DCX — Digital Customer Experience Internal · April 2026
AI Acceleration Proposal · Built on the DCX Time & Motion study · 57 activities measured

An evidence-based
proposal to reclaim
2,130 man-hours
per month.

Arnesh Mandal · Design Engineer · DCX
01
Section 01

The Data.

Every activity, every minute. Before drawing any conclusion, the actual measured workload from the DCX Time & Motion study.

DCX · AI Acceleration Proposal 02 / 16
Source: DCX Time & Motion Study, Apr 2026

57 activities measured. Function-wide.

57 activities · 14 columns · weighted averages Scroll vertically to read every row · per-person values
Intensity (per activity) Time / Month per person, by team (Hrs) Hours saved / month (Hrs)
# Activity Frequency Per Day (Mns) Per Month (Hrs) Research4 ppl D&AI6 ppl Insights2 ppl Design Ops28 ppl Suggested Tool Research D&AI Insights Design Ops
1Meetings with stakeholders (internal BFL)Daily / Weekly93.4832.7252.1221.5810.5033.92—
2Meetings with stakeholders (external Partner teams)Daily38.5713.5017.7011.6722.5012.65—
3Alignment with team membersDaily602121Custom markdown + SKILL files2.10
4DCX MeetingsWeekly23.338.178.17—
5DCX/Design Reviews (Internal & External)Daily68.3323.9223.92—
6Reviews (Team / Manager / Skip / SMT)Daily69.6424.3824.38—
7Non DCX reviewsWeekly10.333.623.62—
8Grooming sessions with POsWeekly26.679.339.33—
9Grooming sessions with COE, PO & IT teamsMonthly23.608.268.26—
10AI integration meetsWeekly4.831.691.69—
11Creation of research / drafting of protocolsDaily / Weekly91.8832.1627.9940.50Claude Code / Cowork + custom SKILL20.9930.38
12Creation of RP testing reportDaily47.1416.5016.50Claude Code / Cowork + custom SKILL12.38
13Creation of VOC questionnaire / unmod studyDaily25.7099Claude Code / Cowork + custom SKILL6.75
14Configuration of VOC survey on MazeWeekly / Monthly16.965.945.94Claude Code + Playwright CLI2.97
15Maze study monitoringDaily70.4224.6524.65Agent testing (AI as user)6.16
16Moderation sessionsMonthly49.6217.3717.37Agent testing (AI as focus group)4.34
17Branding Study Moderator managementWeekly602121Claude / GPT / Copilot6.30
18Data procurement and alignmentMonthly3010.5010.50—
19RCA Data Gathering and analysisWeekly000—
20Session recording RCA Data gathering & analysisWeekly / Monthly000—
21MOD User evaluation RCAWeekly / Monthly602121Claude Code / Cowork + custom SKILL15.75
22Un-MOD User evaluation RCAWeekly / Monthly000—
23Insighting and analysisWeekly / Monthly68.572424Claude Code / Cowork + custom SKILL18
24Secondary researchQuarterly29.2310.2310.23Claude Code / Cowork + SKILL or GPT Deep Research7.67
25Research on Design related SubjectsDaily3311.5511.55Claude Code / Cowork + custom SKILL5.78
26BenchmarkingWeekly / Monthly19.966.996.99Claude Code + Figma MCP + Playwright CLI + SKILL3.50
27Prototyping for RPWeekly15.355.375.37Claude Code + Figma MCP1.61
28Design explorationDaily34.1711.9611.96Claude Design8.97
29Designing / Design creation & deliveryDaily100.7835.2715.4639.52Claude Code + Figma MCP4.6411.86
30Ideation & BrainstormingWeekly43.1215.0915.09Claude Code / Cowork7.54
31Customer experience, ecosystem & journey mappingDaily19.786.926.92Claude Code + Figma MCP3.46
32UX content generation & verificationDaily19.451.241.24Claude Code / Cowork + custom SKILL1.12
33Library: Components, Icons, AnimationsWeekly15.335.375.37Claude Code + Figma MCP4.03
34Library updation & Logic doc closuresWeekly / Monthly9.693.393.39Claude Code + Figma MCP2.54
35AB test use case creationDaily / Monthly24.161.542.310Claude Code / Cowork + custom SKILL1.730
36AB testing figma creationDaily / Weekly29.264.731.8410.50Claude Code + Figma MCP0.744.20
37A/B, Beta, RP test case creation & solutioningMonthly / Weekly11.654.084.673.95Claude Code / Cowork + custom SKILL2.331.98
38UAT, CUG & Production testingQuarterly8.492.972.97Claude Code + Playwright CLI2.38
39Production testingDaily7.832.742.74Claude Code + Playwright CLI2.19
40Delivery planning (plan creation & workload eval)Daily23.068.078.07Claude Cowork2.82
41Work allocation within teamDaily9.673.393.39Claude Cowork1.02
42Creating presentationsWeekly / Monthly17.776.2214.168.174.67Claude Design8.504.902.80
43Creating frameworks & guidelines for tasksDaily14.505.085.08Claude Code / Cowork + custom SKILL3.81
44DPAOP / LRSMonthly3512.2512.25—
45Design Team InitiativesWeekly12.504.384.38Claude Cowork or Claude Code with SKILL1.31
46Giving downloads (modules, library socialization)Monthly258.758.75—
47Hiring processDaily19.081.211.21Claude Code / Cowork + custom SKILL0.36
48InterviewsAnnual6.172.162.16—
49Off-role managementWeekly11.674.084.08—
50Incentives & PMSQuarterly9.503.333.33Claude Code / Cowork + custom SKILL1.00
51Email managementDaily13.674.784.78Outlook Copilot0.48
52Medalia score & issues tracking—0.830.290.29Claude Code + Playwright CLI0.14
53Maintaining and updating trackersMonthly6.172.162.16Claude Cowork1.51
54Non-meeting stakeholder follow-upsWeekly20.837.297.29—
55Practicing new design toolsWeekly26.679.339.33—
56Learning / Upskilling activitiesWeekly / Daily27.249.536.1210.5010.20—
57Ad-hoc tasksDaily / Weekly54.2718.9911.6211.9626.2521.04Claude Code / Cowork + SKILL or Figma MCP3.493.597.886.31
All values are average per person per month, weighted across teams 03 / 16
02
Section 02

