CS Transformation Leader Shares What AI Can (and can’t) Fix in Customer Success

We turn customer success into a revenue engine, driving millions of dollars in expansion and retention.

Why AI is a double-edged sword in CX

Integrating AI into CX has delivered meaningful gains, but also surfaced some real challenges.

Here's the upside:

• Significant time savings: AI has reduced manual effort across prep, follow-ups, CRM updates, and reporting. In many cases, I’ve seen CSMs save 1–2 hours per day, which they can reinvest into customer-facing work.
• Stronger, faster customer insights: Tools like Claude, Gainsight, Gong AI, and others are surfacing real-time signals — such as churn risk, expansion opportunities, and shifts in customer health — that previously took hours (or days) to uncover.
• Improved consistency and scalability: AI helps standardize outputs like success plans, QBRs, and account summaries, which is critical when scaling teams or supporting different segments.
• More proactive CX motions: Instead of reacting late, teams can act earlier based on AI-driven signals—whether that’s risk mitigation or expansion plays.

But there's a downside too:

• Signal vs. noise problem: Not all AI-generated insights are actionable. Teams can get overwhelmed with alerts that lack context or prioritization.
• Over-reliance on automation: Some teams lean too heavily on AI-generated content (QBRs, emails), which can lead to generic, low-impact customer interactions.
• Data quality dependency: AI is only as good as the data behind it. If your CRM or product data is incomplete or inaccurate, the outputs quickly lose credibility.
• Loss of critical thinking: There’s a risk that CSMs stop thinking strategically and rely on AI recommendations instead of applying judgment.

Why good data is critical for good AI outputs

Here's what every CX leader needs to know: Good data is everything.

I’ve worked with CS organizations that have invested in great AI tools, but their underlying data was messy or incomplete. This completely limited the value that they got from it.

One client didn’t have proper account hierarchies set up in Salesforce. For example, they had 15 separate accounts with the same name that were part of the same customer — but they weren’t connected. When they layered AI on top of that, the system generated insights at the individual account level, but couldn’t provide a holistic view of the customer relationship.

So instead of better insights, they ended up with fragmented and sometimes misleading signals.

It's worth the time investment to collect good data and clean bad data!

How AI and human tasks should be split in CX

AI is most powerful in tasks that are data-heavy, repeatable, or time-consuming…Humans are critical anywhere context, trust, and nuance matter.

Share This Quote on:

• Share on Twitter
• Share on LinkedIn
• Share on Facebook

Andrea BumsteadOpens new window

Founder and CEO of CS Impact

My view on AI versus human in Customer Experience is that AI should handle scale, signals, and speed. Humans should handle trust, judgment, and transformation.

AI is most powerful in tasks that are data-heavy, repeatable, or time-consuming:

• Administrative work: Prep for calls, account summaries, stakeholder research, and post-call follow-ups—including notes, next steps, and CRM updates (Salesforce, Gainsight)
• Content and asset generation: Success plans, QBRs, account plans, white space analysis, and customer communications
• Insight generation: Identifying at-risk accounts (usage drops, champion loss, M&A activity), expansion opportunities, and product feedback trends
• Guidance at scale: Surfacing the right risk or expansion playbooks at the right times
• Digital CX: In-app messaging, onboarding flows, knowledge bases, and support content

Humans are critical anywhere context, trust, and nuance matter:

• Customer conversations and relationship building: Especially with executive stakeholders
• Discovery and problem framing: Understanding not just what’s happening, but why
• Reading between the lines: Picking up on sentiment, risk signals, and what’s not being said
• Consultative guidance: Helping customers make decisions, and driving real business outcomes
• Moments that build trust: Executive meetings, in-person dinners, events, and high-impact interactions

How automated QBRs lead to meaningful customer conversations

Andrea Shares

We have helped companies automate by using AI to gather data and insights, integrating them into a repeatable format, and optimizing delivery for in-person or electronic presentations…As a result, these companies can have more meaningful conversations with more customers — and deliver more value.

