BTS/AI Barometer 2026/Executive Pulse
OCTOBER 2026 BASELINE
THE STATE OF AI AT BTS · EXECUTIVE EDITION

Adoption is here.
Impact comes next.

BTS has already embraced AI. The strategic challenge is to convert widespread individual use into deeper, safer and measurable business value.

127 surveyed13 departments25-question technical assessment

121 OF 127 EMPLOYEES

AI maturity↗
72/100
Healthy self-reported maturity
Depth of use◫
66/100
Main opportunity for greater value
Losing momentum↘
30people
24% expect to use AI less
Critical Shadow AI!
15people
12% at high exposure to uncontrolled use

Organizational AI readiness

Selected indicators, scored from 0 to 100

ATTITUDE & USE
Source: Part 1, indicator overview, pp. 7–8.

AI workforce profile

No AI skeptics in the survey

95% ACTIVE
PIONEERS
5745% of workforce
EXPLORERS
6450% of workforce
OBSERVERS
65% of workforce
The adoption battle is effectively over. The next phase is about use-case quality, engineering capability and responsible scaling.
Source: Part 1, employee types, pp. 9–10.

Future-use momentum

Intent to use AI more, equally or less

EARLY SIGNAL
Building momentumFuture intent exceeds present use
7
StableCurrent and future use aligned
90
Losing momentumPlan to use AI less in future
30

What leadership needs to know

Three signals that change the conversation

Momentum erosion

11 of 20 employees in Market Management anticipate reducing AI use. Purpose-built use cases are a priority.

Expert-led Shadow AI

The 15 critical exposure cases are concentrated among confident users, not inexperienced beginners.

Engineering constraint

Expert design and deployment scores 34.4/100 against 87.7/100 on fundamentals.

Sources: Parts 1 & 3, pp. 7–14, 25–29.
THE STRATEGIC RESPONSE

Three moves. One coordinated direction.

01

Govern what is already happening

Approved-tool catalogue, security guardrails and quick review for advanced use.

02

Move beyond superficial tasks

Expand AI into substantive workflows and measure changes to output and quality.

03

Rebuild the momentum

Target Market Management, close knowledge gaps and activate internal champions.

BTS AI Barometer • Executive Report • October 2026

02 / CRITICAL SIGNALS

Where attention is needed.

AI adoption is widespread, but the findings reveal concentrated loss of momentum, uncontrolled use and uneven capability. These are leading indicators, not confirmed incidents.

Parts 1 & 2 · pp. 10–25
Disengagement signal
24%
30 of 127 expect to use less AI
Market Management
55%
11 of 20 losing momentum
Critical uncontrolled use
12%
15 people in critical risk band
Risk concentration
80%
12 of 15 cases across 3 departments

Who is losing momentum?

Percentage of respondents by major department

MARKET MGMT. OUTLIER
Counts: Market 11/20, Sales 8/32, Finance 2/8, Product & IT 5/26, Software 3/23. Rates rounded in the report.

Critical Shadow AI exposure

15 individual critical-level classifications

POTENTIAL RISK
Important distinction: The assessment identifies exposure to unsafe or unapproved practices, not 15 verified data breaches.
Source: Part 1, uncontrolled-use analysis, pp. 12–14.

Company-wide exposure distribution

Risk-level classification of 127 surveyed employees

61 · 48%Low risk
51 · 40%Moderate risk
15 · 12%Critical exposure

Four issues worth board-level attention

Recommended actions from the report
01

55% momentum loss in Market Management

Run a hands-on workshop to discover actual market-analysis use cases. The study hypothesizes a relevance problem, which should be validated with the team.

02

Concentrated unsupervised AI use

Start with Product & IT Operations, Sales and Software Development, which account for 12 of the 15 critical exposure cases.

03

Financial data is a sensitive hotspot

The report flags a high-exposure individual case within Finance. Handle it privately through Security and the appropriate management channels.

04

Targeted intervention beats generic campaigns

Give technical leaders a secure fast lane, and those struggling or losing confidence role-specific coaching rather than generic messaging.

Interpretation guardrail: “Losing momentum” means people report future-use intention below current usage. It is not evidence of a historical decline. Likewise, a risk classification signals exposure; it does not establish a verified incident or causation.
03 / DEPARTMENT INTELLIGENCE

One company. Different realities.

Explore every department, compare scores and uncover the actions specifically recommended for each team.

Parts 1 & 2 · pp. 14–25
AI execution benchmarkSoftware Development

Highest depth among major departments · 82.6

Momentum priorityMarket Management

55% losing momentum · 11 of 20

Uncontrolled-use hotspotProduct & IT Operations

5 of 15 critical cases company-wide

Select a row to investigate
DepartmentnAI maturityCommitmentDepthMomentum lossCritical exposure

Compare with another department

Two self-reported indexes plus intent and depth

Selected departmentComparison
Small-sample caveat. Eight departments have fewer than eight respondents. Their scores are useful directional signals, not robust departmental rankings.
Scales: Maturity, commitment and depth are 0–100 composite measures from the 127-person Barometer. Momentum loss is the percentage intending to use less AI. Critical exposure is a count of individuals, not a severity rating for a department. Selecting departments does not expose individual employee records.
04 / KNOWLEDGE & SKILLS

Confidence isn't competence.