Where the
gains are.

Not all activities are equal. The data reveals which kinds of work AI can absorb — and which it cannot touch.

DCX · AI Acceleration Proposal 04 / 16
Activities grouped by what they actually are

The savings cluster in two places.

Two work types dominate the savings: design work (because AI accelerates design directly and the team is large) and drafting & analysis (because automation potential is highest). Together they account for nearly 80% of all hours saved.

Grouping activities by what they actually are — not which team does them — makes the pattern clear: the levers a design team should care about most are also where the gains pool up.

Points to: Automate the writing. Amplify the designing. The percentage and the man-hours both agree.
Activity type % of total Man-hrs / mo
01 Design workDesign generation, library, journey maps, presentations, exploration 47% 994man-hrs
02 Drafting & analysisProtocols, reports, frameworks, ideation, test cases, RCA, research 32% 689man-hrs
03 Testing & automationUAT, regression, production testing, Maze configuration, benchmarking 11% 239man-hrs
04 Ops & planningDelivery planning, work allocation, tracker maintenance 5% 117man-hrs
05 Light-touchEmail, alignment, monitoring, light secondary research 4% 92man-hrs
Function-wide · grouped by activity type · sums to 2,130 05 / 16
02 — Where the gains are · The pivot

Less drafting.
More designing.

This is a design team. AI hits us in two places at once — it removes the desk work AND accelerates the design work itself. The hours don't disappear; they return to the work this team exists for.

Drafting work removed ~690man-hrs Per month, function-wide — protocols, reports, analysis, frameworks taken off the plate.
Design work accelerated ~990man-hrs Per month, function-wide — design generation, library, journey mapping, exploration sped up.
Combined leverage ~1,680man-hrs Per month, function-wide — returned to the design team for the work that defines design.
A design team should spend its time designing 06 / 16
The simple rule that explains everything · Insight 02 of 04

AI replaces typing, not talking.

Anything that produces an artifact can be largely automated. Anything that needs real-time dialogue, judgement, or relationship-building stays human. This single rule explains nearly every percentage in the proposal.

