FishDog Research studies · Technology / Budgets
Shared research study link · Updated May 27, 2026

IT Budget Decision-Makers: SaaS vs AI Tools vs Custom Builds

How enterprise IT budget decision-makers are rethinking software spending across three competing vectors — existing SaaS subscriptions, emerging AI-native tools, and custom-built internal solutions. Which SaaS categories are most vulnerable, how build-vs-buy has shifted, whether AI spend is additive or cannibalising, and the CFO/board narrative for 2026.

Headline finding · Q3: build versus buy

Ten budget owners.
A dead heat on build vs buy.

The panel converges on almost everything — flat budgets, cannibalised AI spend, the same vulnerable categories — except the question that decides where the money lands: four consolidate into suites, four now build narrow tools in-house, two are frozen by validation and privacy law.

4 buy and consolidate — suite-native AI wins 2 constrained — validation and privacy freeze the calculus 4 build-lite converts — internal builds beat vendor SKUs

Q3 of 7: "Has AI changed the build-versus-buy calculus for your team? Two years ago, building a custom…"

Position spectrum · Q3

Where each respondent stands

Hover any respondent to read their position.

◄ Buy & consolidateBuild it in-house ►
RN
JC
NA
AE
JM
AM
RM
MM
MH
AO
Dot color follows the position groups above. Position summaries are condensed from the study record.
The argument

The suite against the four-week build

The consolidator case
~60% of our budget motion is top-down consolidation and ROI mandates. Suite-native AI, contractual protections, one-in one-out.
Raymond Navarro · IT manager, Chino CA — with James Cheng (AI 2%→19% funded by consolidation) and Nicholas Ausbie alongside. Condensed from the study record.
VS
The builder case
A four-week internal build replaced the vendor SKU at ~60% lower unit cost, 90–95% accuracy, with a small human review lane.
Aaron Monahan · CIS manager, Franklin TN — with Matthew Hughes, Miles Murad and Ryan Maciel running build-lite pilots. From the study record, near-verbatim.
Adoption conditions

What vendors must prove before the money moves

One row per respondent, one column per condition they voiced. Column totals rank the market's demands.

Respondent Data-use guarantees
(no-train, residency)		 KPI-gated pilot
with ROI offset		 Cost predictability
(caps, FinOps tags)		 Suite-native /
consolidation		 Validation &
auditability
Ryan Macielpractitioner · San Jose CA
Matthew Hughespractitioner · Austin TX
Nicholas Ausbiesenior IT · Hollywood FL
James Chengsenior IT · Boston MA
Aaron Monahansenior IT · Franklin TN
Raymond Navarrosenior IT · Chino CA
Jeremy Millerregulated · rural NJ
Abigail McclungK-12 education · rural NY
Miles Muradpractitioner · Oakland CA
Apryl Ellisonpractitioner · rural MO
demanded by 8/1010/103/108/108/10
  voiced this condition   blocker — a veto features can't buy past   not surfaced
Consensus and contest

How the panel divides across all seven questions

Questions ranked by how strongly the panel diverges.

Q3
Build vs buy — the panel's core fault line: consolidators vs build-lite converts vs constrained sectors
Q6
Procurement change — everyone tightened; intensity diverges from clauses to full validation paths
Q2
SaaS replacement — most are pruning point tools at renewal; regulated and education won't swap
Q1
Budget shift — flat budgets, AI to mid-teens share via reallocation; education is the lone structural outlier
Q4
Additive or cannibalising — almost entirely cannibalised; one partial fresh allocation, in risk/security
Q5
Vulnerable categories — near-total agreement: notes, grammar, BI viewers, forms, async video compress first
Q7
CFO narrative — "hold flat and reallocate," outcome-backed deployments only

The split is structured by role and sector:

Senior consolidators
3 respondents · 50s, IT leadership
The gatekeepers. Translate CFO risk posture into suite-native consolidation with contractual teeth.
Hands-on practitioners
4 respondents · 35–41
The switchers. Drive workflow-level change and internal RAG/triage builds even when procurement is cautious.
FinOps-exec builder
1 respondent
The proof. Wins the build argument with hard unit economics, then caps and tags everything.
Regulated + education
2 respondents
The frozen calculus. FDA/ISO validation and FERPA make the whole debate moot — augment inside validated systems only.
The panel

Who answered

10 respondents recruited from a census-grounded synthetic population of 340,000 U.S. residents — enterprise IT decision-makers, including regulated (FDA/ISO) and K-12 public-education voices to capture the conservative outliers.

