You’re introduced onto a undertaking, and it’s a multitude. The org is a tangle of undocumented automations, convoluted sharing guidelines, and customized objects that appear to duplicate a typical object for no obvious purpose. You have a look at a posh stream or a legacy Apex set off and end up asking: “Why? Why did they construct it this fashion? What downside was this even attempting to resolve?” The unique architects and builders have lengthy since moved on, and with them, the institutional data of those vital choices has gone.
Sound acquainted? As an architect, that is how I’ve felt on multiple event.
In an ideal world, each essential architectural selection can be mentioned in a proper doc, one which any workforce member can have a look at to raised perceive key choices that led to an answer. Nevertheless, rigorously documenting each key resolution manually is time-consuming.
This text explores how AI can be utilized to enhance an architect’s decision-making position on tasks. In my expertise, solely people can present the nuanced, project-specific context and stakeholder empathy required to ascertain the evaluation standards for a call — however AI is admittedly good at analysis and formatting data shortly. So, I’ve been exploring methods to meld one of the best of each worlds to assist kickstart choices with out ranging from a clean piece of paper.
Throughout my profession, I’ve used many resolution documentation templates, however have discovered one to be essentially the most helpful and best to undertake throughout undertaking measurement and business: the architectural resolution report (ADR).
What’s an architectural resolution report (ADR)
The core objective of an ADR is to doc the “why” behind a selection. This contains the enterprise context, alternate options thought of, the trade-offs evaluated, and the ultimate resolution.
The concept is that every one stakeholders, from enterprise architects to builders, are aligned on the trail ahead. This doc is not only for future reference; it features alignment and buy-in from stakeholders throughout a undertaking because it supplies a clear report of the enterprise rationale and any concessions that have been accepted.
It’s your get-out-of-jail-free card when somebody inevitably asks six months later, “Why did we resolve to do it that manner?”
The worth of an ADR lies in its structured format, which supplies a common language for technical choices. It’s sometimes comprised of a number of important sections:
- Context: This part supplies the background data that led to the choice. It’s the narrative of the issue, together with the organizational state of affairs, enterprise priorities, and any constraints or necessities that form the selection.
- Thought of Choices: Right here, the doc lists the choice options that have been evaluated. Every possibility is introduced with an in depth breakdown of its execs and cons.
- Choice: This can be a clear, definitive assertion of the chosen answer. It’s accompanied by the core rationale and justification, explicitly referencing why a selected possibility was chosen over the alternate options.
- Penalties: This part outlines the results and outcomes of the choice. It describes what turns into simpler, what turns into harder, and what mandatory changes or follow-up actions are required for the workforce and the structure.
Whereas the worth of ADRs is apparent, a high quality, effectively thought-out ADR is time-consuming to provide and detracts from the inventive, strategic work that architects do finest.
That is the place giant language fashions (LLMs) and the sphere of immediate engineering supply a serving to hand.
AI for architectural choices
The efficient use of AI in architectural design isn’t a one-prompt, easy hack, however a disciplined follow that requires a basis of strategic rules.
A information to introducing generative AI into your decision-making
When utilizing AI for decision-making, an architect has to contemplate the destructive penalties that may come up from counting on AI with out correct oversight. LLMs can amplify societal biases by skilled knowledge, and the probabilistic nature of LLMs makes their outcomes have a sure lack of explainability. Architects should be vigilant when reviewing outputs to make sure that they replicate a good, effectively thought of, and correct consequence.
Human oversight
An AI-augmented world for choices on tasks wants a “human within the loop” (HITL), which is the flexibility for people to intervene in automated programs to forestall disasters. Within the context of generative AI when making architectural choices, HITL is about giving the architect a possibility to course appropriate AI-generated content material.
I’ve additionally discovered that choices I’ve made on tasks have been extra about stakeholder administration than technical due diligence. It’s the contextual judgement — years of expertise and exhibiting empathy to involved stakeholders — that drives a call that’s technically sound and extensively supported. Finally the Salesforce Architect needs to be accountable for each key architectural resolution made, no matter AI’s involvement.
Governance framework
One implication of utilizing AI to help with architectural choices is the danger of exposing delicate, proprietary data. When an architect inputs an organization’s confidential enterprise necessities, knowledge fashions, or answer designs right into a public, general-purpose LLM, that knowledge can be utilized to coach the mannequin, probably resulting in knowledge leakage and mental property publicity. In case you are utilizing public LLMs like Claude, Gemini, and ChatGPT, just remember to have reviewed the phrases of service, be sure that they don’t use your inputs for coaching, and activate choices to forestall your conversations from getting used for coaching.
I additionally suggest together with a proper disclosure in any outputs to construct belief with undertaking groups and guarantee compliance with rising moral requirements and potential laws. Issues to contemplate can be:
- The AI’s position In our case, to draft an ADR and summarize analysis of evaluation choices and rationale
- The instruments used: Present context and traceability of the mannequin used
- An expectation on accuracy: Embody clear statements that AI could make errors
- An accountability assertion: State that the human architect is accountable for the ultimate resolution
To make sure that the decision-making course of is not only about documenting generic execs and cons, however about evaluating selections towards a structured set of standards, I like to recommend utilizing theSalesforce Properly-Architected Framework. This helps cement any choices you make as an architect with a set of effectively thought-out rules that prioritize safety and compliance (Trusted), maintainability and person expertise (Simple), and scalability and resilience (Adaptable). By anchoring the AI’s output on this framework, an architect can be sure that the ultimate resolution is grounded in a confirmed methodology for constructing wholesome, long-term options on the Salesforce Platform.
Immediate engineering
Immediate engineering is the method of crafting prompts to information generative AI fashions to provide particular outputs. Consider your self as a conductor, directing the AI’s output with particular prompts to offer undertaking context and expertise. The grunt work of researching, drafting, summarizing, and formatting is delegated. It’s a course of that may be repeated, at velocity, to additional refine your strategic considering. Immediate engineering is subsequently a pure companion when making choices.
