Inside Software 03.01.26 - Congrats, you're a leader, and somehow, you're still behind…
Every other week we’ll provide updates on the latest value levers and trends operators are asking us about in Technology and Software. If there are things you want to hear more about - shoot us a note.
Similar to our CIO post a few weeks ago, we’re fortunate to run a number of longitudinal market studies to understand how priorities (and spending) evolve over time. One we’ve been watching especially closely is overall GenAI adoption- across all industries.
This week, we’re sharing the latest readout from Bain’s longitudinal GenAI executive survey, and grounding it in what matters most: not the hype, not the demo - the reality of whether companies are still tinkering… or putting AI into production in ways that show up in real workflows. The headline? For the Taylor Swift fans out there, we’re entering a new era: we are graduation from the “pilot era” to “production era” - and it’s happening fast...particularly for software companies.
One quick note: because this research spans all industries (not just software), it won’t map perfectly to the software‑specific signals we’ve been sharing from our CIO longitudinal survey. Think of this as the “macro weather report” - and we’ll call out where tech/software is moving earlier (or differently) than the rest of the market.
No duh…the prioritization of AI is increasing…
In just over a year, the share of companies ranking AI as a top‑three strategic priority jumped from ~60% to ~74%. And the share calling AI their #1 priority more than doubled - from ~9% to ~21%.
This is the tell that matters: AI isn’t an “innovation team” thing anymore. It’s showing up in budget conversations, operating plans, and exec scorecards.
Adoption is basically ubiquitous and spreading into more domains
The “are you experimenting?” question is over. ~94% of firms say they’re using GenAI in some capacity. And usage keeps creeping into more functions.
A good example: pilots of software development tools rose from ~66% to ~73% over the last year. Similar gains show up in customer service, knowledge-worker efficiency, marketing, IT, and other domains.
Pilots are graduating to production (and software is the front edge)
One of the louder myths right now is: “Everyone’s piloting, nobody’s scaling.” The data says otherwise.
Across the market, the share of use cases moving into production has increased - on average ~22% vs. ~14% a year ago. And the leading use cases are clearly defined:
Software development: ~40% of pilots are moving to production at scale
Customer service: ~32%
A solid second tier (sales, marketing, knowledge-work efficiency, etc.) sits in that ~20%+ “scaling” zone
And the tech/software industry is out in front: tech firms have ~1.7x more use cases in production than non‑tech firms across both internal workflows and customer-facing experiences.
The new bottleneck is security (and it spikes once you scale)
Here’s the catch: scaling changes the risk profile.
Most barriers are slowly easing year over year. Expertise concerns fell (~44% → ~39%) and quality concerns fell (~44% → ~32%). But data security and privacy moved in the wrong direction, rising (~38% → ~45%).
Even more telling: companies already in production are much more worried than those still piloting:
Security concerns: ~49% (production) vs. ~35% (pilot)
Quality concerns: ~37% (production) vs. ~24% (pilot)
This is exactly what we see with our clients: the moment you wire AI into real workflows, you’ve got real data, real users, and real threat models.
Results are real, but ROI is still concentrated
Among the ~59% of companies that say they’re meaningfully adopting GenAI, executives report that ~80% of use cases met or exceeded their expectations - while exciting, we see this as more of a satisfaction test, not the same thing as proven ROI. Why? Because, only ~23% of all respondents can tie GenAI to measurable revenue gains or cost reductions.
And when GenAI doesn’t meet expectations, the failure mode is telling: it’s rarely “the model didn’t work.” It’s often that the AI system + operating model weren’t built to handle the real‑world weirdness - edge cases, messy data, and the hundreds of ways a workflow behaves outside the pilot environment. This is where shadow mode (and gradual rollout) becomes consequential: you surface the weirdness early, harden the system, and avoid the dreaded “thrown over the wall” failure. Roughly one‑third say the use case worked in pilots but didn’t scale, and another one‑third say it was more expensive than anticipated.
