Salesforce Agentforce Partner: 10 Questions Before You Sign
Choosing a Salesforce Agentforce partner? Ask these 10 questions on proof, data, guardrails, cost and handover before you sign a statement of work.
22 Oct 2026·6 min read·Salesforce AI

A partner can demo an Agentforce agent in an afternoon. Running one for a year, inside your real data and your real rules, is a different job.
If you're picking a Salesforce Agentforce partner, that gap is what you're really buying. Most buyers compare logos, certifications and rate cards. I think that's the wrong test. Here's my argument: choose the partner who can show you an agent in production and explain how it will be run after go-live, not the one with the best demo.
One disclosure up front. I run Techila Global Services, a Salesforce Summit Partner, so I'm not neutral. Put these ten questions to us as well.
Key takeaways
- Judge a Salesforce Agentforce partner on agents running in production, not on demos or certificates.
- Test how they decide data readiness, scope, guardrails and testing before they quote a price.
- Ask how they forecast consumption cost and what handover looks like, because that's where surprises hide.
- Score every shortlisted partner on the same ten questions and keep the answers in writing.
Why a Salesforce Agentforce partner is a different hire from a CRM partner
Agentforce projects fail in different places from classic CRM builds, so the partner needs different skills.
A classic Salesforce build is deterministic: you configure a flow and it does the same thing every time. An Agentforce agent reasons over your data and acts, so quality depends on data, permissions, testing and ongoing tuning as much as on configuration.
That's a judgement from delivery work, not a statistic. But it explains why the questions below look different from a standard implementation checklist. You can find consulting partners in Salesforce's own AppExchange partner directory, but a listing only gets a partner onto your shortlist. The questions decide who wins. If you're still deciding whether you need outside help at all, read when to hire a Salesforce AI consultant.
Questions 1 to 3: proof and people
Ask to see real production work, meet the real team, and speak to a real customer.
- "Show me an agent you built that's live today." Not a sandbox, not a slide. At Techila we implemented Agentforce for customer support at Armakuni, a UK cloud engineering consultancy. The agent answers technical FAQs, routes cases and escalates to a human with full context. A good partner can describe their equivalent in that much detail.
- "Who exactly will work on my project, and can I meet them?" Sales teams and delivery teams are often different people. Ask for names, roles and Salesforce credentials, which you can check against Trailhead profiles.
- "Can I speak to a customer whose project finished recently?" Ask that customer what went wrong, not what went well.
My take: Certifications are the entry ticket. A partner who can't name a live agent and let you talk to its owner is selling a roadmap, not experience.
Questions 4 to 6: data, scope and guardrails
A good partner defines the agent's job, checks the data and limits its access before building anything.
- "How do you decide whether my data is ready, and whether I need Data 360?" The honest answer is "it depends on the use case". Be wary of a partner who says you always need it, or never do.
- "What's the agent's job description?" I think of an agent as having six parts, like a new hire: a brain (the model, which also shapes Salesforce Claude fees and Agentforce pricing), a job description, tools, knowledge, a trigger and guardrails. If the proposal covers only the model and the tools, the rest will be improvised on your budget.
- "How will you test it, and what does it do when it's unsure?" Look for test cases built from your real conversations, defined escalation rules, and read-only access first with a human approving actions. I run three agents in my own business this way, and nothing is sent, paid or changed without my approval. Agents should earn promotions, just like people.
Questions 7 and 8: what it will cost to run
Ask for a consumption forecast in writing, because the build fee is the smaller number.
- "How will you forecast what this costs to run?" Agentforce pricing options vary, so check your contract and Salesforce's Agentforce pricing page. Then run your own numbers in the Agentforce pricing calculator. A partner should show their assumptions: volumes, actions per conversation, and a high case as well as a base case, and they should be able to explain Agentforce pricing, Flex Credits and cost per action.
- "What about API and MCP calls from agents?" Salesforce plans to meter every successful call a registered agent makes to a production org over MCP or the API as a "Headless Platform Interaction", with 30 days' notice before it starts (reported target: November 2026; see Understand AIforce Impact). Check the current wording before you sign. Ask the partner to list every agent and integration that touches Salesforce and to design for batching: one call per task, not per step.
My take: Usage billing rewards tidy design. A partner who has never thought about calls per task will hand you an agent that's expensive to run.
Questions 9 and 10: life after go-live
The contract should say who runs the agent, who improves it, and what you own when the partner leaves.
- "Who monitors and tunes the agent after launch, and what does support cost?" Agents drift as products, policies and customers change. Ask for a named owner, a review cadence and the metric they'll report, ideally hours saved or cases resolved.
- "If we part ways, can my team run this?" Ask about documentation, admin training, and who owns the prompts, test sets and configuration. A partner confident in their work isn't afraid of a clean exit.
A quick scorecard
Score each partner 0 to 2 on every question and compare totals.
| Area | Strong answer | Red flag |
|---|---|---|
| Proof | Named live agent, reference call offered | Demo org only |
| Team | Delivery people named and available | "We'll staff it after signing" |
| Data | Use-case-based readiness check | Same Data 360 answer for everyone |
| Cost | Written forecast with high case | "Usage is hard to predict" |
| Handover | Documented, trained, owned by you | Dependency by design |
If you want to start small, the AI-Native Enterprise Framework I'm developing is one way to choose the first use case. It's my own proposal, not an industry standard.
Try this week:
- Write your shortlist's answers to questions 1, 7 and 9 side by side.
- List every system and agent that would touch Salesforce and estimate calls per day.
- Pick one boring, frequent, rule-based, low-risk task as the first agent, and write its job description.
- Ask for one reference call with a customer live for six months or more.
If you'd like to test your own shortlist against these questions, get in touch.
Frequently asked questions
How do I choose a Salesforce Agentforce partner?
Shortlist from the AppExchange partner directory, then ask each one for a live production agent, named delivery staff, a cost forecast and a handover plan. Score them on the same questions.
Does a Salesforce Summit Partner badge guarantee a good Agentforce outcome?
No. The tier shows a partner's standing within Salesforce's programme, but it doesn't tell you whether the specific team assigned to you has shipped an agent. Ask for that proof directly.
How much does an Agentforce implementation cost?
It depends on scope, data readiness and the pricing model in your contract, so check your contract and the Agentforce pricing page. Model running costs with the pricing calculator before you approve the build.
Should I hire a partner or build Agentforce in-house?
If your team has strong admins and architects, start in-house with one low-risk use case and bring in a partner for architecture reviews. If you lack that capacity, a partner can move faster, provided you plan for handover from day one.
Want to put this to work in your business?
Join one of my live sessions, or tell me what you're trying to do with AI.
Newsletter
Get the next one by email.
New writing on Salesforce AI, AI agents and the AI-Native Enterprise. Confirm by email; unsubscribe any time.
Keep reading
2 Nov 2026 · Salesforce AISalesforce Workbench: How to Use It Safely (No Undo)Salesforce Workbench explained for admins and developers: run SOQL, export data, retrieve metadata and follow the safety rules most people skip.8 min
1 Nov 2026 · Future of workSalesforce Administrator Certification: Worth It in 2026?Is the Salesforce administrator certification worth it in 2026? See the US$200 cost, exam changes, what employers want and an 8-week plan.7 min
31 Oct 2026 · Salesforce AIHow to Debug a Salesforce Flow: Fix Common ErrorsHow to debug a Salesforce Flow step by step: use debug mode, add fault paths, read debug logs and stop error emails. Fixes for the errors admins hit.10 min