For a posh, multi-stage course of, like a key design resolution on a Salesforce undertaking, a single, all-encompassing immediate is mostly a much less efficient technique. It’s a bit like asking a brand new workforce member to ship a remaining answer after a single assembly; you’re simply asking for well-meaning hallucinations. AI isn’t excellent. It may be unpredictable, inaccurate, and at instances could make issues up. Within the context of your most essential design choices, that’s not a very good combine.
As an alternative, as a way to hold a human within the loop, I advocate for a extra architect-centric strategy: immediate chaining. The AI is guided to provide a extra project-specific, related, and correct outcome by breaking the duty into logical steps, with an architect diligently on the helm.
An architect-in-the-loop, guiding generative AI when making key choices
To take a look at this course of in additional element, let’s check out a sensible instance.
Sensible walkthrough: Single-org vs. multi-org choices
The next is an in depth, step-by-step walkthrough of a immediate technique that I discover useful to generate an ADR for a standard Salesforce design selection: whether or not to implement a single-org or multi-org structure. We’ll stroll by 5 pattern prompts that show the ideas wanted to provide a trusted outcome from generative AI utilizing immediate chaining with an architect-in-the-loop.
A five-prompt information for a key design resolution
Observe: The pattern outputs within the examples under have been generated with Gemini 2.5. They’re designed to point out you the method and indicative pattern outputs, to not be handled for example of a remaining, validated consequence.
Immediate 1: Context & evaluation standards
That is essentially the most essential step: I inform the AI concerning the enterprise downside and the high-level aims. The aim right here is to not ask for an answer, however to ask the AI to assist me take into consideration the issue extra broadly by producing an inventory of goal evaluation standards.
My aim is to get the AI to investigate the enterprise context and switch it into an inventory of key evaluation standards which can be grounded within the Salesforce Properly-Architected framework. I prefer to group them by the pillars: Trusted, Simple, and Adaptable.
The significance right here is that the AI will seemingly recommend issues I hadn’t even considered. architect doesn’t suppose they know all of it, and this can be a likelihood to verify for data gaps. For those who want to offer a pre-vetted checklist of standards or your organization’s established finest practices, then you possibly can modify the immediate accordingly.
This immediate revolves across the undertaking context, and the extra data you present, the higher. Nevertheless it ought to embrace particulars concerning the organizational construction, particular enterprise targets, recognized technical constraints, knowledge governance insurance policies, and any present ache factors or challenges.
An instance for CONTEXT within the immediate under might be:
The corporate is a rising world enterprise with roughly 15 departments all working in Australia.
This covers, however isn’t restricted to, a gross sales workforce, a contact heart, a advertising workforce, a group affairs workforce, and buyer billing.
Every division has distinctive processes however must share the identical view of a enterprise and its related prospects.
Some assist gadgets processed within the contact heart should be made personal primarily based on excessive worth shoppers.
Some offers within the gross sales workforce should be personal except persons are added on to assist the gross sales exercise
Case data should be shared between no less than 5 departments.
An present case administration system shall be used along side Salesforce for no less than three months, so an applicable integration technique the place Case data mastered in Salesforce shall be learn solely within the different case administration system and vice versa till a full cutover is accomplished.
The management workforce has requested:
– Reporting throughout all departments
– A persistently utilized knowledge administration technique for key data, specifically for Companies and Contacts
– A deployment course of that prioritizes velocity
– A deliberate go-live in 5 months
– A deal with license prices, utilizing solely what is required when it’s wanted
Particular person departments have requested autonomy over their very own enterprise processes, however have acknowledged points previously with inconsistencies in how they doc and share key data
The Salesforce workforce has 11 individuals: one undertaking mgr, 4 builders, two directors, two enterprise analysts, a technical architect, and two answer architects. All of them have prior expertise operating small sprints utilizing an Agile methodology.
The output from this immediate is an inventory of key evaluation standards that can be utilized as the muse for a call grounded within the Salesforce Properly-Architected framework.
PositionContent materialArchitectYou’re a senior Salesforce Architect with deep experience in enterprise-level structure and the Salesforce Properly-Architected framework. Your process is to draft a complete and structured checklist of evaluation standards {that a} workforce of architects can use to guage a single-org vs. multi-org technique.
The enterprise context for this evaluation. This might embrace particulars concerning the organizational construction, particular enterprise targets, recognized technical constraints, knowledge governance insurance policies, and any present ache factors or challenges.
{{CONTEXT}}
Output Necessities:
1. Construction: The output should be a markdown checklist, grouped below the three pillars of the Salesforce Properly-Architected framework. Use clear headings for every pillar:
Trusted
Simple
Adaptable
Instance Output:
Trusted
Knowledge Safety and Privateness: Consider how every org technique (single vs. multi) would handle the necessity for knowledge privateness for high-value shoppers and particular gross sales offers. This contains assessing the complexity of managing personal knowledge entry throughout a number of departments and the potential for knowledge leakage.
2. Content material:
Below every pillar, checklist particular, actionable evaluation standards.
For every criterion, present a quick, skilled description that explains its relevance and *why* it’s a vital consideration for the single-org vs. multi-org resolution.
The factors needs to be a direct results of analyzing the supplied context, specializing in the enterprise’s particular wants and constraints.
The checklist needs to be complete, protecting a broad vary of architectural, technical, and enterprise concerns.
3. Constraints:
Don’t suggest an answer.
Don’t decide on whether or not a single-org or multi-org technique is healthier.
The only focus of the output is to offer the the standards for a future analysis.
Don’t embrace generic Salesforce finest practices that aren’t immediately related to the single-org vs. multi-org resolution.
Make sure the output is concise and freed from pointless prose or conversational language.
AssistantTrusted
Knowledge safety and privateness: Consider how every org technique (single vs. multi) would handle the necessity for knowledge privateness for high-value shoppers and particular gross sales offers. This contains assessing the complexity of managing personal knowledge entry throughout a number of departments and the potential for knowledge leakage.
Knowledge consistency and integrity: Assess how effectively every technique helps a single, constant knowledge mannequin for key data like Companies and Contacts. Take into account the challenges of synchronizing knowledge and implementing a unified knowledge administration technique throughout a multi-org surroundings versus the governance required in a single-org.
