It answers with last quarter's facts
Products, prices and policies moved. Nobody connected “we changed the returns window” to “the agent needs re-indexing and re-testing”.
Agentforce · Support & Managed Services
Who owns the agent on an ordinary Tuesday?
Agentforce managed services means a named team responsible for your agents in production — monitoring and incident response, release and regression management, knowledge and action upkeep, consumption reporting, and a service level you can hold us to. Projects end. Agents keep talking to customers. Someone has to be accountable for that on a Tuesday.
12+
500+
40+
250+
Trusted by 500+ organizations — many of whom stayed with us long after the project that introduced us ended.
Service Scope
- Salesforce Partner
- HubSpot Partner
- Monitoring
- Incident Response
- Release Management
- Knowledge Upkeep
- Consumption Reporting
- Quarterly Review
What happens to an agent nobody owns
None of this is dramatic. That is exactly the problem — an unowned agent degrades quietly for months and is switched off in a single meeting. Decay is the default, not the exception.
An action broke and nobody was watching
An expired credential or a changed API silently fails. The agent apologises politely to every customer for two weeks before someone escalates it.
A platform release changed behaviour
Salesforce ships three major releases a year. Without someone reading them against your configuration, changes are discovered by customers.
Consumption climbs and nobody can explain it
Finance asks why the bill grew forty percent. Without usage attributed to topics and design decisions, the only available answer is a shrug.
Permissions drifted after a reorg
A permission set edited for another purpose quietly widened what the agent can read. Nobody re-verified it, and the audit finds it first.
The person who understood it left
One admin knew why each instruction was written that way. When they moved on, nobody dared change anything — so the agent stopped improving.
What It Means
An agent is a product, not a project
Agentforce managed services is a standing commitment to keep your agents working. It covers five things: watching production and responding when something breaks, managing releases so changes ship safely, maintaining the content and actions the agent depends on, reporting consumption so cost stays explicable, and meeting regularly to decide what should change next. It is bought as a service level rather than a project, because the work arrives continuously and unpredictably.
What makes agents different from other Salesforce assets is the number of things that can change underneath them without anyone touching the agent. A report breaks when someone edits the report. An agent degrades when a content owner rewrites an article, when a product launches without documentation, when an admin updates a Flow the agent invokes, when a credential expires, when a platform release changes a default, or when legal amends a policy. Every one of those is somebody's routine Tuesday, and none of them files a ticket.
Salesforce provides the platform, its release cadence, and the observability and testing tools. Twopir Consulting provides the standing team and the routine: a named owner who knows your configuration, monitoring with agreed response times, regression testing before every promotion, knowledge and action upkeep, consumption reporting in language finance accepts, and a quarterly review with decisions on record. You get an agent that is as accurate in month twelve as it was in week one, and somebody to call when it is not.
We are explicit that this should be optional. Some clients take it in-house after enablement, and that is a good outcome — we build the test sets, dashboards and documentation to be handed over. What we will not do is finish an implementation and leave the run state unassigned, because that is the most reliable way we know to waste the whole investment.
Three ways to cover the run state, and what each includes
Scope and response times are agreed per engagement rather than sold from a fixed menu. The shape below is what these tiers typically cover, so you can see where your situation sits before a conversation.
| Covered | Support | Managed | Managed + Improve |
|---|---|---|---|
| Incident response | Reactive, agreed response times | Proactive, with alerting | Proactive, with alerting |
| Monitoring & dashboards | You watch, we respond | We watch and report | We watch and report |
| Release & regression | On request | Every change, regression tested | Every change, regression tested |
| Knowledge & action upkeep | Not included | Re-index and re-test on change | Includes content remediation |
| Consumption reporting | Not included | Monthly | Monthly, with cost tuning |
| Improvement backlog | Not included | Maintained, not delivered | Delivered in tuning cycles |
| Named owner | Shared team | Named owner | Named owner and architect |
The work of keeping an agent honest
Six standing responsibilities. Most of it is unglamorous and most of it is what separates an agent that is still trusted next year from one that quietly gets switched off.
Monitoring & Incident Response
Alerting on the things that indicate real trouble — action failure rates, escalation spikes, containment drops, latency — and a named person who responds inside agreed times rather than a shared inbox.