A separate 25-question technical assessment measures what employees actually know. It reveals where BTS can use AI comfortably, and where building advanced systems becomes difficult.

Parts 3 & 4 · pp. 26–34
Technical respondents
128
25-question assessment
Mean technical score
66.9/100
Median: 70/100
Score dispersion
23.9points
Standard deviation across respondents
Expert build & deploy
34.4/100
Lowest-scoring knowledge block

Where knowledge actually sits

Correct-score index by domain, 0 to 100

ENGINEERING BOTTLENECK
Source: Part 3, technical knowledge blocks, p. 27.

Technical competence distribution

128 assessed respondents

14

Beginner
10.9%

29

Basic
22.7%

37

Advanced
28.9%

48

Expert
37.5%

The technical assessment distribution is not the same as the later 149-person manually reviewed roster included in the report's annex.

The confidence-knowledge gap

Self-reported maturity versus tested knowledge

DIFFERENT MEASURES
Reported maturityTechnical test
Source: Parts 1 & 3, department scores pp. 14 and 28. Difference in points, not proof of overconfidence.

Misconceptions vs. unknowns

Different problems require different training

28%

Misunderstand LLM behaviour

Incorrect belief that models search the web or reason like humans.

27%

Reinforcement learning

Incorrect understanding of the concept.

73%

Production RAG

Admitted knowledge gap (“I don't know”).

68%

Human-in-the-loop

Admitted knowledge gap (“I don't know”).

Other gaps: tokens 22% misconception; long-context memory 20%; ReAct 59% unknown; MCP/multi-agent security 52% unknown.

Training tracks already mapped to people

Click to explore each track

Four-level learning pathway

Every level closes with practical verification, not just a knowledge quiz

PROPOSED CURRICULUM
LEVEL 01

Basic

LLM limits, hallucinations, safe data use, prompting and the 4 D's.

LEVEL 02

Intermediate

Projects, query/action connectors, Jira acceptance criteria, prompt injection.

LEVEL 03

Advanced

Skills, multi-connector flows and, for technical roles, Claude Code and SDD.

LEVEL 04

Expert

Advanced governance, MCP, agent security, RAG, evaluation and SDD at scale.

The 4 D's: Delegation, Description, Discernment and Diligence. Technical teams also develop spec-driven development: specify → plan → break down → implement → verify.
Source: Part 4, training design, pp. 29–34. Primary Claude-based tooling is a training-plan assumption to confirm before rollout.
05 / STRATEGIC ROADMAP

Turn the findings into action.

A recommended sequence over 30, 90 and 180 days. These are proposed actions in the October 2026 report, not a live implementation tracker.

Part 6 · pp. 40–41

Immediate priority actions

Recommended activities for the first month

PROPOSED
WHY THIS PHASE MATTERS

Make AI adoption safe without slowing it down.

Rapid guardrails and a targeted response to disengagement establish a foundation for the next phase of value creation.

PHASE OUTCOMEClear rules.A simple, fast approval model and a first cohort in training.
MEASUREMENT

What success should look like.

Targets in the report are future checkpoints, not current achievements.

Deeper AI usage

3-month improvement target

66→75 / 100

Baseline depth index to proposed target.

Technical knowledge

6-month assessment target

66.9→≥80

Average technical test score.

Expert engineering

6-month target in technical teams

34.4→>50

34.4 is the company-wide baseline for B4; >50 applies to technical teams.

Also targeted at the six-month review: no department below a 65/100 average technical score; reassess the 24% momentum-loss signal; maintain regular Champion activity. The report does not contain budgets, allocated owners for every initiative or verified results from completed actions.
06 / AI GOVERNANCE

Govern the system. Enable the people.

The proposed AI Committee coordinates secure and measurable use, prioritizes investment and keeps training active. It does not approve every everyday use of an authorized tool.

Part 5 · pp. 35–39

Four governance axes, eleven functions

Select an axis to explore its responsibilities

PROPOSAL

KNOW · responsibilities

Click a function to inspect its owner and cadence

Function details

Execution responsibilities proposed in the report

How a new AI use case is reviewed

Simplified governance logic from the report's approval flow

FAST LANE FOR ORDINARY USE
01 · ProposeDepartment specifies tool, purpose, data, autonomy and reviewer
→
02 · ClassifyIs it an already-approved tool with allowed data?
→
03 · Escalate if neededAdvanced agents, RAG, integrations or sensitive data trigger review
→
04 · RegisterCommittee portfolio & inventory; Security/Legal sign-off where relevant
Ordinary authorized usage does not go through the Committee. Security retains veto on security issues, Legal/DPO on personal-data issues.

The Committee decides

Governance, risk and priority decisions

  • ✓Approved, tolerated or retired AI tools
  • ✓Usage policy and data classification
  • ✓Prioritized and funded use cases
  • ✓Advanced automations, agents and integrations
  • ✓Exceptions, training design and schedule

What stays outside

Protecting speed and team autonomy

  • ×Approving every authorized AI interaction
  • ×Implementing the technical solutions itself
  • ×Building every department's use case
  • ×Replacing Legal/DPO's formal legal audits
  • ×Using test results to evaluate individual performance
Operating model: Eleven functions across Know, Govern, Drive and Train, with a quarterly report to Leadership. The committee charter is a proposal, not evidence that this body has already been constituted. Responsibility and cadence shown in this dashboard reflect the October 2026 proposal.