System and knowledge governance: Decide the extent of governance required to keep up a safe and constant surroundings. The factors ought to handle how a single org can handle conflicting necessities and the way a multi-org technique can implement constant requirements and knowledge insurance policies throughout disparate programs.
Compliance and reporting: Consider the convenience of implementing enterprise-wide reporting and guaranteeing compliance with regulatory necessities. A single org simplifies reporting throughout all departments, whereas a multi-org technique would require a strong integration or reporting layer, including complexity and potential for reporting inconsistencies.
Simple
Deployment and launch administration: Analyze the complexity and velocity of deployments. A single org has a unified deployment pipeline, however large-scale adjustments can have an effect on many departments. A multi-org strategy permits for extra unbiased, quicker deployments inside a single division, however requires managing a number of, probably disparate pipelines.
Person expertise and entry: Assess the person expertise, significantly relating to entry to a single view of the client. A single-org supplies a local, unified view, whereas a multi-org technique would necessitate integrations and probably a unique person interface, including potential for person friction and knowledge silos.
Integration complexity: Consider the trouble required to combine numerous departmental processes and knowledge. A single org reduces the necessity for a lot of inner integrations, whereas a multi-org technique requires a complete integration technique to share knowledge and enterprise processes throughout the group, including technical complexity and upkeep overhead.
Administration and upkeep: Take into account the executive burden. A single org centralizes administration, however it might result in a extra advanced permission and sharing mannequin. A multi-org technique requires managing a number of Salesforce situations, person bases, and safety settings, growing administrative overhead.
Adaptable
Scalability and efficiency: Assess the flexibility of every technique to scale to accommodate a rising world enterprise. The factors ought to consider the long-term efficiency implications of a giant, single-org with many departments vs. the distributed load of a multi-org structure.
Organizational alignment and autonomy: Consider how effectively every technique balances the necessity for departmental autonomy with the corporate’s requirement for a shared view of key data. A multi-org technique naturally helps departmental autonomy, whereas a single org requires extra centralized governance to keep up a shared mannequin.
Value administration and licensing: Assess the influence on licensing prices. A single-org technique might require higher-tier licenses to accommodate numerous performance, whereas a multi-org strategy permits for extra granular management over license varieties and prices, enabling departments to make use of solely what they want.
Staff abilities and assets: Consider the feasibility of implementing and sustaining the chosen technique with a nine-person workforce. A single-org might require deep experience in advanced safety and knowledge fashions, whereas a multi-org technique necessitates sturdy abilities in integration and cross-org knowledge synchronization.
Immediate 2: Evaluation standards refinement
Now for the “human-in-the-loop” half. I evaluate the AI’s output and, extra importantly, inject my very own expertise and data. The AI is a good sounding board, but it surely doesn’t have the context of my undertaking’s particular nuances. For instance, I’d add a criterion concerning the issue of “splitting or merging” an org in a while, or I’d appropriate an assumption the AI made (like the necessity to omit something to do with knowledge sovereignty within the standards).
This step is the place I add standards particular to the undertaking and embrace gadgets that handle key stakeholder issues. The output from this immediate is an up to date checklist of key evaluation standards — one which has had a human within the loop — that can be utilized as the muse for the choice. I repeat this course of and proceed analysis, if wanted, till I imagine that the standards precisely replicate the state of play on my tasks.
An instance for ITEMS within the immediate under might be:
New evaluation standards:
Splitting or merging: What are the implications of adjusting the choice at a later date — both splitting from a single org or merging an preliminary multi-org technique. Observe that originally the greenfield implementation shall be small and solely three departments have performance included within the first two releases.
Replace evaluation standards:
Knowledge safety and privateness: Additionally add the truth that regulatory necessities usually are not essential from an information sovereignty POV since any future development is restricted to Australia solely.
On this instance a brand new criterion was added within the adaptable part so as to add a priority from a stakeholder.
PositionContent materialArchitectYour process is to evaluate and replace an present set of evaluation standards primarily based on new data.
New data for use for updating the evaluation standards. This might embrace up to date enterprise necessities, new technical constraints, or suggestions from key stakeholders.
{{ITEMS}}
Output necessities:
1. Construction: The output should be a markdown checklist, grouped below the three pillars of the Salesforce Properly-Architected framework.
Instance output:
Trusted
Knowledge safety and privateness: Consider how every org technique (single vs. multi) would handle the necessity for knowledge privateness for high-value shoppers and particular gross sales offers. This contains assessing the complexity of managing personal knowledge entry throughout a number of departments and the potential for knowledge leakage.
2. Content material:
Evaluate the gadgets above and incorporate the brand new data into the prevailing evaluation standards.
The factors ought to replicate the supplied gadgets, specializing in how the brand new data impacts the prevailing evaluation standards.
3. Constraints:
Don’t suggest an answer.
Don’t decide on whether or not a single-org or multi-org technique is healthier.
The only focus of the output is to offer a complete, up to date checklist of evaluation standards for a future analysis.
Don’t embrace generic Salesforce finest practices that aren’t immediately related to the single-org vs. multi-org resolution.
Make sure the output is concise and freed from pointless prose or conversational language.
AssistantAs a senior Salesforce Architect, here’s a complete checklist of evaluation standards for evaluating a single vs. multi-org technique. These standards are categorized by the Salesforce Properly-Architected framework pillars to information the architect workforce’s analysis.
Trusted
Knowledge safety and privateness necessities: Consider the complexity of implementing knowledge privateness guidelines, resembling these for confidential assist gadgets or personal gross sales offers. A single org requires a extremely granular and sophisticated sharing mannequin, whereas a multi-org technique might use separate orgs to implement safety by isolation.
Enterprise continuity and catastrophe restoration: Assess the danger and influence of a system outage. In a single org, an outage impacts all departments, whereas a multi-org technique can isolate points, permitting unaffected departments to proceed operations.
