The demo always works. The question every buyer should ask is who builds the workflows, guardrails, and integrations after the contract is signed - your team, or theirs.
AI support vendors with forward-deployed engineering assign a dedicated product manager and engineer to build your initial workflows, guardrails, and integrations for you, instead of handing you a builder and a help center. This is the done-for-you model, and in 2026 it is the dividing line between platforms that go live in regulated environments and platforms that stall in a backlog. This guide ranks seven vendors on that single lens: who does the setup.
The implementation model, not the feature list, is the strongest predictor of whether an AI support deployment reaches production in regulated industries.
Forward-deployed (done-for-you) means a vendor PM and engineer build your first workflows, integrations, and guardrails. Self-serve means you build them, sometimes with a paid services package.
Lorikeet runs a forward-deployed model: a dedicated PM and engineer build the initial workflows, guardrails, and integrations, with a sandbox live in 20 to 30 minutes and most accounts operational in roughly one month.
Premium forward-deployed vendors (Sierra, Decagon) embed engineers but at enterprise contract sizes; self-serve and partner-led models (Agentforce, Cognigy, Fin, Ada) shift the build to you or a system integrator.
The honest trade-off: done-for-you means faster, safer launches and less control over the day-to-day build; self-serve means full control and a slower path to a compliant production agent.
Last updated: June 2026
Buying AI support is not like buying software you log into and use. The product you evaluate in a demo is a finished agent someone already built. The product you actually buy is a platform plus the work of turning it into that agent: mapping your ticket types, writing the workflows, wiring the integrations, and proving the guardrails hold before a regulator ever looks. That work is where most deployments die. A self-serve platform with a great builder still needs someone on your side with the time, the engineering, and the domain knowledge to build it. A forward-deployed vendor brings that someone with them. This is a buyer-neutral ranking on one axis only: who does the setup, and how done-for-you the first ninety days really are.
What is forward-deployed AI support engineering?
Forward-deployed engineering is an implementation model where the vendor assigns a dedicated product manager and engineer to build your AI agent for you - the workflows, the integrations, the guardrails - rather than giving you a self-serve builder and documentation. The term comes from enterprise software (Palantir popularized it), and it has become the default for serious AI support deployments because configuring a regulated agent is closer to a build project than a settings change.
The category splits cleanly into three implementation models. Forward-deployed (done-for-you): a vendor team builds the first version with you, then hands over ownership. Self-serve with professional services: you build it, optionally buying a paid onboarding or services package. Partner-led: a system integrator or agency does the build, with the vendor a step removed. The model matters more than most feature comparisons admit, because the best agent platform in the world still produces nothing until someone builds your workflows on it.
Forward-deployed engineer (FDE): A vendor engineer embedded with your team who writes the actual workflows, integrations, and guardrails for your deployment, rather than advising you while you build them yourself.
Done-for-you setup: An onboarding model where the vendor delivers a working, configured agent against your real ticket types, as opposed to a blank builder you populate from scratch.
Lorikeet is an AI customer support platform built for complex, regulated companies like fintechs, healthtechs, and insurers. It runs a forward-deployed model: a dedicated product manager and engineer build your initial workflows, guardrails, and integrations, get a sandbox live in 20 to 30 minutes, and have most accounts operational in roughly one month. Lorikeet resolves issues end-to-end across chat, email, voice, SMS, and WhatsApp, and pairs a Concierge agent with a Coach agent that runs automated quality assurance on every ticket.
Self-Serve vs Forward-Deployed: Which Implementation Model Fits You
Most buying guides treat implementation as a footnote. For AI support it is the main event. Here is how the two models actually differ once the contract is signed.
Who builds the first workflows. In a forward-deployed model, the vendor's PM and engineer map your ticket types and build the initial flows. In a self-serve model, your team does, learning the builder as they go. The forward-deployed path turns institutional knowledge about your edge cases into the agent faster, because the people building it do this full-time.
Who owns the integrations. Wiring an agent into Stripe, Salesforce, a core banking system, or a pharmacy API is engineering work. Forward-deployed vendors do it for you with scoped, least-privilege access. Self-serve and partner-led models put that work on your engineers or a system integrator, which adds a queue and a dependency.
