Easy to ship, modest measurable benefit
Autonomous procurement agent
Approval and audit requirements not yet defined
What comes out of it
Support ticket triage & routing
Recommended first. The volume justifies it, the labelled history makes evaluation straightforward, and the current handling time gives a clean baseline to beat.
Internal knowledge search
Recommended second, once document permissions are modelled. The value is real but the access control work has to land before anything is exposed to users.
Contract clause extraction
Highest value on the board and not buildable yet. We would rather spend a month fixing document structure than a quarter fighting it.
Illustrative candidates showing how the output reads. Yours are scored from your own interviews, volumes, and systems.
Step 02 — test it
This is where most promising ideas quietly fail, and it is far cheaper to find out now. We assess the data behind each shortlisted candidate across the dimensions that actually decide whether it can be built.
Data readiness for the leading candidate
Scored per candidate, because readiness that is fine for one use case is often fatal for another.
Access
Can the systems holding it be reached, legally and technically, without a six-month integration project?
What we found
API access exists; one legacy source needs a nightly export rather than live reads.
Quality
Is it consistent and complete enough that a model would learn the right thing from it?
What we found
Free-text fields are inconsistent across two teams; normalization is a week of work, not a project.
Coverage
Does it represent the full range of cases, including the awkward ones people actually escalate?
What we found
Eighteen months of history covers seasonal peaks and the main edge cases.
Freshness
Does it arrive fast enough for the decision the system would be making?
What we found
Near real-time. Latency is not a constraint for this workflow.
Labels & ground truth
Is there a defensible answer to compare against, so quality can be measured rather than argued about?
What we found
Resolution history works as a proxy, but roughly 300 cases need human review to build a trustworthy eval set.
Sensitivity & governance
What is in it that changes where it can be processed and who can see the output?
What we found
Customer PII is present and unclassified. Redaction and a retention decision are prerequisites, not follow-ups.
A gap is a plan, not a verdict.
Almost no organization scores clean across every dimension, and they do not need to. What matters is knowing which gaps are a week of work, which are a quarter, and which ones make the use case unbuildable until something else changes — before that becomes a delivery surprise.
Gaps sized in effort, not adjectives
Always
Blockers separated from inconveniences
Explicit
Remediation plan included
In the readout
Step 03 — decide it
A fixed shape with a fixed end date. The last session is a decision, not a status update — and the criteria that decision rests on are written down while everyone is still objective about them.
The engagement, week by week
Map the workflows
Interviews with the people doing the work, plus the volume and time data behind what they describe.
Stakeholder interviews
Workflow and friction map
Baseline measurements
Score the opportunities
Every candidate scored on value and feasibility, then reviewed with your team rather than presented to it.
Opportunity matrix
Effort and value estimates
Shortlist agreed
Probe data and feasibility
Data readiness assessed against the shortlist, and a throwaway spike run at the single riskiest assumption.
Data readiness scorecard
Feasibility spike result
Constraint and risk register
Set the bar and decide
Success criteria, sized delivery estimates, and a readout that ends in a go, a wait, or a no.
Success criteria agreed
Sized roadmap
Go / no-go readout
Success criteria, agreed before anyone builds
Every criterion needs a baseline you can measure today, a target worth the investment, and a method both sides accept in advance.
What we measure
Baseline today
Target
Measured by
Median handling time per ticket
11.4 min
< 7 min
Helpdesk timestamps, matched cohort over 30 days
Routing accuracy
82%
≥ 94%
Reviewed eval set of 300 human-labelled tickets
Escalation rate to a human
n/a
< 15%
Logged escalations as a share of handled volume
Cost per resolved ticket
$2.40
≤ $1.60
Token and infrastructure spend divided by resolutions
The readout is allowed to say no.
We write the stop conditions at the same time as the success criteria. If the spike fails, if a blocker turns out to be structural, or if the honest estimate exceeds the value on the table, the recommendation is to wait or to walk — and you will have spent weeks rather than quarters finding out.
Stop conditions written up front
Always
Outcomes we will recommend
Go, wait, or no
Cost relative to the build it protects
A fraction
Where this sits
Everything here happens before a build is funded. We touch the other three pillars only as far as the decision requires — enough to know a use case is buildable, teachable, and governable, and no further.
