Simform's CTO ties a 42% AI abandonment rate to funding capability before data foundations

Two numbers from the same survey do different jobs, and the essay leans on both. The 42% counts companies that dropped most of their initiatives before production; the 46% counts proof-of-concept projects that never got out ^[2][4]. Year on year, the company figure rose 25 percentage points, roughly two and a half times the earlier rate ^[21]. A rise that size is consistent with more than one story. A portfolio being pruned after a year of wide experimentation would also produce it. The essay gives no reasons from the surveyed companies, no name for the manufacturing client and no figure for its rework ^[22].

The explanation comes from Hiren's own engagements. In many enterprise AI programs he has seen, he writes, the model is not what blocks production: the data was not trustworthy and the platform was not reliable ^[6]. He evaluates readiness in three dependent layers, a governed data foundation, a reliable and observable platform, and governed agentic systems on top ^[7].

The manufacturing engagement is the most specific evidence in the piece. Production-scale capital went into AI capability before the data foundation that should have supported it, and the foundation investment was not approved until months into the program ^[10]. By the time the gap surfaced, nothing the model produced could be trusted, because the data lineage was missing ^[11]. Engineering then spent months working around a foundation that had never been built, absorbing unplanned effort the original schedule had never accounted for ^[12].

In my view the sturdier half of the argument is the incentive claim, which does not depend on the survey at all. Visible AI capability competes better for capital than invisible infrastructure, Hiren writes. Model licences and agent capabilities go in first because they produce quick results leadership can point to before the quarter ends ^[8]. A working agent is easier to demonstrate than data lineage, access controls or governance maturity ^[9]. Whoever sets that order sets the architecture ^[13].

His remedy is procedural. A readiness gate governs scale, meaning how much production investment, traffic, autonomy or scope a layer gets, not whether teams can experiment ^[16]. The data gate opens when the team can confirm authoritative sources, lineage, freshness, access controls and ownership for the information the workflow depends on ^[17]. The platform gate opens when latency, availability, observability, cost per transaction and recovery time all sit inside a predetermined operating envelope at expected peak load plus headroom ^[18]. The agentic gate opens when the accountability questions have written answers before the first workflow goes live. Those questions: what the agent may do, when a human takes over, whether consequential actions can be traced or rolled back, who owns a wrong decision and what remediation looks like ^[19].

Hiren concedes the cost: correct sequencing reduces visible feature velocity early, because some investment goes into reusable dependencies instead of deployable capability ^[14]. What he offers against that cost is the rework bill, framed as a time-to-return argument: "Here is what we are betting on, what it requires and what the rework will cost in engineering time and delayed returns if the dependency is not there," he wrote ^[15]. Neither side of that comparison is quantified here. The practical ask stays a narrow one: pair each capability investment with the dependency it requires, the evidence the dependency is ready, and the cost of skipping it ^[20].