Who proves the guardrails. In regulated industries, a compliance team has to sign off before launch. A done-for-you vendor builds and tests the guardrails (disclosures, escalation triggers, action limits) and hands you the evidence. A self-serve model leaves your team to build that test discipline themselves.
Time to production. Done-for-you typically reaches a working production agent in weeks. Self-serve depends entirely on your team's bandwidth and can stretch into quarters, especially when the build competes with other roadmap work.
The honest trade-off. Forward-deployed gets you live faster and safer, but you cede some control over the day-to-day build and you depend on the vendor's responsiveness. Self-serve gives you full control and no external dependency, at the cost of speed and the risk that a complex agent never quite gets finished. If your team has spare senior engineers and a simple use case, self-serve can win. If your tickets are regulated and your engineers are busy, done-for-you almost always wins.
The 7 Best AI Support Vendors by Implementation Model in 2026
1. Lorikeet
Lorikeet is the forward-deployed AI support platform built for complex, regulated companies. A dedicated product manager and engineer build your initial workflows, guardrails, and integrations with you, get a sandbox live in 20 to 30 minutes, and have most accounts operational in roughly one month. The model is done-for-you by design: Lorikeet treats setting up a regulated agent as a build project staffed by people who do it every day, not a help-center task left to your team.
Key Features
Forward-deployed PM and engineer who build the first workflows, integrations, and guardrails against your real ticket types, then hand over ownership.
Combinable natural-language and deterministic structured workflows in a single interaction, all configurable in plain English so your team can own them after launch.
Defense in depth: pre-launch adversarial simulations, inbound message checks, outbound guardrails, and 100% post-facto QA via the Coach agent. The bad paths get tested before you ship.
End-to-end resolution across chat, email, voice (sub-1-second latency), SMS, and WhatsApp, plus outbound re-engagement, on one workflow engine.
Least-privilege scoped integrations with Zendesk, Intercom, Front, Salesforce, Twilio, and core systems; SOC 2, BAA-ready for HIPAA, GDPR-aligned, with data residency in the US, AU, and UK.
Ideal For
Fintechs, healthtechs, and insurers whose compliance team is the toughest stakeholder in procurement and whose engineers do not have spare quarters to build an agent from a blank builder. Lorikeet's customers are roughly 80% US financial institutions and fintechs. A regulated fintech reached around 85% automation with equal-or-better CSAT after deployment, and cross-border payments customers report meaningful retention lifts on AI-handled tickets versus human-handled ones.
Limitation
Lorikeet is purpose-built for complex, regulated workflows. A small team with a simple FAQ deflection use case and no compliance burden may find a self-serve drop-in tool faster and cheaper to stand up, and will not use the depth Lorikeet is built for.
Pricing
Per-resolution: roughly $0.80–$0.95 per chat, email, or SMS resolution and about $1.20–$1.50 per voice resolution; Coach runs around $0.25–$0.30 per ticket. The customer defines what counts as a resolution and escalations are not charged. For comparison, human-handled tickets typically cost $1.25 to $4 each.
2. Sierra
Sierra, the enterprise AI agent company from Bret Taylor and Clay Bavor, runs a high-touch forward-deployed model with embedded staff during the launch period. It is a genuine done-for-you experience at the top of the market, paired with outcome-based pricing.
Key Features
Embedded Sierra implementation staff who build the agent during deployment.
Outcome-based pricing: customers pay primarily when the AI fully resolves a case.
Voice, chat, and email channels with a branded agent persona approach.
Strong enterprise procurement story and analyst attention.
Ideal For
Large enterprises that want a premium forward-deployed build and have the procurement appetite for an enterprise contract on AI support alone.
Pricing
Not published. Enterprise contracts are reported in the low-to-mid six figures annually, with per-resolution rates negotiated case by case.
3. Decagon
Decagon is the high-end enterprise AI agent platform that sells white-glove deployment with embedded engineering during launch. It is forward-deployed in the same premium tier as Sierra. The honest read on this tier: the embedded team is partly a feature and partly a necessity, because the platform is hard to configure alone.
Key Features
White-glove deployment with embedded engineering during the launch period.
