Nine Agents Target ServiceNow. How to Actually Pick One.

Your company standardized on ServiceNow for IT service management years ago. Now the ITSM AI vendors are in your inbox, each one promising to deflect a big share of your tier-1 tickets and shrink the fulfiller-seat bill. Our catalog models seat compression for nine agents against ServiceNow, and they are not nine versions of the same product. The list spans turnkey deflection agents, build-it-yourself conversational platforms, native add-ons, and one outright platform replacement:

• Salesforce Agentforce IT Service: 50% compression on ServiceNow
• Kore.ai: 35%
• Moveworks: 30%
• ServiceNow Now Assist: 30%
• Cognigy: 30%
• Microsoft Copilot Studio: 30%
• Aisera: 25%
• Yellow.ai: 25%
• Microsoft 365 Copilot for Service: 15%

Nine vendors, one tool, nine percentages. The instinct is to run a bake-off, sign the two that demo best, and let them fight it out in production. That instinct sets money on fire. Here is how the pick actually works.

Stacking buys nothing

The first thing a CFO needs to internalize is that signing two of these agents does not add their compression. Our engine resolves overlapping agents on a single tool with one line:

effectivePct = Math.max(...matches.map((m) => m.pct))

If you select Moveworks (30%), Aisera (25%), and Now Assist (30%) all pointed at ServiceNow, the compression that lands is 30%, not 85%. Two agents cannot both own the deflection of the same ticket. Whichever one answers it, answers it. The remaining tickets escalate to humans because they need human judgment, and a second bot installed alongside the first does not make those tickets more deflectable. We wrote up the full reasoning in the MAX-overlap rule explained, but the practical consequence for a ServiceNow buyer is blunt: you pick one, and the second subscription is money spent for zero incremental deflection.

Watch what stacking does at scale. Take a 2,500-seat ServiceNow estate (more on that number below). Selecting Moveworks plus Aisera plus Now Assist gives you 30% compression, the same as Moveworks alone or Now Assist alone. But you pay all three bills. Their combined year-one cost lands near $2.8 million against $900,000 of theoretical deflection value. Net: roughly negative $1.9 million, for the exact same 30% you would have gotten from one agent. Stacking is not a hedge. It is triple billing.

The three things that actually decide it

Once you accept that you are picking one, the decision reduces to three questions.

Fit. Three product shapes hide inside this list, and they are not interchangeable. Four are turnkey deflection agents that sit on top of ServiceNow while it stays: Moveworks, Now Assist, Aisera, and M365 Copilot for Service. Four are conversational-AI platforms: Kore.ai, Cognigy, Yellow.ai, and Microsoft Copilot Studio sell you the tooling, and your team builds and maintains the agent that does the deflecting. And one is a migration: Agentforce IT Service is pitched as a ServiceNow replacement, which is not a bigger version of the same decision. The risk lives in a different column for each shape.

Fully-loaded cost. Flat fee plus setup, or per-user times deployment size. This is where most bake-offs go wrong, because a per-user price that looks cheap on the vendor slide can be the most expensive option once you multiply by your seat count.

Deflection you believe. The compression percentage is a belief about how much of your ticket volume the agent removes. It is not a guarantee. A 50% number that assumes you migrate platforms is a very different bet than a 25% number that assumes tier-1 deflection on the stack you already run. And a 30% number on a platform assumes the bot your team builds on it is good.

Worked example: 2,500 fulfiller seats at the catalog rate

Take an enterprise running ServiceNow for ITSM with 2,500 active fulfiller seats. Our catalog reference rate is $100 per seat per month (ServiceNow is sales-led and does not publish this, so treat it as a modeling anchor, not a quote). That puts the ServiceNow line at $3,000,000 a year. We are isolating ServiceNow here: most of these nine also carry compression numbers against tools this example does not include (Freshservice, Zendesk, the collaboration suites), so their totals would shift on a broader stack. This estate runs its service desk on ServiceNow alone.