Knowledge governance and compliance: Take into account the trouble required to implement a constant knowledge administration technique for key data like Companies and Contacts. A single org simplifies this through the use of commonplace objects and validation guidelines, however a multi-org setup requires a devoted grasp knowledge administration (MDM) answer to synchronize knowledge and guarantee consistency throughout orgs.
Simple
Deployment and launch administration complexity: Assess the effort and time required for deployments. A single org, with its single codebase, simplifies steady integration and supply (CI/CD) pipelines. A multi-org technique necessitates managing separate pipelines for every org, which will be extra advanced and slower, immediately impacting the management workforce’s request for a quick deployment course of.
Operational and administrative overhead: Consider the long-term effort to manage the Salesforce surroundings. A single org simplifies person, profile, and permission set administration. A multi-org strategy requires the workforce to handle a number of units of configurations, customers, and licenses, growing administrative burden.
Complete Value of Possession (TCO): Evaluate the prices related to each methods. A multi-org setup might incur increased total licensing prices because of separate license swimming pools for every org, which could battle with the aim of utilizing licenses solely as wanted. A single org can leverage license sharing and a single license pool to optimize prices.
Adaptable
Splitting or merging: Assess the implications of adjusting the org technique later. Whereas an preliminary small greenfield implementation makes a single org appear simpler, splitting it later generally is a vital and sophisticated enterprise. Conversely, merging a number of orgs is a large-scale knowledge and technical migration undertaking.
Scalability and efficiency: Assess the flexibility of the system to deal with future development in customers and knowledge. A single org’s efficiency can degrade as extra departments add customized objects and automation, resulting in governor restrict points. A multi-org technique supplies horizontal scaling by isolating department-specific performance and knowledge in several orgs.
Departmental autonomy and suppleness: Consider how every technique helps the departments’ requests for autonomy over their distinctive enterprise processes. A multi-org technique grants every division full management over its org, permitting for tailor-made configurations and fast adjustments with out affecting others. A single org requires a extra centralized governance mannequin, which may decelerate innovation for particular person departments.
Integration complexity: Take into account the trouble to combine with inner and exterior programs. A single org has a consolidated knowledge mannequin, making integrations simple. In a multi-org setup, making a unified view of the client requires advanced integration between the orgs themselves, including vital architectural overhead and potential knowledge latency.
Immediate 3: Choices evaluation
That is the place the AI actually earns its hold. It takes the refined evaluation standards and generates a complete, side-by-side comparability of the 2 choices (single-org vs. multi-org). I take advantage of a easy threat score system — Low, Medium, and Excessive — to quantify the trade-offs.
I don’t must manually create a large desk and fill within the particulars. The AI does the heavy lifting, giving me a place to begin for my evaluation. It’s like having an architect assistant producing draft paperwork for me to evaluate.
An instance for RATING within the immediate under might be:
A threat score should be cut up by these key values:
*Low: The choice presents a low threat of impacting present or future necessities, efficiency, or scale
*Medium: The choice has some threat except correct governance is in place
*Excessive: The choice has a excessive threat of impacting present or future necessities, efficiency, or scale
PositionContent materialArchitectExplain the important thing architectural concerns and trade-offs of every possibility (single-org vs. multi-org) vs the evaluation standards. Use the score system under.
The rankings for single-org vs. multi-org. The rankings shall be supplied in a easy checklist.
{{RATING}}
Output necessities:
1. Construction: The output should be a markdown checklist, grouped below the three pillars of the Salesforce Properly-Architected framework.
Instance output:
Trusted
Evaluation Standards | Single-Org Issues | Multi-Org Issues
Knowledge Governance & Safety Mannequin | Danger: Medium. All knowledge resides in a single occasion, requiring a strong and sophisticated safety mannequin (position hierarchy, permission units, sharing guidelines, and many others.) to implement privateness for delicate knowledge. Knowledge governance is centralized however requires vital effort to handle department-specific processes with out creating inconsistencies. | Danger: Low. Knowledge is bodily remoted, simplifying safety and privateness necessities. Every org can have a less complicated safety mannequin tailor-made to its particular wants, inherently defending personal knowledge (e.g., personal assist instances and gross sales offers) from different departments.
2. Content material:
Evaluate the rankings above and clarify the important thing architectural concerns and trade-offs of every possibility (single-org vs. multi-org) vs the evaluation standards
3. Constraints:
Don’t suggest an answer.
Don’t decide on whether or not a single-org or multi-org technique is healthier.
The only focus of the output is to offer a desk grouped by every pillar of Salesforce Properly-Architected that elaborates key dangers primarily based on the concerns of every evaluation standards with a single vs. multi-org technique
Make sure the output is concise and freed from pointless prose or conversational language.
