Graybeard
02.3 How to evaluate it

The AI support landscape for technical fields

Four categories. One map. Find the row that fits your situation. Notice where the row you might be defaulting to actually leaves you.

Chapter 2.37 min readHow to evaluate it

"AI customer support" covers four structurally different categories. Vendor pitches blur the distinctions on purpose, and most evaluation rubrics inherited from B2C and SaaS peers don't map cleanly to built-environment manufacturers.

THE AI SUPPORT LANDSCAPE Operational lift on one axis, content depth on the other. VERTICAL- TUNED content GENERIC content DIY / SELF-DEPLOY your engineering team MANAGED / DONE-FOR-YOU the vendor operates BUILD-YOUR-OWN VERTICAL Custom RAG. In-house authored content. Multi-quarter project. Requires an AI/ML team. High control. High burden. EXAMPLE PATH Pinecone, Vectara, AWS Q MANAGED AI SERVICES FOR TECHNICAL FIELDS Vertical-tuned content. Authored and operated for you. 2-3 week onboarding. Multi-year ops relationship. Fast time-to-value. ★ WHERE THIS BLUEPRINT POINTS ENTERPRISE SEARCH PLATFORMS DIY content. Generic shape. IT team required. Enterprise-tier license. 6-month implementation. No trade-specific authoring. EXAMPLE VENDORS Glean, Coveo, Algolia GENERIC AI CUSTOMER-SERVICE Managed setup. Generic content. Built for digital-native CX. Fast deploy. Strong on FAQs. No depth on multi-revision hardware or schematics. EXAMPLE VENDORS Fin, Decagon, Sierra CRM BOLT-ONS · captive to platform
Operational lift (DIY vs. managed) on one axis. Content depth (generic vs. vertical-tuned) on the other.
Orientation

Reading the map

The vertical axis is content: the bottom is generic, the top is vertical-tuned for a specific trade. The horizontal axis is operational model: the left is DIY (you bring engineering), the right is managed (the vendor brings operations). Each category sits in the quadrant its architecture commits it to.

Category 1

Enterprise search platforms

Bottom-left. Glean, Coveo, Algolia. Enterprise search infrastructure with optional generative layers.

Where it fits
  • Large enterprises with mature IT
  • Multi-quarter implementation runways
  • Generic content (intranet, help-center)
  • Internal-use cases sharing the platform
Where it doesn't
  • 1-50-person support teams without IT
  • Multi-revision documentation
  • Trades needing labeled diagrams
  • Mid-market cost structure
Category 2

Generic AI customer-service agents

Bottom-right. Fin, Decagon, Sierra, Intercom. AI agents designed for digital-native customer service.

Where it fits
  • SaaS, B2C, e-commerce
  • Account, billing, return-shape journeys
  • Chat-style customer interaction
  • Well-structured help-center content
Where it doesn't
  • Built-environment manufacturers
  • Schematic and multi-revision content
  • Field techs needing cited answers
  • Asymmetric cost of wrong answers
Category 3

CRM bolt-ons

Middle of the bottom band. Zendesk Answer Bot, Salesforce Service Cloud Einstein. AI capability built into an existing CRM platform.

Where it fits
  • Existing host-CRM customers
  • Low-complexity FAQ shapes
  • Incremental support-workflow gains
  • No appetite for another vendor
Where it doesn't
  • Companies not on the host CRM
  • Content too rich for thin AI
  • Multi-source or schematic-heavy answers
  • Anyone wanting vertical authoring
Category 4

Managed AI services for technical fields

Top-right. Managed AI customer support purpose-built for trade verticals in the built environment. The vendor authors structured content on top of your existing documentation, and owns setup, monitoring, authoring, and monthly reporting.

Where it fits
  • 1-50-person support teams
  • Security, fire, HVAC, lighting, electrical, sensors
  • Real documentation, no in-house AI team
  • Time-to-value in weeks, not quarters
Where it doesn't
  • Enterprise teams with budget for Glean/Coveo
  • Digital-native SaaS businesses
  • Companies wanting to own the AI stack
  • Industries the vendor hasn't authored for

This category is the newest of the four and currently sparse. The sparseness is structural: vertical-content authoring depth takes years, and it doesn't happen as a side effect of building generic AI customer service.

In practice

Bring this map to your evaluation

The map is diagnostic, not prescriptive.

  • Place your own company on the map. Do you have an in-house IT/ML team? Is your documentation generic or trade-specific?
  • Place every candidate by architecture, not by pitch. A vendor pitching into a quadrant they don't structurally occupy is a signal.
  • Ask explicitly: "Where do you see yourselves on this map, and where do you see us?" Good vendors answer crisply. Over-extended vendors hedge.
The most useful evaluation question

Ask each candidate to name a customer they're not a good fit for. Vendors who answer crisply ("we don't fit if your support content is mostly account-style") understand their quadrant. Vendors who can't are positioning themselves into territory they don't occupy.

Pitch decks look similar. Quadrant placements are structural. Get the placement right early and the rest of the evaluation compresses dramatically.

What good looks like
A CS director using the map well:
  • Has placed their own company by operational model and content type
  • Has placed every candidate by architecture, not pitch
  • Has eliminated quadrants that structurally don't fit before scheduling demos
  • Has used the "name a non-fit profile" question as a signal
  • Has acknowledged the trade-off: a managed service is a multi-year relationship
Next · Chapter 2.4
Build, buy, or hire
Get started

Want help placing your company on the landscape?

We walk through your operational model, content type, and constraints, and show you which quadrant actually fits.

Talk to us →