CategoryAutomation potentialExamples
Written / document work75–90%Protocols, reports, test cases, UX content, frameworks & guidelines
Repetitive testing70–80%UAT, regression, production testing, Maze configuration
Visual / design generation30–75%Figma components, libraries, prototypes, design exploration
Analysis & synthesis50–75%Insighting, RCA, secondary research, benchmarking
Tracker & ops work30–70%Delivery planning, work allocation, tracker maintenance
Live meetings & reviews0–15%Stakeholder meetings, design reviews, grooming sessions
Human relationships0%Interviews, training others, stakeholder follow-ups, learning
Pattern derived from the 57-activity dataset 07 / 16
02 — Where the gains are · Insight 03 of 04

3,400 man-hours
a month go to meetings,
reviews, and grooming.

The single biggest unaddressed bucket. AI can shave 0–15% from this category — useful, but not the headline. The real unlock here is human: meeting hygiene, async-by-default, fewer attendees. A separate conversation worth having.

~3,400Man-hrs / month in meetings
~40%of total function-wide time
0–15%AI automation ceiling
A non-AI problem with a non-AI solution 08 / 16
03
Section 03

How we
get there.

First the tools that win. Then the procurement decision they imply. Then the reasoning behind both.

DCX · AI Acceleration Proposal 09 / 16
Savings, summed by tool stack · Results first

The data picks a winner.

When activity-level savings are summed by the tool stack that does the work — and totaled across the whole function — the order is striking. Two tool stacks dominate; the rest play distinct, complementary roles.

Rank Tool stack Coverage Man-hrs / month
01Claude Code + Figma MCPDesign generation & delivery, library, journey mapping, prototyping798
02Claude Code or Cowork + skillDrafting, analysis, frameworks, ideation, test cases, RCA, secondary research689
03Claude DesignDesign exploration, presentations & pitch material196
04Claude Code + Playwright CLIUAT, regression, production testing, Maze configuration141
05Claude Cowork (standalone)Delivery planning, work allocation, tracker maintenance117
06Claude Code + Figma MCP + PlaywrightBenchmarking — combined visual capture & analysis98
07Agent testing & light-touch toolsMaze monitoring, moderation simulation, email, alignment92
The key observation Claude Code & Cowork together (across all stacks) drive ~85% of total savings. Code, Cowork & Design form a complementary triad — different tools for different families of work.
Total man-hours saved per month, summed by tool stack 10 / 16
One ecosystem, four tools · The recommendation

The procurement story is simple.

~95%of identified savings covered
~2,130Man-hrs / month
~25,500Man-hrs / year
~12 FTECapacity unlocked
Click any tool to see its activity detail · Code & Cowork as paired primaries 11 / 16
A simple frame · Why the ranking looks like it does

Processing. Memory. Hands.

An AI tool is only as useful as how complete it is. Three parts together make an agent; missing any one shrinks what it can do. The ranking on the previous slide tracks this completeness exactly.

Part 01

Processing

The language model itself — the part that thinks, reasons, and writes. Every modern AI tool has this. The cheapest commodity in the stack.

Looks like: Any LLM — the engine answering your question.

Part 02

Memory

Structured context the agent reads from — markdown files, skills, project conventions. Not "what was said earlier" but a reusable, version-controlled brain.

Looks like: Markdown files · skills · CLAUDE.md · custom protocols.

Part 03

Hands

The ability to actually create and modify things outside the chat UI — files on disk, design components, browser actions, database rows.

Looks like: Bash · file system · MCP · Playwright · Figma plugin.

The fewer parts a tool has, the less work it can finish for you 12 / 16
Mapping the tools to the three parts · Capability matrix

Not every AI is an agent.

Tool Processing Memorystructured Handscreates artifacts What it can actually do
Claude CodeCLI · full agent YES YES YES Drafts, generates, tests, builds. Reads markdown + skills as durable memory; uses bash, MCP, file system as hands.
Claude CoworkDesktop · structured drafting YES PARTIAL YES Creates real files. Has skills but lighter on markdown-based memory — ideal for structured drafting, planning, and ops without CLI overhead.
Claude DesignSpecialist · visual YES PARTIAL YESvisual only Generates visual artifacts — designs, presentations, mockups. Memory is design-context only; hands are limited to visual output.
Generic AI chatGPT, Copilot, etc. YES NO NO Suggests and advises. Doesn't read structured memory; doesn't create artifacts outside its own UI. Useful for quick prompts only.
The pattern in the savings data Tools with all three parts produce 50–75% reductions. Tools missing one produce 25–35%. Tools with only processing produce under 15%.
YES = full · PARTIAL = limited · NO = absent 13 / 16
DCX · AI Acceleration Proposal · End of main deck

~2,130man-hrs / month

Roughly twelve full-time people of capacity, returned to a forty-person function — without hiring, without restructure, without changing what the team is asked to deliver. The proposal is on the table.