- Threshold alerting on the metrics that matter
- Triage, diagnosis and fix or rollback
- Agreed response times by severity
- Root-cause notes, not just closure
- Escalation path into your team when needed
Release & Regression Management
Every change to topics, instructions, actions or grounding goes through the same path: sandbox, regression run against the test set, promotion, and a rollback plan that somebody has actually thought about.
- Change intake and impact assessment
- Regression run before every promotion
- Metadata deployment through your pipeline
- Rollback plan per release
- Testing & deployment →
Knowledge & Action Upkeep
The single largest source of quiet decay. We watch for content and product changes, re-index and re-test when they land, and feed the questions the agent could not answer back into the content backlog.
- Re-indexing on content and product change
- Retrieval re-testing after any edit that matters
- Unanswered questions routed to content owners
- Action health checks and credential renewal
- Data & knowledge integration →
Consumption & Cost Reporting
Usage attributed to topics and design decisions rather than presented as one number, so an increase can be explained — and so the levers that reduce it are visible before renewal season.
- Monthly usage broken down by topic and channel
- Cost per resolved conversation, tracked over time
- Variance explained against forecast
- Test-run budget planning
- Early warning before a threshold is crossed
Governance Re-Verification
Permissions drift. We re-check what the agent user can actually read, confirm Trust Layer settings still match policy, and keep the audit evidence current rather than reconstructing it under pressure.
- Periodic permission and sharing re-verification
- Trust Layer configuration checks
- Audit evidence maintained continuously
- Platform release review against your config
- Governance & security →
Quarterly Service Review
The meeting where decisions get made rather than reported. Performance against target, incidents and their causes, consumption trend, the improvement backlog, and what the next quarter should prioritise.
- Performance against agreed targets
- Incident summary with root causes
- Consumption trend and forecast
- Ranked improvement backlog
- Optimization & performance →
The surfaces where trouble shows up first
An agent has more dependencies than most Salesforce assets. These are the ones that break it, and what we are watching for on each.
Agent observability
Containment, escalation, topic mix and latency, watched against thresholds rather than reviewed monthly. A ten-point containment drop overnight means something broke, and it should page someone the same day.
Action health & credentials
Failure and timeout rates per action, plus credential and certificate expiry dates. An expired named credential is the most preventable outage in this whole category and one of the most common.
Knowledge & content changes
Article edits, new publications and retirements all change what the agent says. We watch the change stream and trigger re-indexing and retrieval re-tests rather than waiting for a complaint.
Salesforce release cycle
Three major releases a year, each capable of changing a default or deprecating something you rely on. We read them against your configuration and test in the preview window rather than after.
Automation the agent depends on
A Flow edited for an unrelated reason can change what an agent action does. We keep the dependency map so a routine admin change is assessed against the agent before it ships, not afterwards.
Permissions & the agent user
What the agent user can read, re-verified on a cycle. Permission sets get edited for other reasons, and a widened boundary is invisible in agent behaviour until it appears in an audit.
Consumption & thresholds
Usage tracked against forecast with early warning before a commitment is crossed, so the conversation with finance happens in advance of the invoice rather than in response to it.
Real conversations
Dashboards tell you something is wrong; transcripts tell you what. A standing sample read every cycle is how new failure patterns get caught while they are still small.
Onboarding an agent we did not build
Most managed-service engagements start with somebody else's configuration and no documentation. Two to three weeks of onboarding turns that into something a service can actually stand behind.
Inventory & Document
Every topic, action, library, permission and dependency written down — usually the first time it exists in one place. This becomes the artefact that survives staff turnover.
Establish a Baseline
Current performance measured and a regression test set built from real sessions, so we can prove the agent is no worse after we take over — and better later.
Instrument & Alert
Dashboards and thresholds configured for the metrics that indicate real trouble, with alerting routed to a named owner rather than a distribution list.
Agree the Service
Response times by severity, change intake route, who may approve what, escalation into your team, and the reporting cadence — written down and signed rather than assumed.
Run It
Monitoring, releases, upkeep and reporting on the agreed rhythm, with a quarterly review where the backlog is re-ranked and the next quarter's priorities are decided.
Support that clients kept paying for
Post-go-live ownership has been part of how we work since long before agents existed. These are engagements where staying past launch was the point.
★★★★★
Working with Twopir Consulting was a game-changer for our organization. They designed and implemented a comprehensive Service Cloud + Experience Cloud solution that perfectly aligned with our customer support and partner portal needs. Their proactive approach, attention to detail, and post-go-live support have been outstanding.
