← ccPlanning Academy · Advanced track

Planning for automation & AI deflection

Free visual lesson · about 5 minutes · short quiz at the end

ccPlanning academy · advanced

Planning for automation & AI deflection

Forecasting a baseline that keeps shifting under your feet.

The big idea

Deflection breaks “history repeats.”

Every forecast assumes the past predicts the future. When a bot, an AI assistant or a new self-service flow goes live, it deliberately changes the past’s relationship to the future — volume that used to reach agents now doesn’t. The planner’s job is to forecast through a moving target.

Effect 1

Volume falls — but unevenly.

Automation removes the contacts it’s good at — balance checks, password resets, simple tracking — first and most. It barely touches the complex ones. So total volume drops, but the mix shifts towards the hard stuff. A flat “volume down 20%” assumption misses this entirely.

Effect 2

AHT rises — the residual-complexity effect.

Because deflection skims off the quick, easy contacts, what reaches an agent is the harder residue. Average handle time goes up even though volume went down. Plan for the same total workload reduction as the volume drop and you’ll be short — you removed the cheap minutes, not the expensive ones.

Effect 3

The mix — and the channel — shifts.

Deflected customers don’t always vanish; some escalate to chat or call back angrier, having failed with the bot. Automation reshapes the whole demand pattern across channels, not just the headline number. Watch where the deflected contacts re-emerge.

The moving baseline

Forecast the deflection curve, not a step.

Adoption ramps — a new bot deflects little at launch and more as customers learn it (and as it improves). Model deflection as a curve climbing over months, with explicit assumptions, rather than a one-off cliff on go-live day. And expect the vendor’s promised deflection rate to be optimistic.

How to plan it

Treat it as a tracked what-if.

Build the automation as an explicit scenario overlay (from the capacity track): a deflection assumption on volume, an uplift on AHT, a channel-shift, ramping over time. Then watch actuals like a hawk and re-baseline fast — early deflection data beats any pre-launch estimate.

Why “volume down 20%” under-staffs

The cheap minutes left, not the work

The bot deflects 20% of contacts — but they’re the 3-minute balance checks, not the 12-minute complaints. So volume drops 20% while AHT climbs, say, 15%. Net workload falls maybe 8%, not 20%. Staff to the headline volume cut and you’re short every day.

Automation removes the cheap minutes and leaves the expensive ones. Plan the workload, not the contact count.

The takeaway

Less volume, harder mix, higher AHT — on a ramp.

Automation cuts the easy contacts, raises AHT through residual complexity, reshuffles the channel mix, and ramps over time. Model it as an explicit, time-phased scenario rather than a flat volume cut — then trust early actuals over the vendor’s promises.

Now test yourself ↓

1 / 8

Slides done? Here’s the same idea in a bit more depth — the part worth keeping.

In depth: forecasting a baseline that keeps shifting

Every forecast assumes the past predicts the future. When a bot, an AI assistant or a new self-service flow goes live, it deliberately changes the past’s relationship to the future — volume that used to reach agents now doesn’t. So the planner’s job becomes forecasting through a moving target, and a flat “volume down 20%” assumption misses almost everything that actually matters.

Three effects, all uneven

First, volume falls unevenly: automation removes the contacts it’s good at — balance checks, password resets, simple tracking — first and most, and barely touches the complex ones, so the mix shifts toward the hard stuff. Second, AHT rises through the residual-complexity effect: because deflection skims off the quick, easy contacts, what reaches an agent is the harder residue, so average handle time goes up even as volume goes down — plan for the same workload cut as the volume drop and you’ll be short, because you removed the cheap minutes, not the expensive ones. Third, the channel mix shifts: deflected customers don’t always vanish, some escalate to chat or call back angrier having failed with the bot, so watch where the deflected contacts re-emerge.

Model the curve, trust the actuals

Deflection isn’t a step, it’s a curve — a new bot deflects little at launch and more as customers learn it and as it improves — so model it climbing over months with explicit assumptions rather than a cliff on go-live day, and expect the vendor’s promised deflection rate to be optimistic. The practical method is to treat automation as a tracked what-if: an explicit scenario overlay with a deflection assumption on volume, an uplift on AHT and a channel-shift, ramping over time — then watch actuals like a hawk and re-baseline fast, because early deflection data beats any pre-launch estimate.

The principle to remember: less volume, harder mix, higher AHT — on a ramp. Model automation as an explicit, time-phased scenario rather than a flat volume cut, and trust early actuals over the vendor’s promises.

Quick quiz

Five questions. Pick an answer to each, then check your score.

1. Why does deflection challenge normal forecasting?

Automation moves the target — volume that used to reach agents now doesn’t.

2. How does automation affect contact mix?

A flat ‘volume down 20%’ assumption misses the shift towards complexity.

3. What is the residual-complexity effect on AHT?

You removed the cheap minutes, not the expensive ones — plan workload accordingly.

4. How should deflection be modelled over time?

Adoption ramps; expect vendor deflection promises to be optimistic.

5. What’s the practical way to plan an automation launch?

Early deflection data beats any pre-launch estimate — watch actuals and adjust quickly.