Note: Estimates based on prior team structure and typical AI productivity gains; recalibrate from real time logs after one quarter of usage. Adoption requires SKILL authoring & training (not modeled). Meeting time (~3,400 mh/mo) sits at the AI ceiling — that's a culture conversation, not a tools one.

DCX · April 2026 · Two more slides on procurement & alternatives 14 / 16
The procurement reality · The simplification angle

All of this.
One Claude subscription.

Generic AI chat + Claude Design + Claude Cowork + Claude Code

Every tool in this proposal lives inside the same Claude product family. Generic chat for quick answers, Design for visual exploration, Cowork for drafting and ops, Code for agent-grade work. Acquired together, deployed together.

~2,130 Man-hours / month — unlocked through one Claude subscription
One ecosystem · one procurement decision · one onboarding curve 15 / 16
If not Claude — what then · The alternative path

Many tools or one.

Suggested vs alternative tools · 37 activities Microsoft tools where they fit · gaps where no real equivalent exists
Activity Man-hrs / mo Suggested (one ecosystem) Alternative (multiple vendors)
Designing & design creation / delivery360Claude Code + Figma MCP— no equivalent —
Ad-hoc tasks (Figma-touching)228Claude Code + Figma MCP— no equivalent —
Ideation & brainstorming211Claude CoworkMicrosoft Copilot
Drafting research protocols145Claude Cowork (skill)Microsoft 365 Copilot
Creating presentations142Claude Design— no equivalent —
Benchmarking98Claude Code + Figma MCP + PlaywrightCursor + Playwright
Customer experience & journey mapping97Claude Code + Figma MCP— no equivalent —
Delivery planning79Claude CoworkMicrosoft Copilot in Loop
Library updation & logic doc closures71Claude Code + Figma MCP— no equivalent —
A/B, Beta, RP test case creation69Claude Cowork (skill)Microsoft 365 Copilot
UAT, CUG & production testing67Claude Code + Playwright CLI— no equivalent —
Production testing61Claude Code + Playwright CLI— no equivalent —
Design exploration54Claude DesignGoogle Stitch
RP testing reports50Claude Cowork (skill)Microsoft 365 Copilot
Insighting & analysis36Claude Cowork (skill)Microsoft Copilot
Research on design subjects35Claude Cowork (skill)— no equivalent —
MOD User evaluation RCA32Claude Cowork (skill)Microsoft Copilot
UX content generation & verification31Claude Cowork (skill)Microsoft Copilot
Secondary research31Claude Cowork (skill)GPT Deep Research
Work allocation within team28Claude Cowork— no equivalent —
VOC questionnaire / un-mod study27Claude Cowork (skill)— no equivalent —
Maze monitoring25Agent testing (AI as user)— no equivalent —
Library: components, icons, animations24Claude Code + Figma MCP— no equivalent —
Frameworks & guidelines23Claude Cowork (skill)— no equivalent —
Moderation sessions17Agent testing— no equivalent —
Branding study moderator mgmt13Claude / GPT / CopilotMicrosoft Copilot
Maze configuration12Claude Code + Playwright CLI— no equivalent —
AB testing figma creation11Claude Code + Figma MCP— no equivalent —
Hiring process10Claude Cowork (skill)— no equivalent —
Maintaining & updating trackers9Claude CoworkMicrosoft Copilot in Excel
Design Team Initiatives8Claude Cowork— no equivalent —
AB test use case creation7Claude Cowork (skill)Microsoft Copilot
Prototyping for RP6Claude Code + Figma MCP— no equivalent —
Incentives & PMS6Claude Cowork (skill)Microsoft 365 Copilot
Alignment with team members4Custom markdown + skill— no equivalent —
Email management3Outlook CopilotOutlook Copilot (same)
Medalia score & issues tracking1Claude Code + Playwright CLI— no equivalent —
Italicised cells = no real alternative · the gaps speak for themselves 16 / 16
Detail · Claude Code
01
For specialized, agent-grade work · Tool 01 of 04 Claude Code
~1,040 Man-hrs / month — MCP & CLI work

Full agent. Processing + memory (markdown + skills) + hands (terminal, MCP, Figma plugin, Playwright). Use it when work genuinely needs agent capabilities — Figma component generation, browser-driven testing, multi-step automation, structured memory across sessions. The right pick for anyone doing pro-level work that touches MCPs, CLI, or skills — design, research, ops alike.