Discover
Uncover real value
The entire engagement. We map the workflows, size the opportunities, test whether the data supports them, and define the criteria that will decide whether the build succeeded.
Workflow map with friction and cost quantified
Value-versus-feasibility opportunity matrix
Data requirements and gap analysis
Agreed success criteria and stop conditions
Build
Take ideas to production
We build only enough to remove doubt — a throwaway spike against the single riskiest assumption. Production delivery is deliberately a separate decision, made after this one.
Feasibility spike on the hardest unknown
Sized delivery estimate
Reference architecture sketch
Enable
AI competency for every team
Your team runs the assessment alongside us, so the prioritization method stays behind after we leave and the next candidate can be evaluated without us.
Scoring framework your team keeps
Working sessions with stakeholders
Readout to leadership
Govern
Scale without losing control
Risk is cheapest to handle before the architecture exists. We surface the data, privacy, and compliance constraints early enough to shape the design rather than block it.
Data sensitivity and residency review
Regulatory constraint map
Early risk register
How we deliver it
Discovery only works if it costs materially less than the build it is protecting. That shapes how we sell it: a fixed engagement with a fixed end date, or a workshop that hands your team the method outright.
Project-based
A scoped outcome, shipped
A fixed-scope assessment with a fixed end date. Interviews, data review, a feasibility spike, and a decision-ready readout.
Engagement shape
Typical length
3–5 weeks
Starts with
Workflow interviews
Ends with
Go / no-go readout
What it includes
Stakeholder interviews and workflow mapping across the target area
Opportunity matrix scored on value and feasibility
Data readiness assessment with a gap plan
Success criteria, sized estimates, and a prioritized roadmap
Embedded engineering
Senior engineers inside your team
Discovery is a decision, not an ongoing capacity need, so we do not sell it as embedded time.
Why not, and what instead
An open-ended embedded engagement removes the forcing function that makes discovery useful
If you want continuous opportunity assessment, that belongs in an embedded delivery engagement
We are happy to run discovery first and continue as embedded engineers on what it recommends
Workshop-based
Hands-on teaching on your stack
A facilitated cohort that teaches your team to run this assessment themselves, using your own candidate use cases as the material.
Engagement shape
Typical length
1–2 weeks
Format
Facilitated sessions
Run on
Your use cases
What it includes
The scoring framework taught on your real candidates
Data readiness review performed by your engineers
How to write success criteria that survive contact with delivery
A repeatable intake process for the next set of ideas
Discovery is designed to be a small bet
It should cost a fraction of the build it informs, and it should be able to end with a recommendation not to build. If it cannot do both, it is not discovery — it is the first invoice of a project that was already decided.
What you leave with
The output is not a strategy deck. It is a ranked set of opportunities, the evidence behind the ranking, the constraints that will shape delivery, and the numbers that will tell you afterwards whether it worked.
Clarity
The right problem, chosen deliberately
Candidates are scored on value and feasibility against real workflow data, so the shortlist reflects evidence rather than whoever argued hardest.
Confidence
The risky assumption already tested
The hardest technical unknown is probed during the assessment, so the build starts without the question that usually derails it.
Accountability
Success defined before spending
Baselines, targets, and measurement methods are agreed while everyone is still objective, rather than negotiated after launch.
Protection
Bad ideas stopped cheaply
The assessment can conclude that nothing here is worth building yet, which is the least expensive moment for that to be true.
Signals we help your team move
We make the invisible platform work visible enough to prioritize, fund, and improve.
Start with the question, not the budget
Bring the workflow that keeps coming up in planning, or the AI idea nobody can quite size. In a few weeks you will know whether it holds, what it would take, and how you would know it worked.
Fixed scope and a fixed end date
A recommendation, not a strategy deck
Success criteria agreed before any build
We will tell you when the answer is no
The discovery brief
What we look at, and what lands on your desk at the end of it.
We inspect
The workflows and the people running them
Volume, handling time, and cost of the status quo
Data access, quality, coverage, and sensitivity
Integration, risk, and compliance constraints
You receive
Scored opportunity matrix and ranked shortlist
Data readiness scorecard with a sized gap plan
Feasibility spike result on the riskiest assumption
Success criteria and a sized delivery roadmap
Three to five weeks, fixed scope, and a readout that ends in a go, a wait, or a no. Whichever it is, you own the artefacts.