I mentioned QBRs — we created the 15-minute QBR. It covers:

1. The value the customer has received to date
2. How they get more value
3. Three recommendations for next steps

We have helped companies automate this by using AI to gather data and insights, integrating them into a repeatable format, and optimizing delivery for in-person or electronic presentations. As far as tools, we're aggregating data across Salesforce, Gong, and Gainsight to automatically generate QBR decks using Claude.

As a result, these companies can have more meaningful conversations with more customers — and deliver more value.

Before implementing the 15-minute QBR, I recommend CS leaders focus on five critical foundations:

1. Get clear on the problems you solve: Identify the top 2–3 business problems your product solves — this becomes the anchor for every customer conversation.
2. Define objectives and success metrics: Align on KPIs tied to value realization so you can clearly measure and communicate impact.
3. Track customer progress: You need visibility into how each customer is performing against their goals, not just product usage.
4. Benchmark your top customers: Understand what your best customers are doing differently so you can highlight gaps and opportunities.
5. Build a recommendations engine: Develop repeatable, high-impact recommendations that help customers unlock more value.

From there, I recommend a simple rollout approach:

1. Start small: Pilot with a focused cohort and refine the approach.
2. Automate where possible: Leverage AI to pull data and generate insights at scale.
3. Then scale: Expand across segments once it’s proven.
4. And just as important, enable the team to deliver it effectively: Train them to lead with confidence, ask strong discovery questions, and run the conversation from a consultative, value-driven perspective.

Andrea Shares

Start small. Automate where possible then scale. And just as important, enable the team to deliver it effectively.

Where AI falls short with QBRs and content

One challenge with the QBR workflow is that, while AI excels at generating content, it still struggles to structure it in a usable, business-ready format, like an executive-ready deck.

In practice, AI accelerates the first 70% of the work, but the final 30% — which often makes it usable and impactful — still requires human refinement.

How AI improves forecasting in CX — and what's still needed

I’ve used Gong for forecasting across several organizations, and it has the strongest capabilities I’ve seen in Customer Success.

It aggregates signals across the account: customer sentiment, activity levels, and key data from Salesforce like ARR, TCV, deal stage, notes, and next steps.

That gives CS leaders a much more holistic, real-time view of account health and potential outcomes.

That said, the process is still largely manual. CSMs are still responsible for:

• Managing the contact process in Salesforce
• Updating opportunity stages
• Making judgment calls on renewals — whether an account will come in flat, downgrade, or expand.

The next evolution is AI-assisted forecasting. Instead of starting from scratch, AI should:

• Analyze historical trends
• Incorporate health scores and product usage
• Pull in signals from customer conversations
• And generate a recommended forecast.

At that point, the CSM’s role would shift to refining and validating the forecast, not building it from scratch.

I also see a gap in visibility into change over time. Forecasting shouldn’t just be a point-in-time view — it should clearly show week-over-week movement and exactly what drove that change. For example, did a large account suddenly become at risk? Or did we successfully recover revenue we previously flagged as churn?

That level of visibility, combined with AI-driven recommendations, would not only improve forecast accuracy but make it far more actionable, scalable, and aligned with how CS teams operate.

Forecasting shouldn’t just be a point-in-time view — it should clearly show week-over-week movement and exactly what drove that change…Don’t be afraid to experiment and try new things. Don’t be afraid to fail.

Share This Quote on:

• Share on Twitter
• Share on LinkedIn
• Share on Facebook

Andrea BumsteadOpens new window

Founder and CEO of CS Impact

Why CX leaders must become comfortable with failure

Here's my advice: Don't be afraid to experiment and try new things. Don't be afraid to fail.

Try new AI workflows, take courses, and expand your horizons.

Pavilion has a strong course on GTM and AI. And Customer Success Collective events, like their Customer Success Summits across the US, UK, and APAC, focus on AI-driven workflows.

Follow along

Follow Andrea Bumstead on LinkedIn for insights on transforming Customer Success into a revenue engine, building high-impact CS strategies, and leveraging AI to drive customer value. Subscribe to her Substack, Real Talk with CS Impact, for practical playbooks and frameworks. And follow along on Instagram.

More expert interviews to come on The CX Lead!