Per-conversation or per-resolution pricing, customer-selectable.
Voice, chat, and email in one platform.
Production deployments processing large interaction volumes for enterprise customers.
Ideal For
Large enterprises with multi-million-dollar support budgets that want a top-of-market premium vendor and can sustain a months-long, high-touch deployment.
Pricing
No published rates. Industry data suggests a platform fee plus per-conversation or per-resolution fees, with median total contract value reported near $400,000 per year.
4. Salesforce Agentforce
Salesforce Agentforce is the agent layer inside the Salesforce platform. Its implementation model is primarily self-serve plus partner-led: you configure it with your admins, or a Salesforce system integrator does the build. There is no embedded forward-deployed vendor team in the Lorikeet or Sierra sense. Lorikeet coexists with Agentforce in accounts that keep Salesforce as the CRM.
Key Features
Native to the Salesforce platform and data model, ideal for existing Salesforce shops.
Configured by Salesforce admins or implemented through the large SI partner ecosystem.
Deep access to Salesforce CRM data and the broader Salesforce app surface.
Per-action consumption pricing tied to the Salesforce contract.
Ideal For
Enterprises standardized on Salesforce with internal admin capacity or an SI relationship, that want their support agent inside the same platform as their CRM.
Pricing
Consumption-based per action, layered onto existing Salesforce licensing. Effective cost depends heavily on configuration and SI fees.
5. Cognigy
Cognigy is an enterprise conversational AI and contact-center automation platform with a strong low-code builder. Its model is self-serve plus partner-led: enterprises build flows in the visual designer, often alongside a Cognigy partner or SI. The build power is real, but the build is yours or a partner's, not an embedded Cognigy product team's.
Key Features
Low-code visual flow builder for voice and chat automation.
Strong contact-center and IVR heritage with broad telephony integrations.
Large partner and system-integrator ecosystem for delivery.
Enterprise deployment options including on-premise and private cloud.
Ideal For
Enterprises with a contact-center automation focus and an internal team or SI partner ready to own the flow-building work in a low-code environment.
Pricing
Not published publicly; enterprise licensing quoted by sales, typically with partner implementation fees on top.
6. Fin by Intercom
Fin by Intercom is the AI agent layered on Intercom's messenger and helpdesk, and it is the clearest self-serve, drop-in model in this guide. You enable it, point it at your content and actions, and it starts resolving. There is no embedded vendor engineer building your workflows; the trade is speed and simplicity for done-for-you depth.
Key Features
Self-serve activation with a fast trial-to-deployment path.
Low published per-resolution pricing, among the lowest in the category.
Works with Salesforce and HubSpot helpdesks, not only Intercom.
Optional copilot for human agents.
Ideal For
High-volume teams, often already on Intercom, with relatively standard ticket types that want the fastest self-serve path to a live AI agent and the lowest published per-outcome price.
Pricing
Around $0.99 per resolved outcome, plus a helpdesk seat fee if not already an Intercom customer, with optional copilot and analytics add-ons.
7. Ada
Ada is an established AI agent vendor that began as a chatbot platform and expanded into voice and email. Its model is self-serve with a professional-services and onboarding layer: you build in Ada's tooling, with paid services available to help. It sits between pure self-serve and forward-deployed, without the embedded product-team build of the premium tier.
Key Features
Self-serve builder with optional paid onboarding and professional services.
Multi-channel coverage across chat, voice, and email.
Mature integrations with Salesforce, Zendesk, and major helpdesks.
Established deployment playbooks for mid-market and enterprise.
Ideal For
Mid-market and enterprise teams with high chat volume that prefer a long-track-record vendor and have internal capacity to drive the build, with services support as needed.
Pricing
Not published publicly. Marketplace data shows median annual contracts around $70,000, varying with company size and services scope.
If a vendor's first ninety days depend on your engineers having spare quarters, the demo is the easy part. See how Lorikeet's forward-deployed team builds your first workflows, guardrails, and integrations.
How to Choose Between Done-For-You and Self-Serve
The right model is the one that matches your team, your tickets, and your risk profile. These five questions surface the answer faster than a feature matrix.
Who on your side will build the agent?