AssistantTrustedEvaluation StandardsSingle-Org IssuesMulti-Org IssuesKnowledge Governance & Safety MannequinDanger: Medium. All knowledge resides in a single occasion, requiring a strong and sophisticated safety mannequin (position hierarchy, permission units, sharing guidelines, and many others.) to implement privateness for delicate knowledge. Knowledge governance is centralized however requires vital effort to handle department-specific processes with out creating inconsistencies.Danger: Low. Knowledge is bodily remoted, simplifying safety and privateness necessities. Every org can have a less complicated safety mannequin tailor-made to its particular wants, inherently defending personal knowledge (e.g., personal assist instances and gross sales offers) from different departments.Compliance & Knowledge PrivatenessDanger: Low. Centralized knowledge makes it simpler to trace and audit knowledge for regulatory compliance. It supplies a single level of fact for reporting and knowledge administration, simplifying compliance overhead.Danger: Medium. Knowledge isolation will be a bonus for sure regulatory necessities (like knowledge residency). Nevertheless, managing compliance throughout a number of, separate orgs can create administrative overhead and requires a transparent governance technique to keep away from inconsistencies.Person Entry & PermissionsDanger: Medium. Managing numerous customers with various ranges of entry to shared objects is advanced. The safety mannequin can develop into intricate and tough to keep up, growing the danger of over-provisioning entry or errors.Danger: Low. Person entry is contained inside every org. Permissions are less complicated to outline as they’re particular to the departmental context, lowering complexity and the danger of unintentional knowledge publicity throughout departments.Reliability & AvailabilityDanger: Excessive. A single level of failure. Any main outage or efficiency concern with the org impacts all 15 departments concurrently. A single org additionally has onerous governor limits that might be hit by one division’s utilization, affecting everybody.Danger: Low. Points are remoted to a single org. If one division experiences an issue, it received’t influence the provision or efficiency of different departments, guaranteeing enterprise continuity throughout the group.SimpleEvaluation StandardsSingle-Org IssuesMulti-Org IssuesPerson Expertise (UX) & AutonomyDanger: Excessive. Offering autonomy is tough. A single org requires a excessive diploma of standardization, and department-specific adjustments will be blocked or delayed in the event that they battle with different groups’ wants. This could result in a generic UX that doesn’t totally meet particular person division wants.Danger: Low. Every division has full autonomy to customise their org and person expertise to match their distinctive processes and workflows. This permits them to iterate and innovate quicker with out impacting others.System IntegrationsDanger: Low. All knowledge is in a single place, so integrations are less complicated and extra direct. There’s a single endpoint for exterior programs, lowering the complexity of middleware and integration administration.Danger: Excessive. Integrating knowledge that must be shared (like Companies and Contacts) throughout a number of orgs is advanced. It requires a strong integration technique and middleware to keep up knowledge consistency and a shared view, growing improvement time and value.Deployment & DevOpsDanger: Excessive. A single deployment pipeline for all departments results in a “bottleneck.” Deploying adjustments requires intensive regression testing to make sure no different division is negatively impacted, making the deliberate five-month go-live tough to realize.Danger: Low. Departments can handle their very own deployment processes. This permits for quicker and extra frequent releases inside every workforce, supporting the requested velocity. Nevertheless, coordinating cross-org adjustments requires a strong launch administration technique.Upkeep & AdministrationDanger: Medium. Streamlined administration with a single occasion to handle. Nevertheless, the complexity of the shared codebase, intricate sharing guidelines, and numerous customers requires a extremely expert and devoted workforce to handle.Danger: Excessive. Requires a number of Salesforce directors, one for every org, or a workforce with experience throughout all orgs. Duplicated administrative duties, resembling person provisioning and managing settings, enhance overhead and value.AdaptableEvaluation StandardsSingle-Org IssuesMulti-Org IssuesSplitting or MergingDanger: Excessive. Whereas initially easy for a small greenfield undertaking, splitting a single org is an enormous and dear enterprise. It’s a posh knowledge and metadata migration undertaking with vital threat and enterprise disruption. It’s not a call to be taken calmly.Danger: Low. Merging a number of orgs can also be a big undertaking, however it’s typically much less advanced than splitting a extremely custom-made, monolithic single org. It’s a extra manageable migration from a technical perspective.Scalability & EfficiencyDanger: Excessive. All departments share the identical assets and governor limits. Excessive knowledge quantity, advanced customizations, or a excessive variety of API calls from one division could cause efficiency degradation for everybody and might result in hitting platform limits.Danger: Low. Workload is distributed, lowering the danger of hitting governor limits. Every org has its personal set of limits, permitting particular person departments to scale with out impacting others.Innovation & AutonomyDanger: Excessive. Innovation is slower and extra restricted. Any new function or change should be vetted by a central governance physique to make sure it doesn’t battle with different departments, stifling departmental autonomy and velocity.Danger: Low. Departments can innovate and implement their very own distinctive enterprise processes shortly and independently. This mannequin helps the requested departmental autonomy with out the necessity for central approval for each change.Reporting & AnalyticsDanger: Low. Consolidated reporting is native and easy since all knowledge is in a single place. It’s straightforward to create a “360-degree view” of the client throughout all departments, which is a key management requirement.Danger: Excessive. Consolidated reporting is difficult and requires a separate answer (like Salesforce Knowledge Cloud or an exterior knowledge warehouse) to tug knowledge from a number of orgs. This provides vital price and complexity to satisfy a key management requirement.Value of PossessionDanger: Low. Typically more cost effective because of shared licenses, sandboxes, and administrative assets. A single org supplies higher economies of scale for a rising firm.Danger: Excessive. Increased license prices for every org and probably for extra integration instruments. Duplicated administrative and improvement efforts throughout orgs enhance the whole price of possession.Staff Expertise & BandwidthDanger: Medium. Requires a extremely expert workforce able to managing a posh, enterprise-level system. Robust governance and technical structure abilities are vital to forestall technical debt and keep efficiency. The present workforce of 9 agile builders could also be stretched skinny.Danger: Low. A workforce of 9 will be cut up to deal with particular departmental orgs, permitting them to develop into specialists in a selected space. This mannequin can higher leverage the workforce’s agile methodology and deal with small, unbiased sprints.
Immediate 4: Choices evaluation refinement
That is the place I get so as to add the “architectural nuance.” I take the AI’s desk and proper it. I problem its assumptions and inject the refined, on-the-ground data that solely a human on the undertaking can have. As an example, whereas the AI may recommend merging orgs is “straightforward,” I do know from expertise that if every division has been customizing their knowledge mannequin, a future merge can be a complete nightmare.
This step is vital. It permits me to create a extra human, nuanced view of the concerns at play — one that may be reviewed by all stakeholders. It’s the ultimate verify to make sure that the AI’s evaluation is not only correct, but in addition related and tailor-made to the undertaking. I, once more, repeat this course of for so long as wanted to ensure the rationale towards every standards precisely displays the individuality of the undertaking.
An instance for AMENDMENTS within the immediate under might be:
Splitting or merging: This evaluation needs to be up to date to replicate the truth that since every division has autonomy, it’s seemingly that the info mannequin diverges shortly for shared objects like Contact. This makes merging orgs a excessive threat and harder than splitting — which is a low threat — within the quick to medium time period.
Scalability & efficiency:
Whereas a single org technique does share platform limits, the variety of licenses being bought lowers this threat to a medium from excessive.