The most interesting nuance: satisfaction tends to rise as companies move from “AI as an assistant” to task automation and then to agentic workflow automation. The closer you get to end‑to‑end automation, the more upside you can unlock - but the harder it is to do safely.
Before we even get to tech vs. non‑tech, the leader/laggard gap is widening. ~55% of “leaders” have a defined vision, use‑case list, and roadmap, while ~40% of laggards don’t have a defined vision. Leaders also run far more experiments across use cases - and their worries shift from “is this worth it?” to “how do we govern it?” (security, quality, and regulation). Quick check: Which one are you? Leader or Laggard?
The tech check: Tech vs. non‑tech (and why benchmarking broad is a trap)
Now, zooming in on tech vs. non‑tech, a few deltas stand out:
More investment (and more people). Across the full sample, GenAI budgets are up ~78% YoY and the average company is budgeting ~$10M annually. Tech respondents are budgeting ~$13M (up ~90% YoY) vs. ~$6.5M in non‑tech (up ~50% YoY). Tech also has a higher share of people spending all of their time on GenAI (~32% vs. ~23%).
More distributed ownership. Non‑tech companies still tend to centralize the GenAI agenda in IT (>50% centralize the GenAI agenda with IT). Tech spreads ownership across AI/ML & data science, engineering/R&D, IT, and strategy - more leaders at the table, focusing on outcomes
More “DIY” building patterns. Overall, companies skew ~63% off‑the‑shelf (e.g. packaged AI features in existing applications) vs. ~37% DIY. Tech builds DIY at a meaningfully higher rate (~40% vs. ~32%). And the most common DIY architectures are prompt engineering, followed by RAG and agentic AI (all ~40%+ of builds) versus fine‑tuning / pre‑training (each <30%).
…The lurking concern: disruption risk is rising
While all these data points suggest the tech market is moving faster than the rest of the economy, there are some lurking concerns - starting with disruption risk. Over the last six months, the share of companies saying GenAI poses a “very high” disruption risk more than doubled (~5% → ~11%). In tech, the signal is even louder:
~17% of tech companies see very high disruption risk (vs. ~5% non‑tech)
~44% see high or very high risk (vs. ~36% non‑tech)
…The implication for software companies: don’t benchmark vs the market
That maps to what we’re all feeling: as agentic AI moves from concept to reality, the definition of “software company speed” is getting rewired.
And here’s the implication we don’t want software leaders to miss: don’t benchmark your AI progress against the overall market. Your competition isn’t a CPG company running a handful of pilots - it’s your tech peers, shipping faster, automating more, and resetting customer expectations in real time.
So yes, you may be ahead of the average company, BUT you should be aware of disruption risk - and use that paranoia productively: disrupt your own business model before someone else does. The goal is to be running faster than the rest of your peers - not just the “general market.”
Congrats, you’re a leader. And somehow, you’re still behind. Because in software, “leading” isn’t a badge—it’s a timer. The minute you slow down, your peers turn your differentiator into table stakes. The opportunity is massive if you can harness GenAI to rewrite your own product and cost structure. The risk is just as real if someone else does it first.
Stepping back...what this means for software leaders
Productize “pilot → production.” Buyers don’t need more features - they need deployment patterns: governance, security, controls, and proof that it works at scale.
Sell the workflow, not the widget. “AI-powered” is table stakes. The win is: what process gets faster/cheaper/more reliable - and how do you measure it?
Design for the security reality. If security/privacy is the #1 barrier, your product story needs to include guardrails, auditability, data boundaries, and clear ownership.
Stay on the bleeding edge - and benchmark against your software peers. Don’t grade yourself against the cross‑industry average. Your real competition is other software and tech companies - so keep raising the bar, and disrupt your own category before someone else does.
Note: The opinions expressed in this article are my own and do not represent the views or specific recruiting practices of Bain & Company. The information provided is believed to be from reliable sources but no liability is accepted for any inaccuracies.