Activity Stack Man-hrs / month
Designing & design creation / deliveryClaude Code + Figma MCP360
Ad-hoc tasks (Figma-touching)Claude Code + Figma MCP228
BenchmarkingClaude Code + Figma MCP + Playwright98
Customer experience & journey mappingClaude Code + Figma MCP97
Library updation & logic doc closuresClaude Code + Figma MCP71
UAT, CUG & production testingClaude Code + Playwright CLI67
Production testingClaude Code + Playwright CLI61
Library: components, icons, animationsClaude Code + Figma MCP24
Maze configurationClaude Code + Playwright CLI12
AB testing figma creationClaude Code + Figma MCP11
Prototyping for RPClaude Code + Figma MCP6
Medalia score & issues trackingClaude Code + Playwright CLI1
Right pick whenever MCP, CLI, or browser control is involved •• / 16
Detail · Claude Cowork
02
For drafting, structured documents & ops · Tool 02 of 04 Claude Cowork
~840 Man-hrs / month — drafting & ops

Almost as capable as Claude Code, simpler to pick up. Processing + hands + skills, just lighter on the markdown-memory side. The right pick when you need real document creation but don't need MCP servers, CLI environments, or browser control. This is where most of the writing-heavy work lives — and the man-hours bear that out.

Activity Stack Man-hrs / month
Ideation & brainstormingClaude Cowork211
Drafting research protocolsClaude Cowork (with skill)145
Delivery planningClaude Cowork79
A/B, Beta, RP test case creationClaude Cowork (with skill)69
RP testing reportsClaude Cowork (with skill)50
Insighting & analysisClaude Cowork (with skill)36
Research on design subjectsClaude Cowork (with skill)35
MOD User evaluation RCAClaude Cowork (with skill)32
UX content generation & verificationClaude Cowork (with skill)31
Secondary researchClaude Cowork (with skill)31
Work allocation within teamClaude Cowork28
VOC questionnaire / un-mod studyClaude Cowork (with skill)27
Frameworks & guidelinesClaude Cowork (with skill)23
Hiring processClaude Cowork (with skill)10
Maintaining & updating trackersClaude Cowork9
Design Team InitiativesClaude Cowork8
AB test use case creationClaude Cowork (with skill)7
Incentives & PMSClaude Cowork (with skill)6
How to choose between Code and Cowork If the task needs MCP, browser control, or terminal — Claude Code. Otherwise both work, and Cowork is the lower-friction path. Most drafting workflows in this list belong here naturally.
A first-class option · the largest single tool bucket by drafting man-hours •• / 16
Detail · Claude Design
03
Visual specialist · Tool 03 of 04 Claude Design
~200 Man-hrs / month — focused visual work

The visual specialist. Processing + design-context memory + visual hands. Best for design exploration and pitch-grade visuals where the goal is a strong visual artifact rather than structured component work. Pairs with — does not compete with — Claude Code + Figma MCP.

Activity Stack Man-hrs / month
Creating presentationsClaude Design142
Design explorationClaude Design54
Why a separate specialist Visual generation is a different kind of "hands" than file-system hands. Claude Code + Figma MCP covers structured component work; Claude Design covers exploratory and pitch-grade visuals.
Specialist tool · use alongside Code & Cowork, not instead of them •• / 16
Detail · Generic AI / Copilot
04
Light-touch · likely already deployed · Tool 04 of 04 Generic AI & Outlook Copilot
~50 Man-hrs / month — residual tasks

Processing only. Useful for high-frequency, low-complexity tasks that don't need an artifact or structured context. Likely already available in the org — no procurement effort. Use to mop up the long tail.

Activity Stack Man-hrs / month
Secondary research (light)GPT Deep Research / Claude31
Branding study moderator mgmtClaude / GPT / Copilot13
Alignment with team membersMarkdown + skill files4
Email managementOutlook Copilot3
Why kept light Without hands or memory, these tools advise but don't deliver. Don't overinvest here — the savings ceiling is structurally low.
Likely already in the org · zero procurement cost •• / 16