Andrew Stewart Consultant · Service Cloud delivery & post-go-live support Post Go-Live
Professional Events Organization — CRM Integration
Salesforce–MeetMax integration for a corporate networking organization.
100% Elimination of manual data sync between platforms
360° Unified client view across events & CRM
0 Manual reconciliation tasks post-integration
★★★★★
We engaged Twopir Consulting to conduct a Salesforce audit, and their structured, insight-driven approach exceeded our expectations. Their team quickly understood our complex processes, identified critical gaps, and provided clear, actionable recommendations. A knowledgeable and dependable partner.
Kelly Hale Managing Director · Ongoing Salesforce partnership Dependable
Professional Services Firm — AI Document Automation
Automating email attachment processing and Salesforce data routing.
40% Reduction in manual document processing time
2× Admin throughput without added headcount
100% Automated classification & CRM routing
We would rather be boring and reliable
Managed services is a trust purchase. What matters is whether the same people answer, whether they know your configuration, and whether the thing you were promised actually happens.
A named owner who knows your build
Not a rotating queue. Someone who knows why an instruction is worded that way, which action is fragile, and what your last incident was — because context is most of the value.
We can fix the cause, not just restart it
Our service team is the same practice that builds. When the fault is a Flow, an integration or a data problem rather than the agent, we fix that rather than filing it with someone else.
Nothing ships without a regression run
Every change to a live agent is regression tested against the current test set before promotion. Small changes have surprising blast radius on a non-deterministic system.
Cost is reported, not absorbed
Consumption broken down by topic and explained against forecast, so the renewal conversation is informed rather than defensive — and the levers to reduce it are already on the table.
We build toward you not needing us
Documentation, test sets and dashboards are written to be handed over. Clients who take the run state in-house after a year are a success, not a loss.
Service questions, answered honestly
Yes, and it is a large share of this work. We start with two to three weeks of onboarding: inventory every topic, action, library, permission and dependency, establish a performance baseline, and build a regression test set from real sessions. That produces documentation that usually did not previously exist and lets us commit to a service level honestly. We will not take on a service level for a configuration we have not inventoried — that is a promise nobody can keep.
Broader than an outage, which is what makes agent incident management different. Severity one is the agent being unavailable or giving harmful or non-compliant answers. Severity two is an action failing systematically, or containment collapsing. Severity three covers a specific topic answering wrongly or latency degrading. A single bad answer is not an incident — it is backlog input — but a pattern of them is. Agreeing these definitions during onboarding avoids arguing about them during a real one.
Far less than an implementation and far more than zero. The load is driven by how fast your business changes rather than by conversation volume: a company shipping new products monthly generates continuous content and testing work, while a stable policy-driven service agent may need little between quarterly reviews. The recurring baseline is monitoring, release management when something changes, re-indexing and re-testing after content changes, and the quarterly review. Ask any partner to describe their recurring work concretely — vagueness there is a warning sign.
Often yes, and we build toward it. What an in-house team needs is a named owner with real time allocated — not a fractional add-on to an admin's existing role — plus the test set, the dashboards, the documentation and a habit of regression testing before promotion. We hand all of those over deliberately. The honest failure mode we see is the hybrid where an agent is nominally owned by someone with no capacity; that is worse than either extreme, because everyone assumes it is covered.
Yes — reading each release against your specific configuration, testing in the preview window, and flagging both risks and capabilities worth adopting. Salesforce ships three major releases a year and AI features move faster than the rest of the platform, so a default can change or a new capability can make part of your build obsolete. Agents are more exposed to this than most Salesforce assets because their behaviour depends on so many moving platform pieces.
Managed services keeps the agent from getting worse; optimization makes it better. The first is a standing commitment covering monitoring, incidents, releases and upkeep. The second is project-shaped work: a diagnostic, a ranked backlog, and tuning cycles measured against a baseline. They combine naturally — the top tier includes improvement cycles — but they answer different questions, and buying optimization without an owner for the run state means the gains erode within months.
Next Step
If nobody can name the owner of your agent, that is the conversation
Tell us what is live, who currently looks after it, and what happened the last time something broke. We will tell you what a realistic service looks like — including whether you would be better off running it yourselves.
Salesforce, CRM & AI delivery for growing and mid-market companies · Contact the team