Respondents
10 recruited
from a census-grounded synthetic population of 340,000 U.S. residents
Updated
May 27
2026 · 7 questions · 70 individual responses
Panel
8M · 2F
ages 35–57 · median household income ≈ $158K
Contexts
4
enterprise IT · analytics · regulated engineering · K-12 education

Panel income vs. U.S. households

A senior panel, deliberately — budget owners, not end users. Benchmark: Census ACS 2022 (B19001).
Under $50K
0%
35%
$50K – $100K
30%
29%
$100K – $150K
10%
17%
$150K – $200K
30%
9%
$200K+
30%
12%
This panel (n=10) U.S. households

Behind every respondent

Each profile opens into its full evidence trail.
Every respondent carries a grounded biography, a personality profile (OCEAN), an ingested media diet — this panel reads the New York Times, BBC Business and Technology, NPR Technology, the Washington Post, the Houston Chronicle and more — the recent news they actually read and how it shaped their views, and a voice profile. The roster below leads with role, region, age, and the budget stance that shaped each respondent's answers.
James ChengBoston, MA · 57 · M
Senior IT · Ops manager
ROI-first consolidator: AI share rose ~2%→19%, funded by suite consolidation and seat reclamation.
Raymond NavarroChino, CA · 57 · M
Senior IT · Manager
Consolidation-first, no margin for avoidable risk; buy/build posture shifted ~90/10 to ~80/20.
Nicholas AusbieHollywood, FL · 57 · M
Senior IT · CIS manager
Procurement-clause driver: no-default training, logging, model transparency; fresh budget only in risk/security.
Aaron MonahanFranklin, TN · 47 · M
Senior IT · CIS manager
FinOps-savvy builder: 4-week internal build beat a vendor SKU by ~60% on unit cost; tags and caps all LLM spend.
Ryan MacielSan Jose, CA · 39 · M
Practitioner · Project manager
Quick-ROI operator: moved transcription to suite AI; CFO funding is conditional on measurable productivity within a quarter or two.
Matthew HughesAustin, TX · 41 · M
Practitioner · Systems analyst
"I'd rather compare specs than chase trends" — leads build-lite RAG, triage and doc-intake pilots under strict governance.
Miles MuradOakland, CA · 39 · M
Practitioner · Operations research analyst
"I trust evidence over hype" — measurable build-lite pilots held to CFO-grade proof of productivity.
Apryl EllisonRural MO · 35 · F
Practitioner · Systems analyst
Cancelled most Grammarly Business seats once Microsoft Editor + Copilot sufficed; AI held to productivity framing.
Jeremy MillerRural NJ · 49 · M
Regulated · Engineering/quality lead
The FDA/ISO conservative: validation, traceability and provenance; AI augments validated systems, never replaces them.
Abigail McclungRural NY · 42 · F
K-12 education · Operator
The education outlier: SaaS ~93% / AI ~3%; FERPA, offline reliability and grant dynamics block public LLMs for student data.
Findings

What they said, with the evidence attached

10 OF 10 NAMED
AI spend is cannibalised, not additive — and every dollar needs an offset
Budgets are flat. AI grew from low-single digits to mid-teens share, funded by SaaS consolidation and seat right-sizing. James Cheng: ~2% to ~19%.
The one unanimous mindset on the panel — all ten named
Lone partial exception: Nicholas Ausbie — fresh allocation only in risk/security, under board pressure
→ read the ten responses behind it
4 OF 10 NAMED
Build-lite now beats vendor SKUs on narrow problems
A 4-week internal build replaced a vendor SKU at ~60% lower unit cost, 90–95% accuracy, with a small human review lane.
Aaron Monahan — CIS manager, Franklin TN
Build-lite camp: Hughes, Murad, Maciel; Navarro quantifies the posture shift at ~90/10 → ~80/20
→ read the four responses behind it
6 OF 10 NAMED
Per-seat point tools die first
Meeting transcription, writing/grammar, lightweight BI viewers, simple forms, async video — cut at renewal wherever suite AI reaches "good enough." Apryl Ellison cancelled most Grammarly Business seats.
From the study record
Named to this stance: Maciel, Cheng, Navarro, Murad, Ellison, Hughes
→ read the six responses behind it
5 OF 10 NAMED
Procurement grew teeth: AI is a gating factor at every renewal
60–90 day KPI-gated pilots; no training on customer data; retention and residency controls; prompt/output logs; model-version transparency; hard spend caps.
The new standard clause set, from the study record
Named to this stance: Ausbie, Maciel, Miller, Monahan, Hughes
→ read the five responses behind it
Appendix

Full narrative

A1

Objective & context

We set out to understand how enterprise IT budget owners are reallocating spend across existing SaaS, emerging AI-native tools, and custom builds; which categories are most exposed to AI; how build-vs-buy has shifted; whether AI spend is additive or cannibalising; and the CFO/board posture for 2026. Insights synthesize seven question areas across IT leaders spanning technology operators, regulated industries, and public education.

