- Sheridan, Wyoming, US
- [email protected]
- Call Us Anytime+123 (4567) 890
- Email Us[email protected]
- Opening HourMon - Sat 8:00 - 6:30, Sunday - CLOSED
Get latest update for our trusted applications
Insights
What 120 value assessments have actually taught us.
Not thought leadership. These are the findings that keep repeating across engagements, including the ones that were unflattering to us and the ones that reduced the size of our own proposals.
120+
assessmentsThe evidence base for these findings
18
industriesAcross which the patterns repeat
5
findingsThat change how we scope work
We publish what we have measured, not what we have read.
Our industry produces an enormous volume of content about artificial intelligence, most of it a restatement of vendor announcements. We have deliberately kept this page small: five findings, each drawn from measured outcomes across multiple engagements, each of which changed how we scope work.
Two of them reduce the size of our proposals and one of them is an argument for spending less money with us. That is roughly the ratio you should expect from findings that are genuinely derived from evidence rather than from marketing.
The most useful thing they told us cost them the larger version of the project.
The adoption factor is the number your business case is missing.
A system used by 60% of intended users returns roughly 60% of its modelled benefit. Almost no business case includes this, which is why so few of them materialise.
Across 120 value assessments the single most common cause of a business case failing to materialise is not model quality, integration difficulty or scope creep. It is that the modelled benefit assumed universal adoption and the actual adoption was 40 to 70 per cent.
We now apply an explicit adoption factor to every benefit line before quoting anything, which reduces our own proposals by roughly a third. It also changes what we build: adoption becomes a design constraint with a measured target rather than a training exercise scheduled for the week before launch.
0.7
typical factorAdoption discount applied to modelled benefit
If you have never run a holdout, you have never proved anything.
Every client who has kept a control group has discovered at least one initiative that was quietly destroying value. Every client without one has an unfalsifiable benefits case.
Holdouts cost a small, deliberate amount of upside. In exchange they answer the only question that matters at renewal time: would this have happened anyway? Seasonality, market movement, a competitor's mistake and a pricing change will all flatter a benefits case that has no control group.
The uncomfortable finding is consistent. In our portfolio, every client who has maintained holdouts across a portfolio of interventions has found at least one that was net negative — usually a retention or discounting programme aimed at customers who were never going to leave.
19%
spend removedRetention discounting to customers who would stay
Foundations should be load-bearing, which implies something standing on top.
The eighteen-month data platform that has never served a decision is the most expensive failure pattern in enterprise data, and it is a sequencing error rather than a technical one.
We have reviewed programmes several million dollars into delivery with elegant lakes, beautiful lineage and two hundred pages of governance documentation, where not one business decision was being made differently. The teams were competent. They had been asked to build a foundation for everything before anything was allowed to depend on it.
Restarting from a single funded question — which customers to call today, and why — produced a working decision in seven weeks, after which the platform grew behind live use cases that each paid for the slice beneath them.
7 wks
to a live decisionAfter restarting a stalled 18-month programme
Autonomy is a business decision, not a technical capability.
The question is never whether an agent can complete a task. It is which decisions your risk function will allow a machine to make, and what happens at the boundary.
Every agentic deployment we have put into production was gated by a policy conversation rather than by model capability. Which actions are permitted, what value threshold forces human approval, which validators must pass before an action commits, and what the escalation carries with it.
Teams that treat this as compliance overhead ship slowly and get blocked. Teams that treat it as the design specification ship faster, because the objection that stops most AI launches has already been answered before anyone asks it.
68%
fully automatedTasks completed without human touch, within policy
A third of your inbound demand may never have been answered.
In four separate engagements across property, hospitality, education and lending, unanswered inbound contact was the largest single value line in the model — and in every case it was a surprise to the client.
Businesses instrument their marketing funnels obsessively and their telephone systems almost never. The result is that spend continues to be optimised at the top of the funnel while a fifth to a third of high-intent contact is lost at the point of answering.
It is the least sophisticated finding in our portfolio and reliably one of the most valuable. Before commissioning any AI programme, log your inbound contact for two weeks and count what went unanswered.
31%
unansweredShare of inbound calls lost, property client
Five findings, and what they should change about your next decision.
120+
Value assessmentsThe evidence behind these findings
1 in 5
Concluded: do not proceedWritten recommendations to stop
$1.7B
Client value generatedWhere the findings were applied
Where to go next.
Test these findings against your own numbers.
The two-week assessment applies exactly this method to your business: adoption factored, baselines verified, holdouts designed and an honest verdict at the end.
