Mode 01 ~60% of cases
Starved data
The agent has the right prompts and no read access to the records it needs to act on. Most common.
Path A · Agentforce stalled
Your Agentforce pilot is stalling. The architecture is the reason.
Six-month pilots that do not deliver are usually starved data, not bad prompts. The assessment finds out which, in writing, in three weeks. Independent. Fixed-fee. NDA-default.
From €18k
Investment
3 weeks
Duration
1 architect
Team
< 24h
Reply SLA
Why it stalls
Every stalled engagement is one of these four.
The operating theory the assessment runs on. Four failure modes, in order of frequency. You will know which one by Friday of week 2.
Mode 02 ~20% of cases
Over-served data
The agent sees too much. Sharing rules leak PII into responses. Hidden until audit.
Mode 03 ~12% of cases
Identity collapse
Cross-cloud identity resolution never finished. Agents address ghosts.
Mode 04 ~8% of cases
Action vacuum
The agent can see and reason, but the Flow or Apex actions it needs were never built.
If you do not recognize any of these, the architecture probably is not the bottleneck. The discovery call says so honestly, before any contract gets signed.
What you get
Four written artifacts. Yours.
Documents the engagement leaves behind. No retainer required to keep them. No follow-on dependency by design.
- D1
Data access map
Every record the agent reads, every join it walks, every row it cannot see. Diagram plus table.
- D2
Security model audit
Sharing rules, Einstein Trust Layer config, scope creep, against AI-era patterns.
- D3
Production-readiness score
Scored checklist across access, identity, security, observability, rollback. Numeric, defensible against an external auditor.
- D4
Written report
A 25 to 40 page document your CTO can take to the board. Not a slide deck.
Schedule
Shippable artifact at each milestone.
The timeline below is fixed. If discovery surfaces something that warrants a longer engagement, the SoW gets amended in week 1, not by surprise.
- Wk 1 Discovery
Stakeholder interviews. Sandbox access. Agent telemetry review.
→ Kick-off note and scoped questions - Wk 2 Diagnosis
Written analysis. Risk register. Architecture remediation map.
→ Mid-point review with your CTO - Wk 3 Roadmap
Sequenced fixes with budgets, dependencies, build vs. configure calls.
→ Final report and read-out - Wk 4+ Execute
Optional. Direct or alongside your SI partner.
→ Statement of work, by request
Methodology
Why the work is shaped this way.
Independent
No license resale. No implementation team to feed. The recommendation is allowed to be "stop."
Architecture-first
Prompts get tuned by anyone. Data access models get fixed by someone who has shipped them. Different problem.
Written, not presented
A report a CTO can mark up and take to the board survives the next staff change. A deck does not.
What clients say
“His insights are always spot-on. He sees the architecture problem before anyone else can articulate it, and his recommendations are framed in language a non-Salesforce CTO can actually action.”
Juno Noh · Senior IT Manager, L'Occitane Korea
See full case studies → Best for / not for
Honest qualification.
A short list of who this engagement was built for, and a shorter list of who it was not.
Best for
- You bought Agentforce licenses on an AI mandate.
- A pilot is past month four and still not in production.
- Your SI is pushing more scope, and you suspect a redesign.
- You need a second opinion before the next steering committee.
Not for
- You have not bought Agentforce yet. That is Path B territory.
- You want someone to manage the SI for you. Different engagement.
- You want a prompt-engineering workshop. Not the work.
- The decision is already political and the answer is fixed.
vs. an SI assessment
How this differs from a Big-Four engagement.
Independent architect
Big-Four assessment
€18,000 · 3 weeks
€80,000+ · 8 to 12 weeks
One architect, full ownership
Pyramid: 1 partner, 6 associates
"Stop" is a valid recommendation
Recommends more SI work, always
Written report, 25 to 40 pages
Slide deck, 80 to 120 pages
Reports to your CTO
Reports to the engagement partner
Decision in 3 weeks
Decision in 3 months
Questions
Questions that come up in procurement.
If yours is not here, ask on the call. The answer will be specific.
Q01 Do you need admin access to our Salesforce org?
Read-only access is enough for the first two weeks. Some artifacts require sandbox metadata exports; those are scoped in the NDA. Production access is never required.
Q02 What if your recommendation is to kill the project?
Then that is the recommendation. It has happened twice. Both sponsors re-deployed the budget into Data 360 foundations and shipped within the original window.
Q03 Can you work alongside our existing SI?
Yes. The engagement is structured so the SI has visibility into findings but does not control the analysis. Most SIs welcome it; the ones that do not are usually the reason you called.
Q04 What does production-readiness actually mean?
A scored checklist across data access, identity resolution, security, observability, and rollback. The score has a number. The number is defensible against an external auditor.
Q05 Is the work covered under our existing MSA?
Usually no. The work runs under a fixed-fee statement of work with a default NDA. Most procurement teams turn it around in under a week.
Q06 Can my SI deliver this instead?
They can deliver something. They cannot deliver an independent finding, because the recommendation that follows is usually their next statement of work.
Q07 Where are you based?
Paris and Seoul. The work is delivered remotely with two on-site days in week 1 if your team is in EMEA.
Q08 What languages do you work in?
English and French, written and spoken. Korean conversational.
Before you go
Take the AI Readiness Assessment → How AI-ready is your Salesforce org?
9 questions, 2 minutes, instant results. No signup required.