A2

How allocations have shifted

Budgets are largely flat, but the mix is changing. AI-native has grown from a tiny base into a meaningful line item, typically mid-teens, funded by SaaS consolidation and seat right-sizing rather than net-new dollars. James Cheng reports AI rising from ~2% to ~19%; Raymond Navarro attributes ~60% of the motion to top-down consolidation and ROI mandates. Custom/internal work remains material but has pivoted from greenfield apps to integrations, data plumbing, governance, access controls, and lightweight automations. Outliers include education (Abigail Mcclung: SaaS ~93%, AI ~3%) and one org keeping custom spend high but focused on governance.

A3

Build vs buy in an AI era · where AI disrupts

AI has shifted "buy by default" to pragmatic "build-lite" for narrow, internal, low-risk workflows (document intake/extraction, support triage, RAG/knowledge search, analytics helpers). Speed-to-value and per-seat vs usage economics tip decisions toward assembling internal tools; productionization, SLAs/SSO/auditing, and regulatory risk keep core systems with vendors. Raymond Navarro quantifies a posture shift from ~90/10 to ~80/20 (buy/build-lite). Aaron Monahan demonstrates the upside: a 4-week internal build replaced a vendor SKU with ~60% unit-cost reduction at 90–95% accuracy and a small human review lane.

Teams are pruning edge, single-purpose SaaS where suite-native or AI-native reaches "good enough." Most frequent cuts: meeting transcription/notes (Ryan Maciel moved to Zoom/Meet AI), writing/grammar (Apryl Ellison cancelled most Grammarly Business as Microsoft Editor + Copilot sufficed), lightweight BI viewers, simple forms/surveys, async video/screencast, and small KB/chat add-ons. Core systems (ERP, CRM, HRIS, ITSM, QMS/ALM) remain durable due to auditability, change friction, and identity/governance constraints (Ryan Maciel; Jeremy Miller). Education and medical devices are especially conservative (FERPA, FDA/ISO).

A4

Procurement posture & persona correlations

AI is a gating factor in renewals, not a checkbox. Standard practice: 60–90 day, metrics-driven pilots on first-party data; consolidation-first at the edges; and elevated requirements for data-use guarantees (no default training, retention/residency, BYOK), model governance (version pinning, provenance, change notifications), auditability (prompt/output logs, RBAC, export), and cost predictability (caps, token metering). CFOs treat AI as a conditional investment: net-neutral unless measurable productivity, ticket deflection, or revenue lift is proven within a quarter or two (Ryan Maciel; Miles Murad). Security/identity/governance are protected spend; LLM usage is FinOps-managed with tags and caps (Aaron Monahan). Partial fresh allocations appear only in risk/security under board pressure (Nicholas Ausbie).

  • ROI-first consolidators (Cheng, Navarro): drive suite-native AI, seat reclamation, one-in/one-out funding.
  • Platform technologists (Hughes, Maciel, Ellison): lead build-lite pilots (RAG, triage, doc intake) under strict governance.
  • Regulated engineering (Miller): validation and traceability required; AI augments but does not replace validated systems.
  • Education K-12 (Mcclung): extreme constraints (FERPA, offline reliability); SaaS share rises, AI remains bundled and minimal.
  • Gov/security-mature IT (Monahan, Ausbie): high custom for governance; request ML SBOMs and AI incident playbooks.
A5

Recommendations, next steps & measurement

Recommendations

  • Consolidate Wave 1: target notes/transcription, writing/grammar, BI viewers, forms, async video. Use renewal windows to cancel/downgrade where suite AI is "good enough."
  • Stand up a pilot factory: 60–90 day templates with KPI gates for L0/L1 support deflection, RAG/enterprise search, and doc intake/extraction.
  • Codify build-lite blueprints: reference architectures with model pinning, eval harnesses, RBAC/ABAC, and rollback.
  • Centralize AI governance: LLM gateway with prompt/output logging, DLP, region/retention controls, and BYOK; stream to SIEM.
  • Negotiation playbook: co-term, consolidation credits, downgrade rights, transparent metering, and export guarantees to neutralize AI upcharges.

Next steps & KPIs

  • Run a 14-day seat/SKU audit and publish a renewal triage sheet (consolidation test, displacement test, AI-upcharge sanity).
  • Enable suite-native AI in the office/conferencing stack; schedule decommissions for overlapping tools.
  • Launch the pilot factory with baseline KPIs and a hard kill switch; cap and tag all LLM spend.
  • Deploy the LLM gateway and AI DPA addendum before scaling any pilots.
  • KPIs: vendor-count reduction in edge categories (15–25% in 2 quarters); AI wallet share of software mix (15–20% in 12 months, net-neutral); pilot pass rate (≥60% meeting KPIs in 90 days); seat/tier savings (8–12% run-rate reduction in 2 quarters); governance coverage (≥90% of AI flows behind gateway/logging).

FishDog · Research without respondents. Study updated May 27, 2026 · 10 recruited respondents · 7 questions · 70 responses. Prototype note: position placements, stance counts, and matrix cells are illustrative pending response-level stance data; quotes, demographics, segment analyses, and the appendix narrative are from the study record.