Compression value first. Each agent removes floor(2,500 x pct) seats worth of the ServiceNow bill:

• Agentforce IT Service at 50%: 1,250 seats, $1,500,000 a year
• Kore.ai at 35%: 875 seats, $1,050,000
• Moveworks, Now Assist, Cognigy, and Copilot Studio at 30%: 750 seats each, $900,000
• Aisera and Yellow.ai at 25%: 625 seats each, $750,000
• M365 Copilot for Service at 15%: 375 seats, $450,000

Now the year-one fully-loaded cost, which is where the ranking scrambles. Our engine charges 12 months of subscription plus one-time setup. Flat-fee agents default their setup to the larger of a $15,000 floor or one month of subscription unless the catalog carries an explicit figure; per-user agents carry no separate setup:

• Copilot Studio ($200 a month flat): $2,400 subscription plus $15,000 setup floor, $17,400 total
• Aisera ($9,000 a month flat): $108,000 plus $15,000 setup, $123,000
• Yellow.ai ($12,000 a month flat, explicit $30,000 setup): $144,000 plus $30,000, $174,000
• Agentforce IT Service ($20,000 a month flat, explicit $50,000 setup): $240,000 plus $50,000, $290,000
• Kore.ai ($25,000 a month flat): $300,000 plus $25,000 setup, $325,000
• Cognigy ($25,000 a month flat): $300,000 plus $25,000 setup, $325,000
• Moveworks ($33,000 a month flat): $396,000 subscription plus $33,000 setup, $429,000
• M365 Copilot for Service ($30 per user): 2,500 x $30 x 12, $900,000
• Now Assist ($75 per user): 2,500 x $75 x 12, $2,250,000

Subtract cost from theoretical deflection value and the theoretical net looks like this:

• Agentforce IT Service: positive $1,210,000
• Copilot Studio: positive $882,600
• Kore.ai: positive $725,000
• Aisera: positive $627,000
• Yellow.ai: positive $576,000
• Cognigy: positive $575,000
• Moveworks: positive $471,000
• M365 Copilot for Service: negative $450,000
• Now Assist: negative $1,350,000

The vendor with the cheapest per-seat sticker (Now Assist, the native ServiceNow add-on at $75 a module) is the worst deal on this estate by a wide margin. And the strangest row sits one line from the top: a $200-a-month credit pack posts the second-best theoretical net in the field. That row is not the same kind of product as its neighbors, and it gets its own section below.

The per-user trap hiding in this list

Now Assist and M365 Copilot for Service both quote a per-user rate. Per-user pricing scales with the population that deploys the agent, and our engine sizes that population as the largest active-seat count among the tools the agent compresses. On a ServiceNow-only estate, that is all 2,500 fulfillers. So the $75-per-user native add-on becomes $2,250,000 a year, and it deflects $900,000. The $30-per-user Microsoft option becomes $900,000 and deflects $450,000. Both lose, and both lose because the per-seat number that sounded modest got multiplied by the whole estate. The full mechanic is in the deployment-size trap.

One honest asterisk on M365 Copilot for Service: if you already license base Microsoft 365 Copilot org-wide, the Service role folds in at close to zero marginal cost, which flips it positive. The engine models the standalone $30, so check whether you are already paying for the base seat before you write it off.

What realization does to the ranking

Theoretical net is the ceiling. Year one rarely hits the ceiling, because deflection ramps as the agent learns your ticket taxonomy and your team reroutes work. Our action-plan math discounts new agent deployments to 40% of theoretical in year one. Apply that 0.4 factor to the deflection value and the year-one picture changes again:

• Copilot Studio: $360,000 realized minus $17,400, positive $342,600
• Agentforce IT Service: $600,000 realized minus $290,000, positive $310,000
• Aisera: $300,000 realized minus $123,000, positive $177,000
• Yellow.ai: $300,000 minus $174,000, positive $126,000
• Kore.ai: $420,000 minus $325,000, positive $95,000
• Cognigy: $360,000 minus $325,000, positive $35,000
• Moveworks: $360,000 minus $429,000, negative $69,000
• M365 Copilot for Service: $180,000 minus $900,000, negative $720,000
• Now Assist: $360,000 minus $2,250,000, negative $1,890,000

The ramp year punishes heavy flat fees. Moveworks was positive $471,000 on paper and flips to negative $69,000 in year one; its 30% deflection is real, but $33,000 a month is a heavy fee to carry while the agent is still learning. It clears easily in year two when setup is amortized and realization steps up, which is exactly the kind of timing a CFO should plan for rather than be surprised by. Kore.ai takes the same hit without flipping: positive $725,000 theoretical shrinks to positive $95,000 in year one, because $325,000 of platform cost meets only $420,000 of ramped value. And Copilot Studio, whose cost is too small for the ramp to hurt, takes over the top of the table. The reasoning behind the 0.4 factor is in why deploy realization is 0.4, not 1.0.

