I'm a big believer is not pre-scripting any story line. I figure out a) Who is involved in a situation, b) What their goals are, c) How they operate ( their preferred, emergency, and avoided tactics ), d) What they know/believe, and e) What resources/abilities they have. 

This makes it simple figuring out what the central conflict is ( where the goals collide ), and it's pretty easy to tell when the conflict is over ( one of the groups has accomplished their goals, or everyone has withdrawn from the conflict ). And it means that I merely have to keep asking myself "given that they want, and know, what will they do next to try and move themselves closer to their goal"? This means that I never have Player derail what I expected, as I don't expect anything, I react to everything - but I make sure that all the factions and NPCs constantly attempt to move toward their goal, so that the narrative doesn't wander around randomly.

I also try to keep story pacing decently managed ( although I have a lot of work to bring my practice and my intent up to match one another ).

• I want fast paced scenes to have terse descriptions, lots of action words, lots of "calls to action", and to not give the Players much time to react: The prisoner flips over the grocers stand, scattering produce in your path, as his accomplice leaps off the wagon at you, drawing a pair of daggers - what do you do?!

• I want slow scenes to be more involved and information dense. That's not having long-winded multi-sense descriptions - still use language economically - but have much more to describe, such a complex setting, with many possible choices for interaction or branching exploration. Likewise, NPC conversational scenes are slow, but contain a lot of interactions, and usually a lot of potential information. Or - I can have a slow scene with a single piece of information that I really want to drive home - taking several detailed sentences to keep pounding that one fact into the Players: By evening, the cold biting wind out of the north is like an icy knife. The horses start first shivering, then stumbling and faltering. Your fingers and toes start that sharp prickling that tells you that frostbite is not far from setting in. As you top the next rise, looking further up the winding mountain pass, bracing against the knife-like blast of wind, your guide Felgar points up the slope of the mountain, "The horses are starting to founder, they won't last much longer, and neither will we. We could shelter in the caves along the pass ... I've heard that the Goblins lair there, I don't know if we should risk it, but we can't stand much more of this cold". Note: this still involves a conflict and a choice - not the same as a detailed transition which sets up a new scene.

• I want fast and slow scenes inter-cut with each other. I might just go back and forth fast/slow/fast/etc. - or I might try and mimic the pacing of known genres ( e.g. Horror movies might start slow for setup, have one fast paced "big event", then back to slow, with slowly rising pace over many slow scenes building tension and suspense, then one big fast-paced-burst-of-non-stop-oh-god-we're-all-going-to-die action scene for climax, and a slow scene to relax and wind up threads ).

And I also want to manage table flow, reasonably well ( again - my ideal and practice gap needs a lot of work ). Time blocking out how long a scene should take in real world table time, and trying to limit the time spent of that scene to that limit, works decently well.

Now - if I had a pre-defined, branching story-structure flowchart, pacing & flow would be pretty easy to plan along these guidelines. I could make sure that I'm shuffling in fast and slow scenes, and I could pre-estimate the time each scene might take. I could examine the possible story flows, figure out where I might need to toss in an extra scene between these two scenes, if the Players go down that path, etc.

But how to approach, since I'm not pre-planning?

I'm thinking along these lines:

1. Time block on the fly. This just takes a bit of practice: "They Players are trying to infiltrate the ship, and sneak down the hold, without the crew realizing ... let's try and hold the entire scene to 10-minutes or less".
2. For each iteration of the OK, what will everyone try and do next cycle come up with a Fast Scene Response and a Slow Scene Response. Actually, come up with more than one of each, would be ideal. These all must make narrative sense. Will the local Thieves guild try and assassinate one of the Players as an act of revenge ( combat, fast ), or kidnap a member of their family and send a threatening letter to cause the Party to back off ( social/emotional, slow )? Either would make narrative sense, and be true to the Guild's character; what are the needs of the pacing, right now?
3. Have a collection of semi-random events of each pacing type, to throw in as needed. These could be secondary effects of the Players' ( or NPCs' ) actions coming home to roost ( that grocer comes charging up with the local Guard to accuse the Player of destroying his place of business ), or actually random encounters in the wilderness or dungeon. These could even be  recycled unused actions from point #2 above ( so long as they still make narrative sense ).
4. Related to this, develop tactics to turn a slow-paced scene that the Players' have insistent on going into, into a fast scene: "Suddenly, as you're talking to the Duke, a group of black clad figures swing through the stainglass windows, dropping into half-crouches with blades drawn - roll initiative!", or a fast scene into a slow one "You draw weapons, kick in the door, and charge in! There stand Advesario, standing there, unruffled, smirking slightly as he holds the struggling Damselo in front of his, knife to her throat ... 'That's quite far enough, he drawls ...".

 So - any thoughts, or suggestions? 

What tactics or techniques would you use for juggling Pace and Flow with an improvisational style?

A couple of ideas.... An hourglass behind the screen. Sand falling is a visual queue to speed up. You may not even realise it, but watching that sand will increase your own tension and you will (almost certainly) speed up. Even if the players can't see it.

Second, for the slower ones. Keep a stock of pictures of locations. A nice picture of a musty library. A dingy alley. A tavern common room. Have a bunch of them on an ipad so you can quickly refer to them. Then if a player enters one of those areas, you can describe it in as much detail as you want, looking at the picture. This will only get you so far, obviously, but it's something I like to do, even for people. If I don't need an NPC to be very specific, I'll just get a picture from the net. I might google old female tavern keeper. Or Skinny male beggar. Or whatever. And then I have everything down to their hair colour, skin tone, what they're wearing, how grubby they look. It might not work for everyone, but having a stock of visual stimulus is invaluable to me.

I like those suggestions.

I hadn't thought of psychological tricks to keep me on task - but that's just as valid :)

I like the hourglass idea ( or digital timer ). My DM Screen is a laptop, so quickly Googling pictures for the scene - or random stock photos of people - is a definite idea.

However, I'm not convinced that long descriptions are necessarily a good thing. I'm thinking the formula of General Descriptor + 1-2 distinctive sensory features +  1-2 distinctive actions is enough: Tall thin balding man, with a large paunch and bone white skin, shuffling gloomily down the street in baggy ragged clothes.  A lanky humanoid wolf like creature drops out of the trees onto the forest floor in front of you. It's red eyes glint as it' bares its teeth at you in a snarl and leaps at you! That's enough to set a picture in the Players' heads - and if they want more detail, they'll ask. Often they won't, because they have a fully fleshed-out visual picture built out of their own imagination, already.

When I'm talking about slow scenes, I'm talking about using scenes - or interactions - which have a lot of detail, or complexity, not artificially driving up the descriptive complexity of an otherwise simple scene by inflating the description. You still describe the facets of the scene or situation efficiently - there's just a lot of details or options to describe.  Or I'm emphasizing a particular aspect by restating the same qualities in multiple ways ( see my cold example above ) - which, I guess, is kind of going into more descriptive detail, but I'm not describing more detailed facets, I'm just hammering repeatedly on the one in different ways ( it's really cold! ).

This means that if I want a slow scene, I have to add complexity: social interactions, or lots of options for exploration ( environmental interactions ), but without any obvious time pressure. 

If I want a fast scene, I need to pile on time pressure ( locked door, water rising, pick the lock! ), risk ( combat! ), or a high number of events piled on in a short amount of game ( training montage out of movies ).

Last category is transitions which is a fast scene with no interaction needed: so after 3 weeks of travelling across the desert, you finally arrive at the trading outpost of ....