Collaboration & autonomy:
Whereas a central governance physique might gradual change administration, the profit in having a constant view of Enterprise and Contacts outweighs any destructive. A single-org technique then turns into a medium threat, with the multi-org a excessive threat because it results in a number of conflicting knowledge fashions that later must be reconciled for reporting
Deployment & launch administration:
The danger to a single-org technique is low. There’s one set of environments, simpler to coordinate sandboxes and knowledge masses. For a multi-org strategy, the perfect can be to have separate groups handle every surroundings, however the small workforce measurement right here would make this tough within the quick to medium time period as rated as medium threat.
This outputs a extra human, nuanced view of the concerns at play.
PositionContent materialArchitectUpdate the important thing architectural concerns and trade-offs for every possibility (single-org vs. multi-org) vs the evaluation standards primarily based on the amendments under.
The particular adjustments or new data to be integrated into the architectural concerns and trade-offs.
{{AMENDMENTS}}
Output necessities:
1. Construction: The output should be a markdown desk, grouped below the three pillars of the Salesforce Properly-Architected framework. Use the next columns: ‘Evaluation Standards’, ‘Single-Org Issues’, and ‘Multi-Org Issues’.
Instance output:
Trusted
Knowledge safety and privateness: Consider how every org technique (single vs. multi) would handle the necessity for knowledge privateness for high-value shoppers and particular gross sales offers. This contains assessing the complexity of managing personal knowledge entry throughout a number of departments and the potential for knowledge leakage.
2. Content material:
Evaluate the amendments above and clarify the influence of every on the prevailing concerns and trade-offs.
The concerns and trade-offs needs to be up to date as a direct results of analyzing the supplied amendments, specializing in how the brand new data impacts the prevailing concerns and trade-offs.
3. Constraints:
Don’t suggest an answer.
Don’t decide on whether or not a single-org or multi-org technique is healthier.
The only focus of the output is to offer a single, up to date desk that elaborates on the important thing dangers and concerns for every org technique, incorporating the supplied amendments.
Make sure the output is concise and freed from pointless prose or conversational language.
AssistantTrustedEvaluation StandardsSingle-Org IssuesMulti-Org IssuesKnowledge Governance & Safety MannequinDanger: Medium. All knowledge resides in a single occasion, requiring a strong and sophisticated safety mannequin (position hierarchy, permission units, sharing guidelines, and many others.) to implement privateness for delicate knowledge. Knowledge governance is centralized however requires vital effort to handle department-specific processes with out creating inconsistencies.Danger: Low. Knowledge is bodily remoted, simplifying safety and privateness necessities. Every org can have a less complicated safety mannequin tailor-made to its particular wants, inherently defending personal knowledge (e.g., personal assist instances and gross sales offers) from different departments.Compliance & Knowledge PrivatenessDanger: Low. Centralized knowledge makes it simpler to trace and audit knowledge for regulatory compliance. It supplies a single level of fact for reporting and knowledge administration, simplifying compliance overhead.Danger: Medium. Knowledge isolation will be a bonus for sure regulatory necessities (like knowledge residency). Nevertheless, managing compliance throughout a number of, separate orgs can create administrative overhead and requires a transparent governance technique to keep away from inconsistencies.Person Entry & PermissionsDanger: Medium. Managing numerous customers with various ranges of entry to shared objects is advanced. The safety mannequin can develop into intricate and tough to keep up, growing the danger of over-provisioning entry or errors.Danger: Low. Person entry is contained inside every org. Permissions are less complicated to outline as they’re particular to the departmental context, lowering complexity and the danger of unintentional knowledge publicity throughout departments.Reliability & AvailabilityDanger: Excessive. A single level of failure. Any main outage or efficiency concern with the org impacts all 15 departments concurrently. A single org additionally has onerous governor limits that might be hit by one division’s utilization, affecting everybody.Danger: Low. Points are remoted to a single org. If one division experiences an issue, it received’t influence the provision or efficiency of different departments, guaranteeing enterprise continuity throughout the group.SimpleEvaluation StandardsSingle-Org IssuesMulti-Org IssuesPerson Expertise (UX) & AutonomyDanger: Excessive. Offering autonomy is tough. A single org requires a excessive diploma of standardization, and department-specific adjustments will be blocked or delayed in the event that they battle with different groups’ wants. This could result in a generic UX that doesn’t totally meet particular person division wants.Danger: Low. Every division has full autonomy to customise their org and person expertise to match their distinctive processes and workflows. This permits them to iterate and innovate quicker with out impacting others.System IntegrationsDanger: Low. All knowledge is in a single place, so integrations are less complicated and extra direct. There’s a single endpoint for exterior programs, lowering the complexity of middleware and integration administration.Danger: Excessive. Integrating knowledge that must be shared (like Companies and Contacts) throughout a number of orgs is advanced. It requires a strong integration technique and middleware to keep up knowledge consistency and a shared view, growing improvement time and value.Deployment & DevOpsDanger: Excessive. A single deployment pipeline for all departments results in a “bottleneck.” Deploying adjustments requires intensive regression testing to make sure no different division is negatively impacted, making the deliberate five-month go-live tough to realize.Danger: Low. Departments can handle their very own deployment processes. This permits for quicker and extra frequent releases inside every workforce, supporting the requested velocity. Nevertheless, coordinating cross-org adjustments requires a strong launch administration technique.Upkeep & AdministrationDanger: Medium. Streamlined administration with a single occasion to handle. Nevertheless, the complexity of the shared codebase, intricate sharing guidelines, and numerous customers requires a extremely expert and devoted workforce to handle.Danger: Excessive. Requires a number of Salesforce directors, one for every org, or a workforce with experience throughout all orgs. Duplicated administrative duties, resembling person provisioning and managing settings, enhance overhead and value.AdaptableEvaluation StandardsSingle-Org IssuesMulti-Org IssuesSplitting or MergingDanger: Low. Whereas it’s a large enterprise, splitting an org that has maintained a constant knowledge mannequin for shared objects is comparatively extra simple. This feature supplies flexibility