The cheapest row is a platform, not an agent

Before anyone circles the Copilot Studio number, be clear about what that row is. Microsoft Copilot Studio is a low-code bot-building platform, and the $200 a month our catalog models is a 25,000-credit consumption pack at a cent a credit. It is tooling, not a finished ITSM agent. The 30% deflection is what a well-built bot on the platform can reach in an M365-heavy shop, and the building is your team's job: connector wiring, topic design, testing, prompt maintenance, and ownership when it breaks. That is internal engineering work, it recurs, and the engine's $15,000 setup floor stands in for it rather than pricing it. Consumption billing also means the credit bill scales with conversation volume instead of staying pinned at one pack. We do not track internal build labor in the catalog and we will not invent a dollar figure for it here, but you should budget it, because the $342,600 is a best case for a team that builds well and maintains what it built.

Kore.ai, Cognigy, and Yellow.ai are the same shape at contact-center scale: enterprise conversational-AI platforms where the five-figure monthly fee buys the tooling and the deflection depends on what you build with it. Kore.ai is sales-led, so the $25,000 is our modeling anchor. Cognigy bills voice, chat, and LLM workloads separately on top of the base. Yellow.ai is resolution-priced in practice, which makes its $12,000 flat an anchor too, not an invoice.

Aisera and Moveworks sit on the other side of the line: turnkey, vendor-managed deflection agents. You sign, you integrate, and the vendor's product does the deflecting and the maintaining. Now Assist is the native add-on, and Agentforce IT Service is a migration. That is why the four rows carrying the identical modeled 30% cost anywhere from $17,400 to $2,250,000 in year one, a spread of roughly 130 times. The spread is not one vendor being 130 times greedier. It is four different product shapes wearing the same percentage: a credit pack you build on, an enterprise platform, a turnkey agent, and a per-user native add-on.

The honest read for a CFO: the platform rows price the tooling and leave the delivery risk with you. The turnkey rows price something closer to the outcome, and the higher fee is the vendor doing the work the platform would hand to your team. Which one is actually cheaper depends on whether you already have the platform engineers, not on the sheet alone.

Agentforce is a different decision

Agentforce IT Service posts the highest compression (50%) and the highest theoretical net (positive $1,210,000). But its 50% assumes you move the ITSM workflow off ServiceNow onto Salesforce. The $50,000 setup in the model is optimistic for a platform migration that touches every fulfiller, every integration, and every runbook. If you are already consolidating on Salesforce, that number is a real option worth pricing seriously. If you are not, treat the migration cost as far higher than the sticker and the 50% as the payoff for a project, not a plug-in.

The Monday morning move

If you are keeping ServiceNow, the sheet now forks on one question: do you have a platform team with capacity to own a bot? If you do, and your stack is already M365-heavy, Copilot Studio posts the best year-one number in the field (positive $342,600 realistic) at a cost so low the experiment is nearly free even if it underdelivers; just budget the internal build and maintenance labor honestly, because the sheet does not price it. If you want vendor-managed deflection with no build program, Aisera is the turnkey pick: positive $177,000 realistic, the lowest turnkey cost, and no platform migration. Moveworks buys the same family of deflection at more than three times Aisera's flat fee and runs negative in year one, though it clears in year two. If you are already moving to Salesforce, price Agentforce IT Service as a migration with a real number, not the $50,000 placeholder. Skip the per-user native add-ons on an estate this size unless the base seat is already paid for.

Whatever you choose, choose one. The MAX rule means the second agent adds no compression, and the fully-loaded math means it adds its entire bill. We walk the same ServiceNow conversation from the renewal side in the ServiceNow renewal playbook and the Moveworks unlock math. Run your own seat count and rates through the calculator before you sign anything.