Miguel Dias
v3 deck · Merantix Berlin

Going from Outputs to Outcomes.

Miguel Dias
Independent VSM consultant
12 May 2026

AI multiplies output, but it's not multiplying outcomes. How do we close the gap?

Miguel Dias presenting outputs to outcomes at Merantix Berlin
Outputs to Outcomes · Merantix Berlin
01
The five problems nobody stopped talking about.

Across the panel, the talks, and the coffee-line conversations at Prompt 2026. Ranked loosely by how often they surfaced.

Recurring

Picking what to build.

Getting the why and the what before the how. Working closer with customers to decide what actually matters.

Recurring

Burnout and overload.

Context switching is up. More agents, more concurrent work. One CEO prescribed longer hours. Others pushed back.

Recurring

Harness engineering (the weeds of agents, PRs, CD).

Everyone is building a version of this: orchestration, guardrails, feature flags, evals, the loop that keeps agents shipping safely. Different names, same problem.

Open problem

Quality is degrading.

More code, more bugs. Speakers named agents addressing quality in production as an unsolved problem.

02
First of all, What will your CTO ask end of quarter?

Three-month window. The questions you’ll get — and what you now have answers for.

“Show me the AI ROI.”
More code ≠ more revenue. What’s the conversion mechanism we’re building?
“Are we delivering on our commitments?”
Can we ship what we promised to our biggest customers, on time? If not — where is it stuck, and why?
“_______________________?”
What’s your CTO actually asking? The right row beats both.

Best guess at your CTO’s quarterly. The point of the last row is the one we haven’t guessed.

03
Output vs Outcome Economics

Local optimisation lives on Outputs. CTO targets are Outcomes.

Output economics
  • code shipped
  • PRs merged
  • deploys / week
  • dev velocity
Outcome economics
  • revenue-bearing features
  • bad bets killed early
  • end-to-end TTM
  • quality (fewer incidents, faster recovery)

▸ This deck moves the conversation to the right.

04
Example scenario, B2C

Feature hit rate — what % of your bets paid off?

Now — what if 5–10 became 15?
5–10 extra winners × €500k–1.5M each
≈ €2.5M – €15M / yr upside

Same team. Same budget. That’s the prize this deck is chasing.

05
Three ways to move feature hit rate.

What % of your bets paid off? What can we do to increase it?

a.

Hire more capacity.

  • coordination tax goes up
  • Brooks ramp — new hires get worse before better
  • onboarding dip on the team carrying them
b.

Same capacity — raise hit-rate and time to market.

  • moves the outcomes
  • more money coming in, same capacity
c.

Something else?

  • massive restructuring, team redesign…
— left open. Tell us what’s on the table.

Option (c) is the one we don’t want to assume. Reorgs and restructuring often show up at growth-stage companies but rarely make the slide. Worth naming what’s actually in scope before this conversation closes.

06
What it takes to increase feature hit-rate:

Same capacity, raise hit-rate and throughput = more revenue-bearing features. Two things need to move at the same time.

Lever 1: Faster time to market
  • faster product discovery to experiment results
  • engineering enables product leaders to experiment faster
  • Look at: lead time from idea to production
Lever 2: Quality
  • more code = more quality problems
  • no way around it except improve the system
  • Look at: incidents, time to recover, rework eating capacity

Both are measurable. Both respond to the same intervention: seeing the value stream clearly and managing it together.

07
Example — how a 2-week cadence injects waste.

Lever 1 in action. One pattern — yours may be different. Mapping the value stream is how you find out.

SPRINT 1 SPRINT 2 SPRINT 3 SPRINT 4 SPRINT 5 SPRINT 6 t idea spec build release measure decide ← killed "waiting for next sprint" "CAB on Thursday" killed at month 3 active work waiting (sprint boundary) killed late

One pattern. Yours may be different — mapping the value stream is how you find out which structural waste is actually costing you. Local dev speed is not the bottleneck; the cadence between phases is a cross-stream constraint.

08
Same stream — released from the cadence.

Same items, no sprint trap. Below the Gantt: where the measurement hooks would attach to a generic delivery stack.

CONTINUOUS FLOW t idea spec build release measure decide ← kill at week 1 YOUR DELIVERY STACK Idea tool Spec tool Build tool Deploy tool Measure tool Decide tool hook hook hook hook hook ← measure

Start by talking to teams about what data matters and how to collect it (this will change over time and is never finished). You’ll likely need to build custom aggregations on top of the tools you already have. Improve them continuously. Start small. Big-bang adoption fails.