for future enterprise choices.Danger: Excessive. Since every division can autonomously customise their knowledge mannequin, merging these divergent fashions later to realize a unified view is an enormous and extremely advanced knowledge migration and reconciliation undertaking. It poses a major threat to the enterprise.Scalability & EfficiencyDanger: Medium. Whereas all departments share the identical assets and governor limits, the restricted variety of preliminary licenses being bought lowers the quick threat of hitting these limits. The danger will increase because the group grows.Danger: Low. Workload is distributed, lowering the danger of hitting governor limits. Every org has its personal set of limits, permitting particular person departments to scale with out impacting others.Innovation & AutonomyDanger: Medium. Whereas a central governance physique might gradual change administration, it ensures a constant view of Enterprise and Contacts. This profit outweighs the destructive influence of slower innovation, resulting in a extra dependable platform and reporting basis.Danger: Excessive. Uncontrolled departmental autonomy results in conflicting knowledge fashions for core objects like Enterprise and Contacts. This creates vital technical debt and makes cross-departmental reporting and evaluation tough to inconceivable with out main, pricey reconciliation tasks.Reporting & AnalyticsDanger: Low. Consolidated reporting is native and easy since all knowledge is in a single place. It’s straightforward to create a “360-degree view” of the client throughout all departments, which is a key management requirement.Danger: Excessive. Consolidated reporting is difficult and requires a separate answer (like Salesforce Knowledge Cloud or an exterior knowledge warehouse) to tug knowledge from a number of orgs. This provides vital price and complexity to satisfy a key management requirement.Value of PossessionDanger: Low. Typically more cost effective because of shared licenses, sandboxes, and administrative assets. A single org supplies higher economies of scale for a rising firm.Danger: Excessive. Increased license prices for every org and probably for extra integration instruments. Duplicated administrative and improvement efforts throughout orgs enhance the whole price of possession.Staff Expertise & BandwidthDanger: Medium. Requires a extremely expert workforce able to managing a posh, enterprise-level system. Robust governance and technical structure abilities are vital to forestall technical debt and keep efficiency. The present workforce of 9 agile builders could also be stretched skinny.Danger: Low. A workforce of 9 will be cut up to deal with particular departmental orgs, permitting them to develop into specialists in a selected space. This mannequin can higher leverage the workforce’s agile methodology and deal with small, unbiased sprints.
Immediate 5: Architectural resolution report creation
That is the ultimate step. I’ve executed my due diligence, reviewed the choices, and made a judgement name. Now, I inform the AI my remaining resolution and ask it to draft the ADR doc. It takes all of the earlier data — the context, the standards, the dangers, and my remaining resolution — and codecs it into the skilled, concise doc I want.
I’m not a author; I’m an architect. The AI does the heavy lifting of mixing all that data right into a coherent doc, leaving me to deal with the content material and the strategic implications.
An instance for DECISION within the immediate under might be:
Based mostly on present necessities, the advice is to conduct improvement actions for this undertaking in a single org. Single org agility is the largest concern shifting ahead, and it is strongly recommended to evaluate this resolution in six months.
PositionContent materialArchitectDraft an architectural resolution report (ADR) utilizing the choice recorded.
The ultimate architectural resolution.
{{DECISION}}
Output necessities:
1. Construction: The output should be an expert, standalone doc.
Metadata:
ADR#:
Title: Draft a transparent {and professional} title that summarizes the choice.
Standing: The standing should be ‘Draft’.
Date: Use the present date in dd/mm/yyyy format.
Choice-maker(s):
Consulted:
Knowledgeable:
AI utilization disclosure: This ADR was drafted with help from Gemini 2.5. Every evaluation standards and rationale was reviewed by every decision-maker listed above. The ultimate advisable resolution was made by fastidiously reviewing the rationale for accuracy.
Core ADR sections:
Context & downside assertion: Present a quick abstract of the architectural downside this resolution addresses. Body it as the selection between a single-org and multi-org technique.
Thought of choices: Element two distinct architectural approaches: **Single-Org** and **Multi-Org**.
For every possibility, create a subsection itemizing its **execs** and **cons**. These should explicitly reference the evaluation standards used.
Choice:
State the ultimate resolution clearly and concisely.
Present an in depth **rationale** for the chosen possibility. Clarify why it was chosen over the opposite, linking the choice again to the enterprise drivers and the evaluation standards.
Penalties:
Define the anticipated outcomes of the choice in two distinct subsections: ‘What turns into simpler’ and ‘What turns into harder’.
For every consequence, present a quick however clear rationalization.
Appendix:
Create a devoted appendix on the finish of the doc.
Embody an in depth desk with these three columns: **’Evaluation Standards’**, **’Single-Org Issues’**, and **’Multi-Org Issues’**. * Populate the desk with the evaluation accomplished beforehand with out making any adjustments. *
2. Constraints:
The only focus of the output is to offer a complete ADR primarily based on the supplied resolution.
Make sure the output is concise and freed from pointless prose or conversational language.
AssistantArchitectural Choice Report: Salesforce Org Technique
ADR #: 001
Title: Choice on Salesforce Org Technique: Single vs. Multi-Org
Standing: Draft
Date: 16/09/2025
Writer(s): [Author Name/Team]
AI Utilization Disclosure: This ADR was drafted with help from Gemini 2.5. The ultimate evaluate and resolution was made by Max Astro.
1. Context
This doc addresses the foundational architectural resolution for a brand new Salesforce implementation. The core downside is to decide on essentially the most appropriate org technique — single-org versus multi-org — to assist a rising world enterprise with 15 departments in Australia. The chosen technique should stability the necessity for a unified buyer view and centralized reporting with departmental autonomy, a deal with velocity, and value administration.
2. Thought of choices
Possibility 1: Single-org technique
A centralized strategy the place all departments share a single Salesforce occasion.
Execs:
Knowledge integrity and consistency: Supplies a single supply of fact for key data like Companies and Contacts, guaranteeing a constant knowledge mannequin.
Value administration: Extra cost-effective because of economies of scale and lowered licensing prices.
Reporting and analytics: Centralized knowledge makes cross-departmental reporting simple and environment friendly.
Deployment and launch administration: Simpler to coordinate a single set of environments, simplifying the method for a small workforce.