If the honest answer is a senior engineer with months of dedicated time, self-serve is viable. If it is a support leader between other priorities, a done-for-you vendor that brings the PM and engineer is the safer bet. Be realistic about whether the build will get the bandwidth it needs once the novelty wears off.
How complex are the tickets you need automated?
FAQ deflection is buildable self-serve. Multi-step regulated workflows (KYC unlocks, disputes, claims, transfers) that chain several tool calls and need state and recovery are a build project. The more complex the ticket, the more a forward-deployed model earns its place. See also: end-to-end resolution.
Does a compliance team need to sign off before launch?
If yes, ask who builds and proves the guardrails. A done-for-you vendor builds the disclosures, escalation triggers, and action limits and hands you evidence from pre-launch simulations. A self-serve model leaves that discipline to you. In regulated industries this is usually decisive.
Who owns the agent after launch?
Forward-deployed should not mean permanently dependent. Ask whether you can own and edit the workflows after handover. Lorikeet's plain-English natural-language and structured workflows are built so your team owns them post-launch; a model where every change is a vendor ticket is a hidden cost.
What does time-to-production actually look like?
Ask for a realistic date to a production agent on your hardest ticket type, not a sandbox demo. Done-for-you vendors should commit to weeks. If a self-serve answer is vague, that vagueness is the timeline.
Questions to ask your vendor
Demos are built to look finished. These questions reveal who does the real work.
Who writes my first workflows and integrations: your engineer or mine?
Will you build my guardrails and hand me the pre-launch test results, or is that my responsibility?
What is a realistic date to a production agent on my hardest ticket type?
After handover, can my team edit workflows ourselves, or does every change route back to you?
If I have no spare engineers, can you still get me live, and how?
Is your services help included or a separate paid package?
Lorikeet's Take on Forward-Deployed AI Support
Most vendors will sell you a builder and a promise that it is easy. The builder usually is easy. Turning it into a regulated agent that your compliance team signs off on is not, and that is the work that decides whether a deployment ever reaches production. The reason Lorikeet runs a forward-deployed model is that the customers we serve, mostly US fintechs and healthtechs, do not have spare engineering quarters to spend learning a workflow builder, and their tickets are too consequential to ship on faith.
So a dedicated PM and engineer build the first version with you, prove the guardrails with adversarial simulations before launch, and hand over workflows you can edit in plain English afterward. The goal is a working agent in roughly a month, owned by your team, not a permanent dependency on ours. If your hardest tickets are regulated and your engineers are busy, that is the model that gets you live. See how Lorikeet handles end-to-end resolution.
Key Takeaways
Implementation model is the most underrated line item in an AI support purchase, and the strongest predictor of reaching production in regulated industries.
Forward-deployed (done-for-you) means a vendor PM and engineer build your workflows, integrations, and guardrails. Lorikeet, Sierra, and Decagon sit in this tier; Lorikeet is purpose-built for regulated workflows with a sandbox live in 20 to 30 minutes and most accounts operational in about a month.
Salesforce Agentforce and Cognigy are primarily self-serve plus partner-led; Fin by Intercom is self-serve drop-in; Ada is self-serve with a professional-services layer.
The trade-off is real: done-for-you is faster and safer with less day-to-day control; self-serve is full control at the cost of speed and a risk the build never finishes.
Whatever model you choose, confirm you can own and edit the agent after launch so forward-deployed does not become permanently vendor-dependent.
Conclusion
The seven vendors above split cleanly by who does the setup. Lorikeet, Sierra, and Decagon embed a team that builds for you; Agentforce, Cognigy, Fin, and Ada hand the build to you or a partner with varying levels of help. Neither model is wrong. The wrong move is choosing on features and discovering in month three that nobody had the time to build the agent the demo promised.
If your tickets are regulated, your compliance team is your toughest stakeholder, and your engineers are already booked, the done-for-you model is built for exactly that situation. Lorikeet is the answer for complex, regulated companies that want a forward-deployed team to build the first version and a platform their own team can own afterward. The other six are credible depending on your in-house capacity and how standard your tickets are.
If you are evaluating AI support and want to know what done-for-you actually looks like, book a Lorikeet demo and bring your hardest workflow - our forward-deployed team will scope the build with you.