09
Example — where quality degrades.

Lever 2 in action. How more AI-generated code, without system-level feedback, leads to more incidents and harder recovery.

vicious circle More code from AI more surface area for defects Harder to catch engineers didn’t write it More incidents longer recovery, rework eats capacity Batch sizes grow rework delays everything Feedback slows loops break down, learning stops Quality degrades customers feel it, churn rises

This is a system problem. You can’t fix it by reviewing harder. You fix it by making the system visible: where defects enter, how long they take to find, what the rework costs. That’s a value stream conversation.

10
A B2B example: where both levers show up.

A real company. Complex release involving six teams in sequence. Touch time per stage is a few hours. Cycle time per stage is weeks. That gap is the waste.

S1 S2 S3 S4 S5 S6 S7 App Team A App Team B Operations QA & Perf Test Implementation Consultants testing at the END firefighting: 25–50% of every sprint actual work (hours) waiting / queue / rework (weeks) killed late
Why
·
No production-like test environments or test data.
·
No continuous testing — testing happens at the END, not throughout.
·
Big-batch release train: high error surface, impossible to isolate which change caused the defect.
·
Last-minute features pushed by business → product → devs → shipped untested.
·
Technical debt dragging every team down.
100 days
Total lead time
< 1 day
Total work time
< 1%
Activity ratio

Business complains: “development and testing take too long.” “Deployments always go wrong.” But we are all causing the problems we complain about.

11
Manage the value stream — as a team.

Why we invite all these teams to talk to each other.

Each team optimizing locally was hurting the whole company — despite seemingly helping themselves.

Business saw
How last-minute feature pressure caused the quality problems they complained about.
Product saw
How skipping testing created the rework that made “development takes too long” true.
Engineering saw
How the lack of tech investment (test envs, test data, continuous testing) was the root cause of customer churn.
Everyone saw
One day of work. One hundred days of waiting. The constraint isn’t speed — it’s the system.
Lever 1: Faster TTM
  • faster feedback
  • kill bad bets earlier
  • team time goes to features, not firefighting
Lever 2: Better quality
  • fewer production incidents
  • less rework eating capacity
  • happier, returning customers
  • necessary tech investments become visible + justified

Siloed visibility: each team thinks they’re doing fine.
End-to-end visibility: teams pursuing local goals are hurting each other and the company.

Value stream metrics + talking to your teams show you where the problems are — and help teams find the solutions. I’m not a product coach. You need to get great at product discovery. But the value stream shows you where to look.

12
From conversation to receipts.

The path from a VSM conversation with your teams to the numbers your CTO is asking about.

VSM conversation with the teams doing the work handoffs, waits, rework — named together
Flow metrics cycle time, batch, hit-rate leading indicators that change weekly, not yearly
Revenue + hit-rate % features generating revenue lagging, 12-month window — the number your CTO already cares about
The Steer

Flow metrics aren’t a dashboard you admire. They’re how you kill bad bets earlier, ship the good ones faster, and walk into the quarterly review with the receipts.

Two phases to instrument first. Let’s pick them before our next session.

13
What working together looks like.

Two modes. A 90-day shape. How outcomes get measured. No price on this slide — that’s a conversation.

Mode 1
Workshopknowledge work · days
  • remap one stream with your team
  • agree the hooks together
  • define what gets measured
Mode 2
Embeddedoutsourced work · weeks
  • ~3 days/week in your standups
  • instrument the hooks across your stack
  • run the loop with one team · handoff at the end
90-day shape (illustrative)
Week 1
Workshop
remap one stream · agree hooks · define what gets measured
Weeks 2–12
Embedded
instrument hooks across your delivery stack · run the loop with one team
Week 13
Handoff
metrics live in your tools, owned by your team

Outcomes measured by:

No price on the slide. The conversation about scope, sequence, and commercials lives in the call.

14
The loop that turns output into outcome.

Closing figure — the mechanism slide 02 promised, drawn. Step 4 is the bridge.

Value stream map · idea → production → learning → idea
close the loop — and start again 1 Decompose task → pieces 2 Assign one agent per piece 3 Commit small flags + TDD, to trunk 4 Capture outcomes usage, reliability, $ the bridge 5 Feed back learning → next loop TABLE STAKES Platform engineering self-serve envs, evals, CI DevOps flow, feedback, continuous learning Continuous delivery trunk, small batches, fast feedback
Output is steps 1–3. Outcome is step 4. The loop closes at step 5.