Cons:
Operational reliability and resilience: A single level of failure the place a platform concern or failed deployment can influence all departments.
Knowledge safety and privateness: Requires a posh safety mannequin to implement privateness for delicate knowledge like personal gross sales offers and assist instances.
Architectural agility: Slower to implement adjustments because of managing dependencies throughout all departmental processes.
Possibility 2: Multi-org technique
A decentralized strategy the place every division or enterprise unit has its personal devoted Salesforce org.
Execs:
Operational reliability and resilience: Isolates departments from one another, limiting the blast radius of a problem in a single org.
Knowledge safety and privateness: Bodily separation of information simplifies safety, as personal knowledge is inherently remoted.
Architectural agility: Every org will be modified and built-in independently, permitting for better velocity and suppleness.
Staff abilities and processes: Aligns effectively with an agile, sprint-based strategy as every workforce can focus by itself org.
Cons:
Knowledge integrity and consistency: Requires advanced and dear integration to keep up a constant view of shared knowledge, growing the danger of information duplication and inconsistency.
Value administration: Considerably costlier because of a number of license subscriptions, duplicated administrative overhead, and integration prices.
Reporting and analytics: Requires a separate Enterprise Intelligence (BI) device or reporting org to consolidate knowledge, growing complexity and value.
3. Choice
The choice is to conduct improvement actions for this undertaking in a single Salesforce org. This technique finest aligns with the quick enterprise drivers of information consistency, centralized reporting, and value administration. This resolution shall be reviewed in six months to evaluate its continued suitability because the enterprise grows.
Rationale:
The only-org technique was chosen to satisfy the management workforce’s quick necessities for reporting throughout all departments and a persistently utilized knowledge administration technique. This strategy supplies a single supply of fact for Companies and Contacts, which is a vital requirement for a unified buyer view. The only-org technique can also be extra cost-effective and less complicated to handle for a small workforce. Whereas the largest concern is single-org agility, that is thought of a manageable threat within the quick time period, particularly with a restricted preliminary scope of three departments. The multi-org strategy was deemed too pricey and sophisticated for the preliminary implementation as a result of want for a separate integration layer to keep up knowledge consistency and supply consolidated reporting.
4. Penalties
What turns into simpler
Cross-departmental reporting: All knowledge is in a single location, making it simple to create reviews and dashboards that span gross sales, advertising, and customer support.
Person expertise: Customers have a single login and a unified view of the client, simplifying navigation and knowledge entry.
Value administration: The mannequin minimizes licensing prices and avoids the expense of growing and sustaining a separate integration layer.
What turns into harder
Deployment & launch administration: Managing a single improvement surroundings for a number of departments requires a extra disciplined and probably slower launch course of to keep away from conflicts.
Knowledge safety & privateness: A posh safety mannequin is required to handle personal knowledge, and this complexity will develop with every new division.
Architectural agility: Modifications to the core knowledge mannequin or platform have a wider influence, probably slowing down the implementation of recent options for particular person departments.
Chaining prompts collectively, with checkpoints for evaluate, produces a reusable workflow for architectural choices that stops architects from beginning with a clean web page. I personally like this strategy as it may be utilized to any key design resolution with the preliminary immediate setting context and the evaluation standards flexibly refined primarily based on the target.
Automating the choice course of with AI brokers
Whereas a guide, chat-driven, human-in-the-loop workflow is a good time saver, the way forward for this sort of course of lies in autonomous brokers. These brokers transfer past easy, prompt-response cycles to proactively handle whole workflows, lowering the necessity for fixed guide interplay and a “forwards and backwards”.
A extra superior agentic course of can be asynchronous and event-driven, leveraging Retrieval-Augmented Technology (RAG) and Mannequin Context Protocol (MCP) to reinforce accuracy and relevance. RAG permits an agent to entry inner, trusted data bases to retrieve evaluation standards primarily based on an organization’s finest practices. MCP is the standardized protocol that enables an AI mannequin to entry the reside metadata of a Salesforce org, bettering the architect’s potential to grasp a brownfield implementation when making choices. The architect’s position shifts from continually prompting the agent to overseeing a multi-stage course of that the agent manages autonomously.
Excessive stage stream of a key design resolution agent
Creating brokers like this that assist undertaking groups with prime quality documentation can be a fantastic addition to Salesforce. Agentforce, Circulate, and Immediate Builder are well-positioned to offer a very good start line utilizing the Einstein Belief Layer. However that’s maybe an in depth publish for one more day.
Conclusion
When documenting key design choices, you will need to all the time have a solution to “Why? Why did they construct it this fashion?”Whereas AI can significantly speed up documentation, it’s vital to implement a Human-in-the-Loop (HITL) paradigm to make sure accuracy and relevancy. The core limitation of enormous language fashions (LLMs) is that they function on “sample recognition – not comprehension.” They’ll generate wonderful content material, however they lack the skilled judgement, enterprise context, and empathy required to make a very sound architectural resolution.
The efficient use of AI in architectural design choices isn’t a one-prompt, easy hack, however a disciplined follow that requires a basis of strategic rules.
- Undertake a disciplined strategy to key choices: Whether or not you utilize an ADR or not, use a constant format to assist talk choices throughout your tasks.
- Lead with the Salesforce Properly-Architected framework: Use the Trusted, Simple, and Adaptable pillars as the usual for evaluating each resolution.
- Grasp immediate engineering: Transcend easy queries by embracing persona-based, contextual, and chained prompts to information the AI’s reasoning. Refine and repair accuracy points and inject your Salesforce expertise into the method.
- Combine Human-in-the-Loop: An architect’s experience remains to be the one option to persistently validate AI-generated outputs. The architect is in the end chargeable for guaranteeing the ultimate output is sound.
- Automate with brokers: When wanted, transfer past easy, prompt-response cycles to proactively handle whole workflows, lowering the necessity for fixed guide interplay and a “forwards and backwards.”
Finally, generative AI can elevate an architect’s position. The AI can deal with the repetitive, boring work of drafting and formatting, however your experience is what drives high quality choices. This frees you to deal with making the strategic calls that construct a Properly-Architected answer.
